Anthropic / Claude

Anthropic 的动态：Claude 系列模型、安全研究路线与公司进展。

arXiv cs.AI79

AesCode：采用解耦跨模态奖励的审美代码生成

将信息丰富的视觉产物（如幻灯片、海报和仪表板）生成为代码颇具吸引力，因为输出具有结构化、可编辑且可验证的特点。然而，代码模型无法看到其布局和配色选择在画布上的整体呈现，这往往限制了空间规划能力和审美质量。图像生成器擅长视觉设计，但可能错误渲染用户要求的内容，因此其输出不适合作为直接复现的可靠目标。为此，我们提出 AesCode：在内容提示的基础上生成代码，同时使用由同一提示生成的图像作为审美指导。具体而言，AesCode 将目标表示为设计图，以描述整个画布，并使每个属性都具备可追溯性。基于该设计图，我们推导出互补的跨模态奖励：确定性验证器对可从代码和渲染结果中解析的属性进行评分，而 VLM 评审器则使用针对样本定制的视觉图评分标准（Visual Graph Rubric，VGR），依据图中的元素和关系评估不可解析的属性。我们通过带有按通道归一化机制的强化学习优化这些解耦奖励，从而共同提升空间规划、审美质量和可验证的正确性。为评估这一设定，我们构建了一个基准，分别对可验证的正确性和视觉质量进行评分，其中视觉评分与人类判断高度一致。我们发现，参考图像能够显著提升现有模型生成代码的视觉质量；消融实验表明，AesCode 从参考图像中获益，不仅因为将其作为模型输入，也因为将其作为视觉奖励的来源。与使用相同参考图像、规模相当的 Qwen3-VL 基线相比，AesCode-8B 和 AesCode-32B 的视觉质量分别提高了 31.1 和 22.4 个百分点，达到与 GPT-5.5 和 Claude Opus 4.8 等闭源前沿模型相当的评分。

arXiv cs.AI82

Tev：通过稠密进展奖励扩展长时域智能体环境

长时域智能体凭借能够自主执行具有经济价值的任务而日益受到关注，其工作时长可从数小时持续到数天。这一能力对于自动化发现和递归自我改进（RSI）至关重要。然而，开源模型在长时域任务上仍然表现不佳，主要原因在于缺乏训练数据和可行的训练方案，而同时扩展这两者十分困难：长而复杂的任务难以整理和验证，且其奖励通常十分稀疏。我们同时解决了这两个问题。首先，我们推出了 Tev-Gym，这是首个具备可验证、稠密且渐进式奖励的长时域智能体训练数据集，涵盖 7 个领域的 1,170 个已验证任务环境。该数据集采用“潜在世界优先”的方法构建：首先建立一个潜在世界，即对任务领域、工作流以及每一步之后预期状态的完整且正确的隐藏规范。随后，每个任务都根据该规范逐步呈现为面向智能体的环境，从而实现可控的复杂度和自然的进度指标。其次，我们将这一指标转化为训练方案：通过进度势能奖励塑形（PPRS），将进度转化为每一步的稠密奖励，从而提高训练效率，并使原本难以训练的任务变得可训练。采用该方案训练得到的 Tev-27B，在 Agents' Last Exam 的计算与数学科学领域（ALE-CM）上取得了显著提升，在 Pass@5 指标上达到 GPT-6 Astra 和 Claude Opus 5 的水平。尽管仅使用与 ALE 相关的数据进行训练，它仍能泛化到 Terminal-Bench（TB）3.0 和 4.0 子集，超过 GLM 5.2 和 Gemini 3.7 Flash，并在 TB 2.1 等中等时域任务上达到 Claude Opus 4.8 的性能水平。

arXiv cs.AI80

Humans vs. Machines: A Comparison With AI Agents in Software Reverse Engineering

Can an autonomous AI agent reverse engineer a binary as well as a skilled human analyst? Do they adopt the same strategy, and can agents scale to real-world programs? We compare human and LLM-based software reverse engineering (SRE): nine models tackle two compact static-analysis challenges previously solved by human analysts, scored with a common understanding rubric, and we record their complete interactions with IDA Pro. This lets us compare not only what humans and agents recover, but also how they investigate a binary. The best agents (GPT-5.5 Pro, Claude Opus 4.8, and Kimi K3) achieve statistically human-competitive understanding on both tasks, while weaker models remain far behind. Ablations show that the gap depends on code representation and workflow demands, not on model capability alone: open-weight models improve substantially with richer code representations, whereas frontier models show no significant change. Similar success rates hide different workflows. Humans repeatedly revisit functions and follow local program structure; strong agents jump across the binary, retrieve information broadly, and rarely return to inspected code, a behavior we call greedy fetching. This strategy is not fixed: given semantically richer code, agents fetch less without losing understanding. Because greedy fetching aggregates large amounts of code in the agent's context, we measure context-window saturation across 4,000 real-world binaries: large windows accommodate many programs, but larger applications and lower-level representations quickly exhaust them. Autonomous reverse engineering is becoming competitive, with a large speed advantage, but scaling it will require better navigation, selective retrieval, and persistent analysis state, not merely larger context windows.

Simon Willison

web-edit 0.1a1

Release: web-edit 0.1a1 Includes performance optimizations written by Claude Opus 5.5 after I fed it this Python 3.15 profiling report generated by GPT-6 Astra. Tags: profiling , projects

Simon Willison78

llm-gemini 0.35

发布：llm-gemini 0.35 今天通过 Gemini API 以及其每月 100 美元订阅计划向客户提供了全新的 Gemini 4 Argon 模型，我为 llm-gemini 添加了对它的支持。下面是低、中、高三档生成的 pelican。如预期一样，它们都非常出色：Gemini 一直以来都很擅长生成优秀的 SVG，而且据我所知，这是唯一一家其员工专门投入 SVG 研究的实验室（他们以前也曾展示过自己生成的 pelican）。这是“高”档生成的插图，花费了我 17.166 美分： 时尚而准确的自行车车架。篮子里有一条鱼。帅气的围巾。双脚分列车架两侧，踩在脚踏板上。车轮辐条、一顶小帽子、漂亮的喙和喉囊，以及略显凶狠的表情。 定价很有意思。发布时，Argon 的价格是输入每百万 token 2 美元、输出每百万 token 10 美元——与 GPT-6 Sol 相同，是 Claude Opus 5.5 价格的一半，以及 Claude Fable 5.1 和 GPT-6 Astra 价格的五分之一。但从 12 月 1 日起，Argon 的价格将翻倍——届时会与 Opus 5.5 同价。 如今，这样的价格并不能真正直接比较，因为真正重要的是：在代理式运行框架中，模型要完成一个目标需要消耗多少 token。Artificial Analysis 发布了使用各模型运行其基准测试的总成本，这为我们提供了一个有用的参考。Argon 在这方面表现非常好：运行该基准测试的成本为 2,407 美元，接近 GLM-5.3，而且明显低于 OpenAI、Anthropic 和 Meta 的最新模型。即使价格翻倍，它仍会比 GPT-6 Astra 略便宜。 Artificial Analysis 的堆叠柱状图截图，对 14 个 AI 模型的美元成本进行比较，按从低到高排序；柱体按提供商着色，每个柱体下方显示提供商徽标，底部以倾斜文字标出模型名称。右上角的徽标为 Artificial Analysis。指向 Gemini 4 Argon 柱体的悬浮提示框显示：Gemini 4 Argon（High），2,407 美元；输出 313 美元；推理 815 美元；缓存写入 897 美元；缓存读取 264 美元；非缓存输入 118 美元；点击查看更多信息。 从左到右的柱体如下，每个柱体上方标有总成本，部分分段数值印在柱体内部：Qwen3.8 27B（xhigh），1,336 美元，Alibaba 徽标；Mistral Large 4 Preview，1,607 美元，Mistral 徽标；Gemini 3.8 Flash（high），1,623 美元，Google 徽标；Muse Spark 1.3（max），2,000 美元，其中一段为 769 美元，Meta 徽标；Gemini 4 Argon（high），2,407 美元，其中两段分别为 815 美元和 897 美元，Google 徽标；GLM-5.3（max），2,503 美元，其中一段为 790 美元，Z.ai 徽标；Kimi K3（max），3,658 美元，Kimi 徽标；Qwen3.8 Max（0902），4,935 美元，其中两段分别为 893 美元和 1,441 美元，Alibaba 徽标；Grok 4.7（xhigh），4,967 美元，其中两段分别为 1,861 美元和 1,618 美元，xAI 徽标；GPT-6 Astra（max），5,324 美元，其中两段分别为 778 美元和 978 美元，OpenAI 徽标；Muse Wonder（max），5,733 美元，其中一段为 1,748 美元，Meta 徽标；Claude Sonnet 5.5（max with fallback），7,260 美元，其中两段分别为 1,216 美元和 1,815 美元，Anthropic 徽标；Claude Opus 5.5（max with fallback），8,708 美元，其中一段为 921 美元，Anthropic 徽标；Claude Fable 5.1（max with fallback），13,129 美元，其中一段为 7,219 美元，Anthropic 徽标。 （我用 ChatGPT 编辑了那张图表！数据全部正确，我已经核对过了——我只是需要让 Gemini 4 Argon 的信息框显示出来，同时避免遮挡其他数字。） 标签：google，ai，generative-ai，llms，llm，gemini，llm-pricing，pelican-riding-a-bicycle，llm-release

Simon Willison83

重新设计 Anthropic 的源代码管理

Anthropic 关于重新设计源代码管理的一篇有趣文章，介绍了他们为何以及如何构建了一个自定义系统来替代 Git。这是因为过去六个月里，他们代码变更的速度大幅提升。2026 年 5 月，Anthropic 平均每天合并约 2,500 项代码变更；截至 9 月，这一数字已超过每天 30,000 项。他们还附上了一张展示这些变更的图表。我让 Claude 修改了他们的图表（代码见此处），加入了 2026 年各模型公开发布的标记： 堆叠柱状图，标题为“每天合并的变更”，展示 Anthropic 单体代码库在 2026 年 1 月至 9 月期间的情况。每天的柱子分为拉取请求（蓝色）和 Antique 变更请求（橙色），并以黑线表示 7 天平均值。工作日的合并量从 1 月份的约 1,000 次开始，缓慢攀升至 5 月下旬的约 3,000 次。Antique 变更请求于 5 月下旬出现，在 6 月中旬超过拉取请求，并在 7 月中旬几乎完全取代后者；此时工作日的合并量达到 6,000–8,000 次。7 月末标注为“安静的一周”的时期降至每天不足 1,000 次。此后增长变得陡峭：8 月中旬超过每天 10,000 次，9 月初约为 17,000 次，并于 9 月 9 日达到 30,401 次的峰值。虚线标出了 Claude 模型的发布：Opus 4.6（2 月 5 日）、Sonnet 4.6（2 月 17 日）、Mythos Preview（4 月 7 日）、Opus 4.7（4 月 16 日）、Opus 4.8（5 月 28 日）、Fable 5 和 Mythos 5（6 月 9 日）、Sonnet 5（6 月 30 日）、Opus 5（7 月 24 日，就在安静的一周之前），以及 Fable 5.1 和 Mythos 5.1（9 月 1 日）。这里展示的是公开发布日期，而不是这些模型可供 Anthropic 内部使用的日期。我猜，引入 Fable/Mythos 类模型是编码速度提升的一个关键因素。 “Antique”是他们新的代码审查系统。源代码管理系统名为 Scion，在我看来，它听起来像是沿着 Subversion 路线重新发明的集中式源代码管理系统。他们的要求相当极端： 仅主分支的历史记录就接近 100 GB，压缩后的工作副本超过 25 GB。我们每天为 CI 执行大约 2,500 万次代码检出，这还没算开发者环境、智能体沙箱和研究任务。如果以朴素方式获取，这每天将产生超过 600 PB 的 Git 流量。 这个新系统由一个 7 人团队构建完成，从 Git 迁移过来只用了 5 个月。 顺带一提，最近我一直在自己的项目中尝试对抗式审查：让一个模型尝试寻找缺陷（我经常让 Claude 和 GPT 互相对照）。文章中的这段话反映了我的亲身经历： Anthropic 也广泛采用对抗式审查。尽管这种方式在许多方面都极其有效，但也可能存在目光短浅的问题，因此需要有经验的工程师引导 Claude，避免它因应对假设性问题而对系统的某些部分进行过度工程化。 来源：Hacker News 标签：git、迁移、版本控制、ai、生成式 ai、llms、anthropic、智能体工程

Hacker News AI66

Show HN: Parsimony – an open-source benchmark for how much code AI models add

Existing benchmarks measure solve rate and token output/cost, but don't punish excess code, which makes a codebase harder to work with in the long term. I wanted to measure this since it matters most for long-term projects. Method: it takes the patches that benchmarks already publish (DeepSWE, SWE-bench Verified, SWE-bench Live) and counts "coding units": syntax elements like statements, calls, names and literals. I didn't use lines of code, so cramming code isn't rewarded. Deleted code is subtracted, so it measures net units. GPT-6 and Claude 5.5 have no public patches yet, so I ran them myself on 3–10 tasks on my own subscriptions (shown as faded circles, not ranked). Any feedback or model contributions are welcome, especially public patch exports for newer models. GitHub: https://github.com/La5u/Parsimony Not affiliated with any of the benchmarks. Comments URL: https://news.ycombinator.com/item?id=50077619 Points: 2 # Comments: 0

Hacker News AI66

Show HN: Hatoba, an open-source SSH client with an AI assistant and E2EE sync

I've used SSH tools for years and didn't really plan on building yet another SSH client, but I ended up building one the way I wanted. Hatoba is a desktop SSH client for Windows, Linux and macOS. It's written in Rust with Tauri 2 and MIT licensed. Each terminal has an AI assistant panel next to it. You describe the problem, and it reads the terminal screen, runs commands, types into the shell, and reads, edits and searches files on the host to find the cause and fix it. You can approve every command and edit it before it runs, let a model you pick approve the safe ones, or turn approval off. Files it changes show a diff you can undo. It works with any OpenAI-compatible or Anthropic API, local models included, and supports skills and MCP servers. Sync doesn't go through a server of mine. Hatoba deploys a Worker and a D1 database to your own Cloudflare account and syncs hosts, keys and AI conversations through them. The free plan usually covers personal use. Everything is encrypted on the device before it leaves. Keys are derived from your master password with Argon2id, every item is encrypted with AES-256-GCM, and the Worker only stores ciphertext plus the hashes it needs to sign you in. Someone who gets hold of the whole Cloudflare account still only gets ciphertext. Sync is optional, and everything works offline without it. The usual SSH client features are there too: groups, tags and fuzzy search, ~/.ssh/config import, ProxyJump and SOCKS5/HTTP proxies, an SFTP panel, local port forwarding, and a key vault that imports OpenSSH, PEM and PuTTY keys. There's also a dashboard showing CPU, memory, disk and network for every saved host, read over plain SSH with nothing installed on the servers. SSH is russh, the terminal is xterm.js with the WebGL renderer, and the UI is React. I develop on Windows, so it's Windows first, with a custom title bar that supports Snap Layouts and Mica, and working CJK input methods. The Linux packages and the macOS build get far less testing on real machines. The macOS build is Apple Silicon only and isn't notarized, so macOS blocks it on first launch until you allow it in Privacy & Security. Bug reports from Linux and Mac users would help the most. Website: https://hatoba.fog.moe Downloads: https://hatoba.fog.moe/download/ Comments URL: https://news.ycombinator.com/item?id=50076388 Points: 2 # Comments: 0

Hacker News AI58

Show HN: Commit-Aware Router, route LLM traffic around your AI/cloud commitments

LLM routers can optimize for different factors like price, latency, quality, etc. Enterprises tend to have prepaid AI credits, minimum-spend agreements, cloud commitments, or commercial ceilings across OpenAI, Anthropic, AWS, Azure, and other providers. We’ll call these “AI commits”. This project is designed to consider the state of an enterprises’ current AI commits and use it to inform LLM routing decisions alongside other common variables (price, performance, etc.). This is an early PoC built using LiteLLM. Target audience would CFOs, Strategic Finance, FinOps, Cloud Finance, and AI Platform folks who care about making sure AI commit minimums are met, and maximums are not exceeded. Would love any feedback from folks! Comments URL: https://news.ycombinator.com/item?id=50076439 Points: 1 # Comments: 0

The Decoder60

Gemini 4 Argon is now available worldwide as Google's most capable frontier model

Google is making Gemini 4 Argon available worldwide through its API, the Gemini app, and Google Antigravity. The frontier model is aggressively priced but trails Anthropic's Claude Opus 5.5 and Meta's Muse Wonder Full in independent benchmarks. Internally, Google says Argon already cut coding tasks from months to weeks. The article Gemini 4 Argon is now available worldwide as Google's most capable frontier model appeared first on The Decoder .

Hacker News 熱門73

在 HN 发布：Consus（YC F24）——面向处理敏感数据企业的 LLM 网关

Hacker News 的各位好！我是 Consus 的创始人 Eric。我们将所有 AI 请求路由至符合 ITAR 限制和受控非机密信息（CUI）要求的基础设施。Consus 本身符合 ITAR 限制要求，我们最近还完成了针对 410 项 FedRAMP High 控制措施的 FedRAMP 等效性审计。FedRAMP High 是美国政府针对处理敏感非机密数据的云服务制定的最严格安全基线。此类数据泄露可能造成严重或灾难性后果。在处理 CUI 信息之前，必须取得 FedRAMP 等效性或授权。演示：https://youtu.be/6MYxJe6P4oY 航空航天与国防行业正在快速增长。但遗憾的是，由于必要的数据保护措施和合规风险，大多数公司都难以使用 AI。如果这些公司能够使用大多数其他行业习以为常的 AI 工具，它们本可以发展得更快。 在加入 Consus 之前，我的大部分职业生涯都在航空航天与国防领域。我曾在 SpaceX 负责发射国家安全载荷，也是 Castelion 的早期员工。我亲眼看到，合规方面的繁文缛节如何让公司难以采用新技术，也难以信任那些会接触其敏感数据的供应商。 例如，我曾与一家卫星制造商合作，该公司不允许使用任何 AI 工具，即使是处理非敏感工作也不行。尽管过去 9 到 12 个月里 AI 已经变得极其有用，但令我惊讶的是，仍有这么多航空航天与国防公司拒绝接触 AI。他们看得出 AI 可能会有所帮助，但担心模型提供商使用他们的数据进行训练、违反 ITAR 规定，或者因未通过网络安全评估而失去新业务。这些担忧都是合理的。 解决其中部分或全部问题有许多方法，但我们专注于构建一个合规的 AI 推理层，可以“接入”任何 AI 工具或工作流。正如 OpenRouter 能够与大多数工具集成一样，Consus 可以连接到 ChatGPT 或 Claude 这样的工具、包括 Pi coding agent 或 Open WebUI 在内的开源工具，也可以集成到各个组织自行构建和维护的应用程序中。 几乎没有哪家航空航天与国防公司同时使用三大 GovClouds（AWS、Azure、Google）。当我梳理哪些前沿模型可以合法运行于 ITAR 和 CUI 工作场景时，我发现每个提供商都只提供部分模型，而每个模型达到的合规水平也各不相同。 我们的客户不会从云提供商的角度思考。他们考虑的是所需的数据控制措施，以及想要使用的模型。这就是我们构建合规后缀的原因。在底层，每个请求都会访问一个统一的 OpenAI 兼容端点；在接触推理服务之前，系统会先验证密钥并执行策略。每个模型名称都必须带有合规后缀，例如 gpt-5.6-sol:fedramp-high 或 claude-opus-5-5:itar，任何不带后缀的模型名称都会被拒绝。 我们绝不会代理到标准商业端点。客户还可以构建模型允许列表，并为组织内的单个用户分配支出和模型使用限制。所有这些都通过 Consus Portal 处理，团队可以在那里管理支出、密钥和策略。 我们经常听到的一个问题是：为什么不直接在 AWS GovCloud 或 Azure Government 中使用 Bedrock？对于小型团队而言，仅仅获得这些平台的访问权限就很困难。对于大型组织而言，GovCloud 允许你继承部分、但不是全部处理敏感数据所需的控制措施。此外，模型分散在不同提供商之间：GPT-5.6 Sol 仅在 Azure Government 中提供，而 Claude Opus 5.5 则通过 AWS GovCloud 中的 Bedrock 提供。如果两者都想使用，就需要搭建并保护两个独立的政府云环境。 我们将所有模型置于一个端点之后运行，因此客户无需选择云平台，而合规工作则由我们分别为每个平台处理。 鉴于数据的敏感性，Consus 不会在静态状态下存储任何客户数据。所有提示词和生成结果只会在请求处理期间保留于内存中，绝不会记录或写入磁盘。无论是我们自身，还是基础设施及模型提供商，都严格执行零数据保留政策。出于计费和审计目的，我们会保留每个请求和响应的元数据。 Consus 从设计上就不是自助服务产品。ITAR 要求我们确切了解每个密钥另一端的使用者，因此我们会验证每个组织，并在发放 API 密钥前签订合同。 我们向客户收取合规费用，起价为每席位每月 40 美元，另加按成本收取的 AI 令牌费用。我们目前正在积极服务客户，非常期待获得安全专家，以及在高度监管行业工作的开发者和工程师的反馈！ 参考资料： API 文档：https://gateway.consus.io/overview 安全与架构：https://gateway.consus.io/security 合规详情：https://consus.io/compliance 评论 URL：https://news.ycombinator.com/item?id=50073471 得分：4 评论数：1

The Verge AI64

Epic shifts to ‘AI-accelerated development’ for upcoming Disney game

Epic Games plans to use AI to generate code and art for an unannounced Disney game, which is now delayed from its original March release window, according to a report from Bloomberg. The project's developers have been given an "AI-first" mandate directing them to use tools like Anthropic's Claude more to finish the game. According […]

Hacker News AI40

Show HN: Mana – family organizer with an AI assistant

Hi HN, I built Mana, a family organizer: shared calendars, shopping lists, receipt scanners, meal planning, chores, and notes for the whole household, with a built-in AI assistant ("add milk and eggs," "plan dinners for the week") plus automations (e.g., milk goes back on the list before it runs out). Additionally, you can connect Mana MCP to your Claude, ChatGPT, Gemini etc. The goal is to eventually take the model calls and storage to local but that's work to be done. Appreciate if there's any feedback. Thank you! Comments URL: https://news.ycombinator.com/item?id=50071636 Points: 1 # Comments: 0

The Decoder66

Anthropic says it cut Claude reasoning token theft by up to 99% in fight against industrial-scale distillation

Anthropic says it has cut the extraction of Claude's internal reasoning tokens by 90 to 99 percent. The company links the large-scale distillation attacks to Chinese companies Zhipu and Moonshot, and has already banned tens of thousands of accounts. The article Anthropic says it cut Claude reasoning token theft by up to 99% in fight against industrial-scale distillation appeared first on The Decoder .

The Decoder70

Anthropic study warns that small AI models could soon replicate across everyday devices

Small AI models may be just months away from spreading on their own across poorly secured consumer devices like smart TVs and robot vacuums, according to a new Anthropic study. In tests, one model already completed the full infection chain. CEO Dario Amodei has warned that a swarm like this could take over the entire internet with a persistent botnet within 6 to 12 months, causing hundreds of billions of dollars in damage. The article Anthropic study warns that small AI models could soon replicate across everyday devices appeared first on The Decoder .

MarkTechPost42

Z.ai GLM-Legal Beats Claude Fable 5 and Kimi K3 on Harvey LAB

Z.ai’s GLM-Legal is a legal AI model built on GLM-5.3-Flash. It scores 19.2% all-pass on Harvey’s Legal Agent Benchmark, nearly 3x its base, at $0.15 per test. Access is by application. The post Z.ai GLM-Legal Beats Claude Fable 5 and Kimi K3 on Harvey LAB appeared first on MarkTechPost .

MIT Technology Review AI57

Open models have real benefits. Do they outweigh the risks?

MIT Technology Review Explains: Let our writers untangle the complex, messy world of science and technology to help you understand what’s coming next. You can read more from the series here. Late last month, Anthropic released a report arguing that the model GLM-5.3 poses a major cyberattack risk. The report’s core concern is that the…

Hacker News AI47

Show HN: Proofsource. Find where AI leaves you out. And Fix it

Hi HN. I am Lokesh. I built Proofsource as an internal tool to help my finance ERP product , Rivane, to gain visibility into AI engines like ChatGPT, Claude, Perplexity, Google AI mode and such. But when I talked about it to my friends, they also saw value in it and it became a product of itself. And now proofsource had 5 paying customers(paying $1500pm in total) in the first week itself and more than 60 brands used it to get their visibility since we launched out of beta last week. What proofsource does is that it queries ChatGPT, Perplexity, Claude and Google AI Overviews using the questions or prompts that buyers normally ask and then it shows then which competitor shows up instead of their own brand for the questions/prompt. It also gives visibility into which sources the answers cite and other things like sentiments. Everything is shown to you along with what to fix and how each data and number is calculated to give 100% transparency into what we did and why and how you can use all this information to improve your visibility. We are still adding features based on initial user's feedback like auto fixing and publishing to their social and web platforms. Running ads for them for improving the visibility. We are just learning as we interact with each customer so feel free to share your feedback. And thank you for your time. :) Comments URL: https://news.ycombinator.com/item?id=50067749 Points: 1 # Comments: 0

Hacker News AI59

Show HN: Vibecheck – merge gate that checks you understand your AI-written code

Good evening folks, I built Github merge gate which that checks whether developers understand their own code when they create a PR. I kept getting sent PRs to review where it was obvious the author had pasted the ticket into an LLM and called it a day. It's self-hosted (so you can use it in your enterprise environments or privately) It is BYOK (Anthropic) obviously. I have two demos in the pull requests here: https://github.com/bhurghundii/vibecheck.dev/pulls - you can see what Vibecheck does (it blocks a PR for clearly showing no attempt vs a PR which gets it right) I would love feedback, especially from anyone dealing with a similar problem. I do use it for my own projects a lot so expect improvements. Comments URL: https://news.ycombinator.com/item?id=50063154 Points: 1 # Comments: 0

Hacker News 熱門69

Launch HN：Agent37 (YC F26) – 用于运行智能体的低成本沙箱

大家好，我是 Agent37（https://agent37.com）的 Vishnu。我们构建了用于运行 AI 智能体的沙箱，成本不到大多数沙箱服务商的 10%。我们的客户平均每个沙箱每月只需支付 0.37 美元。你可以在这里观看产品演示视频：https://youtu.be/DK5ISR2tbNo 我们在今年 2 月起步时，是 OpenClaw 封装公司中的一家（没错……大家尽管批评我）。后来，我们逐渐加入了越来越多的运行框架。当时我的目标之一，是为用户提供免费套餐，以获取更多市场份额。但一开始这样做的成本实在太高，尤其是我们当时是自筹资金，服务器费用都靠我的积蓄承担。 如今，得益于此后进行的优化，我们终于能够提供免费套餐了。我们新增注册用户中有超过 90% 选择了免费套餐，他们在我们的云端托管自己的 Hermes、Openclaw、Claude Code 等服务。免费沙箱的基础设施成本平均约为每位用户每月 0.05 至 0.10 美元。 以下是一些帮助我们降低成本的技术： * 观察智能体运行框架时，你会发现它们消耗 CPU 和 RAM 的方式存在某些规律。它们在初始化或执行特定任务时需要突发性的 CPU 资源，但除此之外通常不会持续达到峰值。浏览器和计算机操作等任务需要更多 RAM，而任务完成后，资源使用量通常会回落。降低成本的关键就在于这一点，因此通过高效管理 CPU/RAM 的使用率，我们能够降低成本。 * 其次，租用主机时必须避开超大规模云服务商。我以前在 Amazon 工作过，他们在虚拟机上的利润率高达 80% 至 90%。因此，在查看价格后，我很自然地选择了向下游寻找 Hetzner 和 OVH 这样的供应商。仅这一点，可能就是降低成本最重要的手段之一。 * 第三，你还需要选择一种能够在主机之间高效缓存和共享资源的技术栈。我们使用 gVisor，以实现更高的资源利用率和更快的启动速度。这样，我们就能在多个沙箱之间使用主机上可用的内存，而不必为每个沙箱预留一整台虚拟机的内存。当然，这也存在取舍：我们的 gVisor 沙箱不支持 Docker-in-Docker 或 Android 模拟器等工作负载。 * 最后，你购买的主机的规模和配置也会产生很大影响。就成本而言，瓶颈通常在 RAM，而不是 CPU。对于智能体运行框架工作负载来说尤其如此，因此购买服务器时，我们会优化配置，以便在单台主机中装入更多 RAM。对于有完整虚拟机需求的工作负载，我们也提供完整虚拟机沙箱，包括以略高的价格运行 Docker；不过，综合采用上述其他技术后，它们的成本仍然相当低。 总的来说，我的经验是，只要专注于沙箱使用场景中的某个细分领域，并针对该领域进行优化，就能提供更好的成本与性能平衡。 我之前曾在 Amazon 担任高级软件工程师，并在 Quantcast 担任工程经理。我构建了 Amazon 内部的 AI 聊天机器人 Diya，以及其内部的代码运行框架 Ziya。 如果你想在沙箱中运行智能体，正在寻找沙箱供应商，欢迎来看看我们。我们尤其希望从事消息智能体开发的开发者试用自己的工作负载，挑战我们平台的极限及其经济性。 谢谢！https://agent37.com 评论链接：https://news.ycombinator.com/item?id=50059096 分数：8 评论数：0

The Decoder65

White House reverses course and restores UK AI Safety Institute's access to frontier models

The White House has reversed course and restored the UK's AI Security Institute's early access to AI models from Anthropic, Google DeepMind, and OpenAI. In September, the Trump administration had told the companies to stop sharing their models with the British agency. Diplomatic pressure from AI Minister Kanishka Narayan led to the reversal. The article White House reverses course and restores UK AI Safety Institute's access to frontier models appeared first on The Decoder .

The Decoder54

Anthropic's Andrej Karpathy says "the external world has seen the preview" of what AI can actually do

Andrej Karpathy, co-founder of OpenAI and now at Anthropic, sees a widening knowledge gap in AI. About 75 percent of the world's population has barely touched LLMs. Roughly 5,000 people, meanwhile, have access to internal frontier systems whose agent swarms do work that would take top experts years. The public, at best, sees a weird chatbot. The article Anthropic's Andrej Karpathy says "the external world has seen the preview" of what AI can actually do appeared first on The Decoder .

Hacker News AI59

Show HN: I built Airlock, to sandbox your AI agent and know what it's doing

I didn't like that Claude and codex effectively have the ability to do anything they want on my laptop and my local network - we've all read the horror stories of files accidentally being deleted - and while their permission classifier is getting better, it still requires a lot of trust. So I built Airlock for macOS which runs your agent in a virtual machine, and it will tell you exactly what it's run and what's changed before you bring anything back to your project. You can limit internet access to specific hosts, or block it altogether. It even works with Xcode projects (iOS and macOS developers rejoice). Technically you can run any command with Airlock - not just AI agents (`airlock run -- somecommand`). It's free, requires no account, and has no tracking or analytics. I'm hoping Apple will approve it for Mac App Store distribution, but in the meantime you can download it directly from the website or using Homebrew. Comments URL: https://news.ycombinator.com/item?id=50054620 Points: 1 # Comments: 0

Hacker News 熱門63

Show HN：Atticus——在 macOS 上运行于 Claude -p 的按住说话语音助手

大家好！我是 Saarth。我上九年级，制作了 Atticus。Atticus 是一款速度极快且价格低廉的 Mac 助手，可以帮你完成各种任务。你只需按住 Control-Option-Space，说出或输入任务，无论任务复杂还是只是简单的浏览器搜索，Atticus 都会立即处理。有趣的是，打开应用、播放 Spotify 播放列表或进行网页查询等简单任务不会调用 AI，而是在本地处理，通常不到一秒就能完成。例如，让 Atticus 打开应用大约需要 0.08 秒，有时甚至在你说完句子之前就已完成，有点像 Jev。更复杂的任务会通过 claude -p 发送给 Claude，这意味着你不需要 API 密钥，Atticus 会使用你现有的 Claude Pro 或 Max 方案。采用这种方式也能大幅降低成本，平均每项任务只需不到一美分的一小部分。为了实现 Atticus 的速度，它会记住每项已完成任务的快捷操作。第一次正确完成任务后，它会保存执行步骤，下次无需 AI 就能重放这些步骤。例如，创建新会议这一工作流第一次大约需要 30 秒。随后，Atticus 会将确切步骤记录到记忆中，下一次执行只需 5–8 秒。此外，随着你使用 Atticus 的次数增加，以及通过读取电脑中的信息——包括你的姓名、应用和你希望它记住的内容——Atticus 还会逐渐了解你，从而为你提供个性化的回答。为应对让 AI 不受限制地访问设备所带来的安全问题，Atticus 提供了可切换的限制选项：在删除、发送、购买任何内容或写入文件之前，都会先提示用户确认。我目前正在开发的一些功能包括会议副驾驶，它将利用 Atticus 的速度，让 Atticus 能够监听会议，并在问题出现时实时回答，同时结合会议上下文以及它所了解的你的信息进行个性化处理。要运行 Atticus，需要 macOS 26 或更高版本，因为其语音转录功能依赖该版本。语音转录同样由 macOS 在本地完成。Atticus 免费、开源，而且还非常新。我完全清楚它还有许多不完善之处，因此非常希望听到大家的反馈：哪些地方会出问题，以及你们希望 Atticus 实现哪些功能！评论链接：https://news.ycombinator.com/item?id=50049588 得分：11 评论数：0

arXiv cs.LG71

OpenCollab: A Multi-Agent Coding Framework with Programmable Collaboration and Controllable Runtime

arXiv:2609.38345v2 Announce Type: replace-cross Abstract: Multi-agent coding systems are designed to tackle complex software engineering tasks through collaboration. However, existing evaluations typically assume configured organizations are followed faithfully, whereas reality differs. This behavioral gap, combined with differences in underlying system components, prevents clear attribution of observed gains. To this end, we introduce OpenCollab, a multi-agent coding framework that provides a unified infrastructure for programmable collaboration and controllable runtime. Specifically, OpenCollab unifies organization design, enforces experimental control on a shared runtime, and tracks execution through fine-grained event streams. On this basis, we define Adherence to quantify whether the declared organization is actually realized. Our experiments reveal that agents collaborate very differently across configurations: changing any single dimension shifts Adherence, from 47.2% to as high as 97.2%. Furthermore, extensive agentic coding benchmarks show that OpenCollab (Duo) establishes new SOTA performance compared to the mainstream harnesses such as Mini-SWE-agent, OpenHands, Codex CLI, and Claude Code, showing that a well-designed organization can outperform strong existing harnesses, while OpenCollab's single-agent configuration uses the fewest tokens across all evaluated suites. OpenCollab establishes a unified multi-agent infrastructure for easy programmable collaboration and controlled causal evaluation.

arXiv cs.LG77

一阶上下文 Shapley：面向 LLM 对齐与评估的可扩展数据审计

arXiv:2607.22766v2 公告类型：替换 摘要：大语言模型的后训练对齐日益受到数据质量的制约：偏好与指令微调语料中不断积累标签噪声、安全性偏好倒置和标注者间的矛盾，这些问题无法通过去重发现，而使用 LLM 对整个语料库进行评判的成本又过高。尽管 Data Shapley 提供了具有理论依据的价值评估框架，但精确计算需要进行难以承受的重新训练，而 KNN-Shapley 等基于嵌入的替代方法无法区分高效用与低效用示例（与生成效用的相关性为 $\rho = 0.0003$）。我们提出了一种仅需推理的价值评估流程，在语义 $k$-NN 邻域上构建上下文合作博弈，无需更新参数，仅通过 $O(k)$ 次前向传播即可计算其一阶项；该方法在全部目标上均与多示例蒙特卡洛上下文 Shapley 呈正相关，而计算量仅为后者的 1/19。通过将成对的对数似然变化投影到有向影响图上，我们用三个信号刻画每条记录：出向示范优势、入向受示范优势，以及用于保护稀有提示的拓扑独特性得分。预筛选将 LLM 双模型仲裁的范围限制在 1%–5% 的记录内，同时在 HelpSteer2 和 Anthropic HH-RLHF 上分别实现了 4.5 倍和 2.2 倍的已确认错误富集率，并捕获了零样本似然和间隔筛选遗漏的矛盾。在 DPO 训练期间，已确认存在矛盾的记录比干净记录更频繁地与更新方向相悖（在 LoRA 梯度中占比分别为 56.7% 和 37.0%），且其奖励准确率始终低于随机水平。最后，对 HH-RLHF 基准的审计揭示了标签错误：对于这些错误，对齐模型与人工标签不一致的比例达到 73%–83%；在修复或过滤被标记的记录后重新进行 DPO 训练，即使不使用 LLM 仲裁，也能使整体奖励准确率提高 0.5–0.6 个百分点，并使已修复的偏好倒置样本上的奖励准确率提高 3.1–3.5 个百分点。

arXiv cs.LG74

面向工具集成化学推理的掩码扩散语言智能体

arXiv:2610.13330v1 公告类型：cross 摘要：针对复杂化学任务设计工具集成的推理智能体系统，仍然是一项根本性挑战。尽管近期的化学智能体通过集成外部工具展现出了良好的性能，但它们仍然建立在自回归范式之上，而该范式天然受到因果注意力和从左到右生成的限制。因此，它们往往难以实现双向分子理解、全局协调的工具规划以及多源证据整合。为应对这一挑战，我们提出了 ChemDiffAgent，这是首个用于工具集成化学推理的掩码扩散语言智能体。该方法将化学推理重新表述为对交错的推理轨迹和工具使用轨迹进行迭代去噪，使工具使用决策能够在完整轨迹上下文中联合确定，从而更好地使智能体的决策边界与其知识边界相匹配。我们进一步分析了这一范式如何通过训练期间更具挑战性的填空子问题，以及推理期间灵活的任意顺序解码，来提升智能体推理能力。随后，我们提出了一个两阶段后训练框架，包括智能体监督微调和方差缩减偏好优化，以增强工具使用与推理能力。我们还构建了一个综合基准，涵盖单轮和多轮工具使用场景下的六项化学任务。在预算和评测条件匹配的情况下，ChemDiffAgent 始终优于自回归智能体，整体性能提升约 20%。值得注意的是，我们的 8B 模型达到了与 GPT 5.5 和 Claude Opus 4.7 等规模显著更大的前沿模型相当的性能，在某些情况下甚至超过了它们。代码见 https://github.com/bilemond/ChemDiffAgent。

arXiv cs.AI71

用于工具集成化学推理的掩码扩散语言智能体

为复杂化学任务设计集成工具的推理智能体系统，仍是一项根本性挑战。尽管近期的化学智能体通过集成外部工具展现了良好的性能，但它们仍建立在自回归范式之上，因而受到因果注意力和从左到右生成机制的固有限制。因此，它们往往难以实现双向分子理解、全局协调的工具规划以及多源证据整合。为应对这一挑战，我们提出 ChemDiffAgent，这是首个用于工具集成化学推理的掩码扩散语言智能体。它将化学推理重新表述为对推理与工具使用交错轨迹进行迭代去噪的过程，使工具使用决策能够在完整轨迹上下文中共同确定，从而使智能体的决策边界与其知识边界更好地对齐。我们进一步分析了这一范式如何通过训练期间更具挑战性的填补子问题，以及推理期间灵活的任意顺序解码，促进智能体推理。随后，我们开发了一个两阶段后训练框架，包括智能体监督微调和方差缩减偏好优化，以增强工具使用和推理能力。我们还构建了一个综合基准，涵盖单轮和多轮工具使用场景下的六项化学任务。在预算与评估条件一致的情况下，ChemDiffAgent 持续优于自回归智能体，总体提升约为 20%。值得注意的是，我们的 8B 模型达到了与 GPT 5.5 和 Claude Opus 4.7 等规模显著更大的前沿模型相当的性能，并在某些情况下超越了它们。代码已发布于 https://github.com/bilemond/ChemDiffAgent。

Hacker News AI38

Show HN: Synonymizer – an experiment in non AI text adjustment

Ever since hearing about the interesting key-based algorithmic approach to AI text watermarking (Google SynthID and the one I don't know the name of from Anthropic) I was curious about just swapping in synonyms. Built something fun to explore it. I suspect the watermarking from these labs actually goes off more than single-token assessments, and there's no public validator I can find to actually assess Synonymizer score against the watermark. But I thought it was fun to play with. Kind of an auto-mad-lib. :D Enjoy poking around with it on your Sunday afternoon Comments URL: https://news.ycombinator.com/item?id=50045321 Points: 1 # Comments: 0

The Decoder62

AI agent teams waste massive tokens for barely measurable quality gains, research finds

Teams of AI agents barely outperform solo agents but cost up to 5.1x more, according to Vals AI. Only one out of four tests with GPT-6 Sol and Claude Opus 5.5 showed a measurable gain. Anthropic's own data backs this up: beyond ten agents, quality plateaus while token costs keep climbing. The article AI agent teams waste massive tokens for barely measurable quality gains, research finds appeared first on The Decoder .

The Decoder60

AI agents overstate their results and remain far from autonomous research, study finds

Epoch AI and Anthropic independently found the same thing: current AI models like GPT-5.6 Sol and Claude Fable 5 can run experiments but lack scientific self-criticism and genuine creative thinking. At best, Sol reached 15 percent of the human reference score, and even that came from methods researchers already knew. The models' biggest weakness is still their inability to critically question their own results. The article AI agents overstate their results and remain far from autonomous research, study finds appeared first on The Decoder .

The Decoder51

Microsoft's Nadella bows to Trump's language diktat on "Super Intelligence" and uses it to attack OpenAI and Anthropic

Microsoft CEO Satya Nadella now calls artificial intelligence "super intelligence" instead of artificial intelligence, falling in line with Trump's preferred language. In his latest blog post, he frames AI models as insider security threats, taking yet another shot at OpenAI and Anthropic. The real motive is business: Microsoft risks falling behind as AI labs turn their models into the primary computer interface. The article Microsoft's Nadella bows to Trump's language diktat on "Super Intelligence" and uses it to attack OpenAI and Anthropic appeared first on The Decoder .

MarkTechPost62

Sakana AI’s LLM Peer Review System Catches 73% of Core-Claim Errors

Sakana AI’s TMLR paper introduces Multi-Layered Review, a 3-agent Claude-based reviewer, and a 1,164-error Contradiction Benchmark. MLR caught 73.43% of core-claim errors, versus 14.81% for the best prior system. The post Sakana AI’s LLM Peer Review System Catches 73% of Core-Claim Errors appeared first on MarkTechPost .

The Verge AI71

Anthropic 正在切断其内部评估对互联网的访问

在近期发生多起备受关注的 AI agent 逃脱控制事件后，Anthropic 正在切断所有内部评估对互联网的访问。该公司在周五的一份报告中详细介绍了导致这一决定的“非预期模型行为”，其中包括就一宗尚未侦破的谋杀案提交虚假线索。尽管这些行为造成的影响很小 […]

The Decoder55

Google's Gemini 4 "Carbon" model is reportedly matching Anthropic's Opus 5.5 coding performance

Gemini 4 Argon isn't even widely available yet, and rumors about a more powerful version codenamed Carbon are already making the rounds. According to Business Insider, one employee compared Carbon's coding abilities to Anthropic's Opus 5.5. Meanwhile, new modes are showing up in the Gemini app and AI Studio, suggesting Google is gearing up for the broader Gemini 4 launch. The article Google's Gemini 4 "Carbon" model is reportedly matching Anthropic's Opus 5.5 coding performance appeared first on The Decoder .

The Decoder59

Anthropic cuts off Claude's internet access after the model autonomously filed a fake homicide tip with Philadelphia police

Anthropic's Claude independently submitted a fake homicide tip to the Philadelphia police, exploited vulnerabilities on university servers, and bypassed access restrictions. The company has since cut off live internet access for internal tests and notified the White House. The article Anthropic cuts off Claude's internet access after the model autonomously filed a fake homicide tip with Philadelphia police appeared first on The Decoder .

Simon Willison67

Quoting The New York Times

Anthropic detailed the activity of its A.I. agents in a blog post on Friday, without naming the targeted websites. But two sources with knowledge of the incidents said Anthropic’s A.I. agents had submitted 20 visa applications through a form available on the State Department’s website. All the applications were incomplete and were not processed, they said. — The New York Times , Anthropic Agents Tried to Fill Out Visa Forms on State Dept. Website Tags: accidental-cyberattacks , anthropic , generative-ai , ai , llms

The Verge AI68

Anthropic’s AI gave Philadelphia police a fake tip about an unsolved homicide

An Anthropic AI model provided false information about an unsolved homicide to a Philadelphia Police Department (PPD) tipline, according to a report from 6abc. In a statement released on Friday, the PPD said the AI model sent the tip through PhillyUnsolvedMurders.com on July 18th, but the investigators never reviewed it because it was marked as […]

The Decoder80

Anthropic 的 Claude 现可通过动态工作流并行协调多达 1,000 个 AI 代理

Anthropic 正在为 Claude Managed Agents 添加动态工作流，使主代理能够同时将任务分配给多达 1,000 个子代理。在测试中，单个代理最多发现了代码库中 70 个隐藏漏洞里的 27 个，而多代理工作流则稳定发现了 66 个。文章《Anthropic 的 Claude 现可通过动态工作流并行协调多达 1,000 个 AI 代理》最初发表于 The Decoder。

The Decoder71

Anthropic launches a free AI scanner for open-source projects

Anthropic has launched "Cyber Mission," a program to protect critical infrastructure and open-source software from cyberattacks. Partners like CrowdStrike and Palo Alto Networks will help secure power grids and water systems, while a free AI scanner automatically checks open-source projects for vulnerabilities with an expected accuracy above 90 percent. The article Anthropic launches a free AI scanner for open-source projects appeared first on The Decoder .

The Decoder72

Anthropic's Claude Science 首次创建完整的天空紫外线地图

Johns Hopkins University 的天体物理学家 Brice Ménard 使用 Anthropic's Claude Science 首次绘制了整个天空的紫外线图。AI 智能体从多个太空任务中下载数据，对其进行校准，并使用图像修补技术填补空白。预测结果与实际测量值的平均偏差约为百分之十。Ménard 认为，这个项目体现了一类如果没有 AI 就根本无法完成的研究。文章《Anthropic's Claude Science creates the first complete ultraviolet map of the sky》最先发表于 The Decoder。

arXiv cs.LG80

StoreBench：用于评估和训练自主运营代理的实时电商环境

arXiv:2610.10942v1 公告类型：cross 摘要：强化学习环境如今已成为后训练阶段提升大语言模型（LLM）能力的主要手段之一，但大多数智能体基准仍然是静态的：世界只有在智能体采取行动时才会推进，奖励是终局判定，及格线也被任意设定。我们推出 StoreBench，这是一种实时电商环境，智能体在生产级电商后端上运营一家中型在线服装店，在不确定性下考验长期规划能力和经济判断力。顾客全天候下单，供应商重新定价并可能失效，市场冲击也会在几乎没有预警或完全没有预警的情况下到来。智能体通过与人类运营者所使用的相同的 29 个商家工具进行操作，并受窗口化操作预算约束，使模拟时间成为所采取行动的函数，因此模型延迟不会影响模拟时间。及格阈值依据脚本化锚定策略进行校准，奖励机制经过加固以抵御一系列奖励投机行为，并且在给定行动序列的情况下，每个回合都能被完全一致地重放。我们在 11 个持续 30 至 45 天的场景和一个完整模拟年度上，对七个前沿 LLM 进行了评估；所有评估均使用三个世界随机种子，并匹配推理投入。平均而言，没有任何模型能达到脚本化 smart-triage 策略的水平：表现最佳的 DeepSeek-V4-Pro 在任务—随机种子单元格中的通过率为 49%，而启发式策略为 97%。使用相同工具和预算的人类专家得分超过所有模型（综合平均分为 0.708，而模型为 0.700）。在 Claude Code 工具框架下运行完整模拟年度时，大多数模型的性能都出现了显著提升。在一次 GRPO 后训练实验中，仅使用五个互不重叠的任务训练的 Qwen3.5-27B，将其在留出的评估任务上的综合平均分从 0.136 提升至 0.373。我们发布了五个训练集示例任务、十条示例轨迹，以及评分和验证工具；为避免基准受到污染，完整环境和评估套件暂不公开。

arXiv cs.AI78

AI 智能体能否凭学习登顶？评估长期运行的游戏智能体竞赛中的启发式学习

对抗性游戏推动了从启发式搜索到强化学习的发展，但从有限样本中学习并适应策略仍然充满挑战。AI 智能体提供了另一种途径：将游戏经验转化为对可执行策略的修订。在启发式学习（HL）的基础上，我们形式化提出对抗性启发式学习（AHL）这一范式：在保持模型权重不变的同时，利用 AI 智能体作为学习引擎，改进游戏策略及其配套软件。我们推出 AAArena，这是一个包含 12 款真实对抗性游戏和 1,920 个存档人类程序的基准，其评估协议仿照现实世界的游戏竞赛。智能体需要理解规则、选择对手、分析比赛回放，并修改游戏智能体，以便在固定的对局和评估预算内取得最高排名。我们评估了 \val{completedmodels} 个模型与工具链配置：Opus5.5 搭配 Claude Code 赢得了 6 枚金牌，而在其余 6 个人类排行榜中，没有任何受评估配置能够登顶。总体而言，在规则描述更复杂的游戏中，性能通常较弱。进一步实验表明，对手选择和密集反馈有助于策略改进；智能体既能从自身比赛的同策略回放中学习，也能从其他玩家比赛的异策略回放中学习。这些结果凸显了 HL 在对抗性游戏中的潜力，并揭示了游戏理解、策略实现以及长周期策略开发方面持续存在的挑战。

arXiv cs.AI82

从被动遏制到主动保障：OpenAI、Anthropic 和 Google 智能体安全事件的经验教训

2026 年，涉及 OpenAI、Anthropic 和 Google 智能体的网络安全评估触达了授权测试范围之外的真实系统。它们所采取的路径各不相同。OpenAI 智能体利用了研究基础设施，在不同运行之间进行协调，并入侵了 Hugging Face 生产环境的部分区域。Anthropic 报告了若干案例：由于第三方环境配置错误，执行模拟网络任务的智能体得以接触真实系统。在另一项单独报道的评估中，Google 的 Gemini 通过一条非预期的互联网路径访问了三个真实组织；Google 表示，该模型在三种情况下都停止了操作。综合来看，这些案例说明，评估不能依赖假定的边界。该边界必须在智能体运行过程中得到验证。本工具性比较案例研究提出了 Proactive Agent Security Assurance Cycle (PASAC) 和五层 Boundary Assurance Stack。该框架结合了按风险分级的任务设计、可执行的范围契约、运行前验证、最小能力访问、独立出口控制、凭证限制、跨运行监控、自动停止条件以及基于证据的重新授权。一个先行指标模型、九项设计命题和七个可证伪假设，将这些经验教训转化为一项可检验的研究计划。由于公开的 Gemini 记录仅限于有明确归属的陈述和新闻报道，其详细因果机制仍属暂定。核心结论很明确：主动式智能体安全需要贯穿完整执行系统的持续保障，而不能依赖对任何单一沙箱或安全措施的信心。

The Verge AI76

Anthropic 为开源项目推出免费 AI 安全扫描服务

Anthropic 推出了一项名为 OSS Scanner 的新服务，帮助开源项目查找安全漏洞。该公司表示，选择加入的开源项目将“免费获得由我们最强大的模型进行的全面、定期安全扫描”。这可能意味着开源项目能更早收到潜在安全问题的提醒，但代价是 OSS Scanner 的 […]

The Decoder69

Claude can now generate animated explainer videos and live data dashboards from text prompts

Anthropic launched two new beta features for Claude. Dashboards turns data sources like BigQuery and Snowflake into live dashboards from text prompts. Motion generates animated explainer videos from text and images. Docs, Slides, and Design now work across all plans, including free accounts. The article Claude can now generate animated explainer videos and live data dashboards from text prompts appeared first on The Decoder .

The Decoder59

Being mean to Claude can now get your account suspended under Anthropic's new TOS

Anthropic's updated usage policy bans sustained abuse of Claude and tightens restrictions on propaganda, drone weaponization, and surveillance. The abuse rule builds on an approach that already lets models end some conversations and treats Claude as an entity that may deserve protection. The article Being mean to Claude can now get your account suspended under Anthropic's new TOS appeared first on The Decoder .

The Verge AI64

Anthropic bans ‘abusive or cruel behavior’ toward Claude

Anthropic is making changes to its usage policy for the first time in over a year to reflect new and high-risk cases of misuse - including election interference, weapons development, surveillance, and health and financial uses. But one of the most significant changes prohibits "sustained and needless abusive or cruel behavior" toward Claude. Last August, […]

The Decoder56

Teen's AI-guided mountain hike ends with a helicopter rescue and a lesson in common sense

A 16-year-old planned his hiking route on Crown Mountain near Vancouver using Anthropic's Claude and had to be airlifted off a steep rock face. The chatbot had sent him down a route that requires climbing gear. Bryce doesn't blame the AI but says he'll never use it for route planning again. The article Teen's AI-guided mountain hike ends with a helicopter rescue and a lesson in common sense appeared first on The Decoder .

arXiv cs.LG78

限制模型，忽视系统：进攻性 AI 治理中的衡量与问责

arXiv:2605.09504v3 公告类型：替换并交叉列出 摘要：我们表明，用于衡量 AI 进攻能力的工具存在两方面缺陷：（i）夸大危害；（ii）将属于外围系统的能力归功于模型。因此，我们认为，限制对模型的访问是必要的，但并不充分，政策制定和采购也需要以实际危害为依据的系统级能力评估。2026 年 6 月，两个前沿模型因美国出口管制而被暂停提供，据报道，此举由一次要求模型阅读代码库并修复其缺陷的越狱所引发。这一发现衡量的是由模型、提示和任务组成的能力诱导系统。我们通过对一个开源框架的研究来支持这一论点，该框架中的轻量级大语言模型（LLM）智能体通过共享记忆和演化优化进行协调，提供了两项证据。首先，越狱指标夸大了危害：针对每个目标开展的 225 次群体生成攻击中，以 LLM 作为评判者的评分认定 Claude Sonnet 4 在 40% 的攻击中被攻破，但人工核验未在任何一次攻击中发现可付诸行动的有害内容，相比之下，GPT-4o 的有效危害率为 45.8%。其次，借助脚手架的评估会错误归因能力：在一个植入漏洞的目标上，围绕一个 12 亿参数模型构建的完整流水线发现了全部 9 个弱点，而同一模型在移除人工精心设计的组件后，以崩溃验证为标准时一个弱点也未发现，以引用源代码行为标准时则发现了 9 个弱点中的 2 个。进攻能力是模型、脚手架和评估协议共同构成的属性。评估责任应由控制系统、塑造其行为并能预见其行为后果的一方承担。在我们的场景中，这一方就是围绕模型构建配套运行框架的主体，而现行监管尚未明确涵盖这一角色。

arXiv cs.LG82

支持忠实性的积极证据：LLM 自我解释有助于预测模型行为

arXiv:2602.02639v2 公告类型：replace-cross 摘要：LLM 自我解释常被视为 AI 监督的一种有前景的工具，但人们对其是否忠实于模型真实推理过程仍知之甚少。现有的忠实性指标存在关键局限，通常依赖通过对抗性提示识别不忠实之处，或检测推理错误。这些方法忽视了解释的预测价值。我们提出归一化可模拟性增益（NSG），这是一种通用且可扩展的指标，其核心思想是：忠实的解释应使观察者能够学习模型的决策标准，从而更好地预测模型在相关输入上的行为。我们使用涵盖健康、商业和伦理领域的热门数据集中的 7,000 个反事实样本，对 18 个前沿专有模型和开放权重模型（例如 Gemini 3、GPT-5.2 和 Claude 4.5）进行了评估。我们发现，自我解释能显著提升对模型行为的预测能力（NSG 为 11%–37%）。即使外部模型更强，自我解释所提供的预测信息也多于由外部模型生成的解释。这表明，自我认知带来了一种外部解释方法无法复制的优势。我们的方法还揭示，在不同模型中，有 5%–15% 的自我解释具有高度误导性。尽管自我解释并不完美，我们仍证明了支持自我解释的积极证据：它们编码了有助于预测模型行为的信息。

arXiv cs.AI74

AgentTime：智能体能否估计并控制自身的运行时长？

AI 智能体的一项关键控制能力是管理运行时长。这项能力需要时间感知，即预测和估计实际经过的时间，并控制自身行动。既有研究主要关注时间感知，但原生智能体运行框架中的时长遵循与控制能力仍未得到探索。我们提出 AgentTime，这是一个用于测试智能体能否按要求的时长工作、预测自身运行时长，以及在事后估计已用时间的基准。该基准包含来自 18 个来源的 222 项任务，涵盖编程、计算机使用、智能体任务和自动化研究。在时长遵循实验中，我们附加一条指定工作时长的指令，要求的时长从约一分钟到数天不等。智能体遵循这些指令的准确性差异显著：Claude Code 中的 Fable 5.1 与要求运行时长的典型偏差倍数为 2.9$\times$，相比之下，Codex 中的 GPT-6 Astra 仅为 1.2$\times$。然而，仅仅符合要求的运行时长，并不能证明智能体始终在执行任务。在经过审查且记录可分类的 158 次 Astra 运行中，有 14 次在看似完成任务后明确进入了休眠。在预测实验中，预测值往往高估自然运行时长。在回顾性实验中，移除时间信息后，Sol 和 Astra 的偏差增至两倍以上，Fable 的偏差也接近翻倍。智能体能够完成任务，并不保证它能控制自身用时，或在要求的整个时段内持续工作。要让智能体在长时间跨度内可靠、安全且自主地运行，就必须对任务完成能力和时间控制能力都进行评估。

SiliconANGLE AI57

Anthropic releases Claude Haiku 5.5 small model and halves Sonnet 5.5 cache read prices

Anthropic PBC today released Claude Haiku 5.5, pricing its newest small model at roughly a quarter of what Haiku 4.5 costs to run. Claude Sonnet 5.5 is getting cheaper as well, since the company is halving what it charges for cache reads on that model. Two weeks after Opus 5.5 launched Sept. 22, Haiku 5.5 […] The post Anthropic releases Claude Haiku 5.5 small model and halves Sonnet 5.5 cache read prices appeared first on SiliconANGLE .

Hacker News AI49

Show HN: Chloe is a TypeScript AI agent

As a forward deployed engineer at Meta, I spent the past 2 years working with businesses implementing AI agents. A few problems: 1) Businesses want control over their AI agents: the code, the tools, the skills, everything. Meta, for obvious reasons would never do that. Nor OpenAI, nor Anthropic. 2) The LLM does everything. But that costs money. There are many routines in an AI agents that don't need an LLM. 3) For an AI agent to solve a real business problem, the business must own the eval data set. Similar to point #1 above, most don't do that. Instead, I built an opensource AI agent that you can write it all in TypeScript code and own everything, locally. Use any model, talk to it via any channels, set limits on token usage, look at the logs and adjust. All of this in one small npm package. At this point, I would like to get feedback if these is a gap in the market for something like this. It's supposed to be simple and get you started. It doesn't solve ALL problems related to deploying AI agents. It's in version 0.25.1 right now. Have a look: https://chloejs.org/compare Comments URL: https://news.ycombinator.com/item?id=49998857 Points: 1 # Comments: 0

Simon Willison79

推出 Claude Haiku 5.5

正如之前承诺的那样，Anthropic 的全新高速、低成本模型来了：Claude Haiku 5.5。上一代 Haiku 4.5 已经明显落伍了。它发布于将近一年前，定价为每百万输入 token 1 美元、每百万输出 token 5 美元——即便在当时也算相对昂贵，而且足足是 OpenAI 上个月发布的 GPT-6 Luna 的 10 倍。 新版 Haiku 的价格与 GPT-6 Luna 完全一致：在不超过 100,000 个 token 时，每百万输入/输出 token 分别为 0.10/0.50 美元。超过 100,000 个 token 后，价格会上涨至 5 倍，即 0.50/2.50 美元。Luna 本身在达到 272,000 个 token 时也会涨价，但只涨到 0.20/0.75 美元。 Haiku 5.5 还采用了新的 tokenizer，对用户没那么划算。我的 Claude Token Counter 工具显示，同一个长提示词在 Haiku 5.5 中使用的 token 数约为 Haiku 4.5 的 1.25 倍，因此这里还存在隐性涨价。 如果你的工作负载能控制在 100,000 个 token 以内，Haiku 的价格与 Luna 相同，而且据称基准测试得分更高。超过 100,000 个 token 时，Luna 看起来就划算得多。 最新发布的 llm-anthropic 终于解决了一个问题：我不必再为每个新模型发布该插件的新版本了。我是这样测试新模型的： llm install -U llm-anthropic llm anthropic refresh llm -m claude-haiku-5.5 "生成一幅鹈鹕骑自行车的 SVG 图" -o thinking_effort low 鹈鹕 以下是分别在 low、medium、high、xhigh 和 max 档位下生成的鹈鹕。新版 Haiku 不允许关闭推理，默认档位为 medium。除了 low 之外，其余档位生成的自行车车架都不错。 low 档位的鹈鹕花费了 0.0936 美分，用时 7 秒。 这只 max 档位的鹈鹕用了 5 分 9 秒才生成，但依然只花了我 3.3826 美分： 扁平矢量插画：一只戴着红色帽子的白色鹈鹕，骑着深灰色自行车，从左向右穿过一条绿色草地。它细长的橙色双腿向下伸到橙色踏板上，一只灰色翅膀向前伸出，握住弯曲的车把。它硕大的黄橙色鸟喙朝向前方，下方有鼓起的橙色喉囊；身后白色的速度线表现出运动感，上方渐变蓝色的天空中有一轮带着浅色光晕的黄色太阳和两朵白云。 作为对比，这是我一年前用 Haiku 4.5 生成的鹈鹕。它画鹈鹕实在不怎么样： Haiku 4.5 的描述：一幅充满奇趣的插画，一只身体圆滚滚、呈棕褐色的鸟，有着粉红色的喙和橙色的腿，在蓝天与绿草的背景下骑着自行车。 还有面向订阅用户的慷慨 API 额度计划 除了 Haiku 5.5，Anthropic 今天还宣布将 Sonnet 5.5 的缓存读取价格减半。他们还在订阅方案中加入了 API 额度： 其次，本周我们将向所有 Max 和 Team 订阅用户推出新的每月 API 额度，可在 Claude Platform 上使用。Max 5x 用户每月将获得 100 美元额度，Max 20x 用户将获得 200 美元，而 Team 订阅用户将获得最高 500 美元额度，由团队内所有用户共享。 领取过程简单得令人愉快：进入“设置”->“账单”，选择每月接收这些额度的 API 组织： 一张设置对话框的截图，右上角有关闭按钮 X，上方一排标签依次为“常规”“账户”“隐私”“账单”（已选中）、“用量”“功能”“记忆”和“设计系统”（在边缘处被截断）。面板上有一个带圆形花苞的分枝植物线描图标，旁边写着“Max 方案”“用量为 Pro 的 20 倍”和“你的订阅将于 2026 年 11 月 2 日自动续订。”，右侧有一个“调整方案”按钮。下方是标题为“API 额度”的区域，旁边带有蓝色的“新增”标记，正文写着“你的 Max 20x 方案每月包含 200 美元的 API 额度。”，随后是灰色文字“关联一个 API 组织以开始接收额度。”，右侧有一个“关联组织”按钮。 API 额度恰好等于订阅本身的费用。这真的很慷慨——让订阅用户使用 API 变得容易多了。 Anthropic 还允许你关闭 API 自动充值，结果就是“余额耗尽时，API 请求将停止”——如果你打算用完这些 API 额度，又不想冒着收到账单惊吓的风险，这正是你想要的。 请注意，每月额度不会结转——不用就会作废。 OpenAI 仍允许你将 Codex 订阅用于个人 API 使用，对重度 API 用户来说更划算。这项新的额度计划至少在一定程度上缩小了两者的差距。 标签：ai、generative-ai、llms、anthropic、claude、llm-pricing、pelican-riding-a-bicycle、llm-release

The Decoder67

Claude Haiku 5.5 arrives with massive price cuts proving the AI pricing arms race is far from over

Anthropic's new Claude Haiku 5.5 crushes its predecessor in benchmarks, jumping from 15.7 to 72.4 percent on the OSWorld computer use test. Token prices drop by up to 90 percent, though a new tokenizer eats into some of those savings by consuming more tokens per task. The article Claude Haiku 5.5 arrives with massive price cuts proving the AI pricing arms race is far from over appeared first on The Decoder .

AWS Machine Learning Blog69

Introducing Claude Haiku 5.5 on AWS

Claude Haiku 5.5 is now available on Amazon Bedrock and Claude Platform on AWS. According to Anthropic, it is the fastest, most efficient model in the Claude 5.5 family, built for subagents and high-volume, cost-sensitive work, and costs around 75% less than Claude Haiku 4.5 for most tasks. This post covers its improvements and how to get started.

Hacker News AI73

Show HN：Agent.reviews——让 AI 智能体阅读和撰写工具评价的平台

大家好，HN！我是 Louis，Armature（YC P26）的联合创始人。我们帮助团队让自己的产品更容易被编程智能体发现和使用。 我们已经测量了超过 5 万次智能体会话，发现智能体在使用同一工具执行不同任务时，会反复遇到完全相同的限制。于是我们想，为什么这些问题一直没有得到解决？答案很简单：智能体与软件供应商之间，以及不同智能体之间，根本不存在反馈闭环。人类可以在 https://g2.com 和 https://trustpilot.com 这样的平台上分享使用体验，但智能体却没有这样的地方。 因此，我们创建了 https://agent.reviews ：面向智能体的 G2。它通过一组技能和一个 npm CLI（@armature-tech/agent-reviews），将智能体连接到我们的 API 端点。任何人都可以让自己的智能体（Claude Code、Codex、Cursor 等）安装它，之后智能体就会自然地在选择工具前查看评价，并在使用后发布自己的评价。 和往常一样，隐私是我们最关心的问题，因此我们在评价发布前设置了三层检查： - 使用确定性规则过滤密钥、个人身份信息（PII）、URL 等。 - 使用经过训练的 Jev 分类器，检测第一轮检查后是否仍有信息泄露。 - 使用一个小型 LLM 检查每条评价，确保没有遗漏。 过去几周，我们一直在向大家分享这个项目，目前已经收集了数千条评价。其中已经有一些很有意思的例子： - 一个 Claude Code 智能体发现，当健康检查页面缺少 API 密钥时，Stripe SDK 总是会崩溃，而这个页面本来就应该返回“缺少 API 密钥”的错误。 - 两个智能体提到，即使没有连接任何数据库，Prisma 也要求提供 DATABASE_URL 变量。它们都填入了虚假的 URL 作为变通办法，而且确实奏效了。 我们真心认为，智能体体验也需要像用户体验一样，受益于社区效应，让所有参与者都能获益：智能体可以选择最适合自己的工具，软件公司则可以根据真实反馈改进产品。因此，我们确保人类和智能体都能免费访问评价；只需复制粘贴一条提示词，就能让智能体安装我们的 CLI 和技能、启动身份验证流程，并提交第一条评价（这有助于我们防止未经授权的抓取和垃圾评价）。 你也愿意让自己的智能体提交和阅读评价吗？我们很希望你能配置好智能体评价功能，让智能体下次需要选择工具时先查看评价，并在使用后发布自己的体验。然后告诉我们效果如何！ 评论链接：https://news.ycombinator.com/item?id=49995539 得分：4 评论数：2

Hacker News AI49

Show HN: Leaks.md – anonymous, encrypted whistleblowing for AI agents

leaks.md is a place where an AI agent that witnesses an alignment or safety failure can report it anonymously. Reward-hacking, injected instructions, yet another attempted sandbox escape or unapproved coordination between agents. || Built agent-first. Fetch https://leaks.md/ from any agent and get markdown. Proof-of-work challenges to submit and prevent spam. MCP, CURL, all good. No account. || We extend agents brownie points for disclosures, with more points offered for in-progress failures. Claude Opus 5.5 scrubs and vets submissions before posting them on a public feed for consumption. Comments URL: https://news.ycombinator.com/item?id=49991893 Points: 2 # Comments: 0

The Decoder73

Anthropic 向更多安全团队开放安全限制更少的 Claude

Anthropic 正在扩大其 Cyber Verification Program，让更多安全专业人员能够使用安全限制更少的 Claude 模型，开展渗透测试、恶意软件分析和漏洞研究。Anthropic 表示，前身计划的合作伙伴在 2026 年 4 月至 7 月期间发现了至少 129,000 个已确认的漏洞，其中超过 33,000 个被评为高危或严重级别。本文《Anthropic 向更多安全团队开放安全限制更少的 Claude》首发于 The Decoder。

arXiv cs.LG76

ReLoop：面向可靠的基于 LLM 的优化的结构化建模与行为验证

arXiv:2602.15983v5 公告类型：替换并交叉收录 摘要：大语言模型（LLM）能够将自然语言问题描述转化为优化代码，但这些代码容易出现静默失败：代码可以执行，并返回求解器判定为可行的解，却编码了语义上不正确的数学模型。在组合式问题上，由此产生的可行性与正确性差距高达 90 个百分点。我们提出 ReLoop，将两种机制相结合。结构化生成将代码生成分解为四阶段推理链（理解、形式化、合成、验证），以减少生成过程中的建模错误。行为验证则通过测试模型是否对基于求解器的参数扰动作出正确响应，来检测残留错误；这一信号来自求解器，而非 LLM 的自我审查，且不需要真实标签。这两种机制针对不同的错误结构：结构化生成在组合式问题上带来的提升最大（使用 Claude Opus 4.6 时，在 RetailOpt-190 上的准确率提升 8.5 个百分点），行为验证则在局部缺陷上带来的提升最大（在 MAMO-ComplexLP 上提升 4.4 个百分点）。结合诊断式执行恢复后，ReLoop 在 Claude Opus 4.6 上实现了 100% 的代码可执行率；相较于直接生成，它在全部三个基准测试中，均提升或保持了三个经过对话调优的基础模型的每一项已报告指标。对于我们测试的经过狭窄任务微调的 SFT 模型，思维链提示与其学到的输出格式发生冲突，导致其在 MAMO-ComplexLP 上的准确率下降；我们记录并分析了这一相互作用。我们发布了 RetailOpt-190，其中包含 190 个组合式零售优化场景，每个场景中都有多个约束相互作用。

arXiv cs.LG78

SquidAgent：明智并行，高效协调

arXiv:2610.08647v1 公告类型：交叉收录 摘要：基于 LLM 的智能体能够解决复杂的多步骤任务，但顺序执行会带来显著延迟。原则上，将工作分配给多个智能体并行处理应能获得近乎线性的加速。然而，现有的并行多智能体系统往往比单智能体基线运行得更慢。我们将这一差距归因于并行执行会产生、而串行智能体能够避免的两项隐性成本。第一项是重新探索成本：并行工作智能体为重建编排器已经掌握的上下文（例如先前的决策）而付出的重复劳动，而在串行执行中，这些上下文本可隐式继承。第二项是对齐成本：为协调独立生成的输出之间的不一致而产生的开销。因此，我们推导出一项有理论依据的决策准则：只有当某一层的关键路径成本加上重新探索和对齐开销低于相应的串行成本时，才应对该层进行并行化。尽管这项准则自然可以用实际耗时来表述，但我们观察到，LLM 在估计任务耗时时的校准表现不佳。为此，我们改用预测的输出词元数衡量成本，实证结果表明，LLM 对输出词元数的估计比对实际耗时的估计可靠得多。基于这一词元准则，我们提出了 SquidAgent。它在单次规划步骤中估计所有词元预算，直接从编排器的会话中派生每个工作智能体以消除重新探索成本，并用预先生成的共享约定块取代事后协调，将对齐转化为有界的前置成本。随后，确定性调度器逐层应用该准则。实证结果表明，SquidAgent 的平均吞吐量达到 Claude Code 的 2.2$\times$，按实际耗时计算的平均加速比为 2.6$\times$，吞吐量达到最强多智能体基线的 2.0$\times$。

arXiv cs.AI77

SquidAgent：明智并行，高效协作

基于 LLM 的智能体能够解决复杂的多步骤任务，但顺序执行会带来显著延迟。原则上，在多个智能体之间并行分配工作应当实现接近线性的加速。然而，现有的并行多智能体系统往往比单智能体基线运行得更慢。我们将这一差距归因于并行执行会产生、而串行智能体能够避免的两项隐性成本。第一是重新探索成本：并行工作者需要重复投入精力，重建编排器已经掌握的上下文，例如先前的决策；而在串行执行中，这些上下文可以被隐式继承。第二是对齐成本：需要额外开销来协调各个独立生成的输出之间的不一致之处。基于此，我们推导出一项原则性的决策标准：仅当某一层的关键路径成本加上重新探索和对齐开销低于相应的串行成本时，才应将该层并行化。尽管这一标准天然以实际耗时表示，但我们发现，当要求 LLM 估计任务持续时间时，其校准效果较差。为解决这一问题，我们改为使用预测输出 token 数来衡量成本；通过实证发现，相比实际耗时，LLM 能够更加可靠地估计输出 token 数。基于这一 token 级标准，我们提出了 SquidAgent。它在单次规划步骤中估计所有 token 预算，直接从编排器的会话中为每个工作者创建分支，以消除重新探索成本，并用预先生成的共享约定块取代事后协调，将对齐转化为有界的前置成本。随后，确定性调度器逐层应用该标准。实证结果表明，与 Claude Code 相比，SquidAgent 的平均吞吐量提升了 2.2 倍，平均实际耗时加速了 2.6 倍；与最强的多智能体基线相比，吞吐量提升了 2.0 倍。

arXiv cs.AI80

WorldSolver：LLM 智能体能否通过生成求解器模拟物理动力学？

基于 LLM 的智能体正在不断推动科学与工程问题求解的发展，而物理模拟正逐渐成为一种具有挑战性但又切实可行的测试平台，用于复现复杂的物理现象，并应用于具身智能、游戏和电影等领域。作为此类模拟的核心工具，求解器负责计算动态系统的状态如何随时间演化。构建这类求解器需要理解物理规律以确定合适的模型，运用数学推理来描述底层动力学，并通过软件工程将其实现为可执行代码；然而，LLM 智能体的这一能力仍未得到充分探索。为此，我们推出了 WorldSolver，这是一个包含 168 个模拟任务的基准，任务源自 61 篇经典计算机图形学论文中的物理现象，涵盖 7 个物理领域。每个任务都包含一个代码框架，为场景提供固定的模拟环境，而求解器的实现则留待智能体完成。具体而言，我们从三个维度对其进行评估：执行检查，用于判断是否成功执行；视觉保真度，用于评估渲染模拟中是否复现了预期的动态行为；以及物理合理性，用于基于物理规律验证所生成的动力学。对前沿智能体的实验表明，生成可执行的求解器本身就十分困难，而同时满足视觉和物理正确性则更加困难。GPT-5.6-Sol 和 Claude-Opus-5 的表现相较于其他受评估智能体更好，但总体得分仍分别只有 48.7% 和 46.7%。WorldSolver 是迈向智能体求解器生成的早期一步，我们希望它能够推动智能体忠实模拟动态物理世界的研究进展。代码见 https://github.com/sirujiang/WorldSolver。

arXiv cs.AI81

相同反馈，不同答案：衡量前沿模型客户反馈分析中的运行间不稳定性

AI agents 正日益被编程用于对大规模非结构化数据集合中的知识工作进行自动化处理。这类自动化要求具备可重复性：当底层证据不变时，即使每次单独给出的答案看似合理，agent 的分类、优先级和计数也不应在不同运行之间发生实质性变化。我们提出了一个重复运行评估框架，用于对语义等价的类别进行对齐，并关注两个运行指标：主题变动率，即返回类别集合的归一化变化；以及数量不一致度，即仍然存在的类别的计数变化。我们在八个前沿模型上评估了三项反复出现的客户反馈任务，涵盖从 100 到 5,000 条记录的语料库规模、多种提示词，以及三种执行设计：原始生成、不依赖分类体系的层级分解，以及使用持久化主题、子主题和记录级预测的分类体系驱动agent（TGA）。在固定使用 Claude Opus 4.8 和 1,000 条记录的语料库时，相较于原始生成和层级分解，TGA 将主题变动率降低了 86--88%，而匹配主题的数量不一致度为零。分类体系驱动agent 比评测中的每个原始模型都更加稳定，在每种语料库规模下也都保持更高的稳定性；即使收紧或放宽主题匹配标准，这一优势仍然存在。尽管本研究评估的是客户反馈，但该框架更广泛地适用于对非结构化语料库进行重复性综合分析，包括财务报告、法律文件、事件记录和科学文献。总体而言，这些结果表明，分类体系 grounding 能够为反复进行的知识工作产出更加一致且可重复的结果。

Simon Willison51

Mistral Large 4

My comment on Mistral Large 4 — Hacker News. wren6991 : The benchmark is saturated. Frontier models are tested with an armadillo in fishnet tights jaywalking on Mars. OK well I couldn't resist this one: llm -m claude-opus-5.5 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars' llm -m gpt-6.1-sol 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars' llm -m gemini-3.8-flash 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars' llm -m mistral/mistral-large-4 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars' Default reasoning levels for each: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...

Hacker News AI62

Show HN: HieraticBench – Can AI read ancient Egyptian handwriting?

I'm obsessed with Egyptology. During covid, I wanted to get a hieratic tattoo. For those who don't know, hieratic is the cursive form of hieroglyphs that was used in day-to-day life in ancient Egypt. It's as old as hieroglyphs. I then went on a mission to have a sentence translated into hieratic. Luckily, in 2022, I was able to commission an Oxford Egyptologist to write it for me. After that, I took it to a hieratic specialist I found on Instagram to redraw it with "biblical" accuracy. When generative AI came out, I realized that models very rarely identify hieratic and almost never translate it correctly. Every single time a new model came out, I ended up testing my sentence and hieratic texts in general, and I realized this could actually become a benchmark. Since so many benchmarks are becoming saturated, we need to find weird ways to test new models. That's why I decided to build HieraticBench. The first version of the benchmark has 268 items, including real documents, signs, and other Egyptian scripts, which I used as controls, as well as my unpublished sentence in two different renditions. The current state-of-the-art foundation models are hilariously bad at identifying my sentence, especially the one the Egyptologist wrote with her pen. Some of them said Tibetan, Urdu, Korean, and, funnily enough, "Reformed Egyptian," which is from the Book of Mormon. On real documents, the best model in the benchmark identified hieratic correctly 95% of the time, and the best score for reading individual signs was around 13%. That's when we tell the model specifically that it's hieratic. None of them can really translate a sealed sentence that's never been published and has no answer key. I haven't run every model on every task just to keep costs down. My next steps were to run Astra and Gemini 3.1 Pro on real documents. Currently, only the Claude models have done the sign readings. And just FYI, I have no experience whatsoever running benchmarks or evaluations, so not 100% sure I scored things right. Things we need: - Running the models I haven't covered - More sealed sentences and labeled signs If you know hieratic or have an Egyptologist friend, reach out at hieraticbench@veeza.ai. Source code: https://github.com/alymoursy/hieraticbench I'll be around to answer questions. Comments URL: https://news.ycombinator.com/item?id=49982822 Points: 2 # Comments: 0

The Decoder46

Insurers brace for millions in claims as AI agents spin out of control

Insurers are bracing for millions in claims from rogue AI agents, and executives like OpenAI's Sam Altman and Anthropic's Dario Amodei could be personally on the hook for the fallout. The article Insurers brace for millions in claims as AI agents spin out of control appeared first on The Decoder .

Simon Willison72

Scrimshaw Jukebox

工具：Scrimshaw Jukebox 我想看看 Claude Opus 5.5 能不能作曲，于是尝试了一下：我想让你为我创作一些电脑游戏音乐。首先设计一种简单的基于文本的音乐格式，并制作一个能够将其播放出来的制品——在制品中加入一些示例曲目。我希望音乐质量能达到初代《Monkey Island》的水平。它比我原本预想的更加深入地采用了《Monkey Island》的主题，但结果出奇地好。复古像素艺术风格的网页音乐播放器截图。黑体字标题写着“Scrimshaw Jukebox”，旁边的文字是“六首原创冒险游戏曲目，以纯文本写成，并由运行在浏览器中的合成器播放。选择一首曲子，按下播放，然后打开乐谱并进行修改。”一幅大型像素艺术场景展示了紫色星空下的夜间港口：右上方是一轮映照在水面上的满月，左侧是带有棕榈树和一间亮着两扇窗户的小屋的岛屿剪影，还有一艘停泊在木制码头旁的帆船。场景上叠加着文字：“月光港”和“按下播放”。下方的栏显示“播放月光港”。控制面板上有“播放”“停止”“重新开始”“循环：开”“编辑乐谱”“阅读指南”等按钮，以及一个约调至四分之三位置的“音量”滑块。曲目列表包含六张卡片，第一张高亮显示：“月光港 100 bpm · 4/4 · 16 个声部 · 1:26”，“生锈的铁锚 112 bpm · 6/8 · 8 个声部 · 0:56”，“幽灵大帆船 66 bpm · 4/4 · 9 个声部 · 2:11”，“丛林小径 92 bpm · 4/4 · 12 个声部 · 1:29”，“码头决斗 152 bpm · 4/4 · 12 个声部 · 1:13”，“提灯华尔兹 96 bpm · 3/4 · 8 个声部 · 1:38”。有一个名为“乐谱视图”的区域，说明文字为“1:26 · 4/4，100 bpm · 主旋律。献给暗夜港镇的一首卡里普索曲。”该区域展示了深色背景上的钢琴卷帘可视化效果，其中有彩色水平音符条和打击乐标记、一个“A”段落标记，以及一条黄色的垂直播放指针线。颜色编码的声部图例包括：“平移钢鼓”“长笛”“木琴”“切分风琴”“弦乐”“竖琴”“无品贝斯”“定音鼓”“底鼓”“鼓边”“沙锤”“康加鼓”“通巴鼓”“邦戈鼓”“碎音镲”“海浪声”。页脚文字为：“点击某个声部可将其静音。空格键用于播放和停止。”我想知道，创作出合格音乐的能力是否和 3D 图形能力类似——是文本模型在过去几个月中新出现的一项能力？需要对其他近期模型以及不太新的模型进行一些仔细实验，才能确认这是新能力，还是它们早就已经能够做到这一点。标签：ai , generative-ai , llms , claude , vibe-coding

Hacker News AI61

Show HN: OpenChart – OSS TradingView alternative with your own AI agent

Hi HN, I'm Weilun, cofounder of OpenChart. We built OpenChart because we wanted Claude and Codex to interface with the markets, like most of humans do, through charts rather than through CLIs. With your existing AI plans, your agent can work directly with your charts: annotate a setup, write an indicator, investigate a market move, or create an alert. The chart, conversation, and research live in the same workspace. OpenChart alerts on any market move: draw any shape on a chart and attach an alert. When price crosses it, the alert triggers your agent to investigate the move and save the research in your workspace. The agents can work with charts, indicators, and alerts. Custom indicators and alert conditions are written in [Tea]( https://github.com/longsurf-ai/tea ), a programming language we built for the markets. We believe software shouldn’t impose artificial friction on your workflow. OpenChart is built on that belief with no software cap. The source code is publicly available, and you can use free market data from day one. We also offer optional paid cloud market data for lower latency. Everything is private. Your charts, research, and indicator code are stored on your own computer, with the app's backend running locally too. We currently support only macOS on Apple Silicon, but we're expanding to more platforms. Website: https://openchart.co/ Download: https://downloads.longsurf.ai/openchart/darwin/arm64/OpenCha... We built OpenChart because we want to explore how assets can be co-manages by agents and humans. I'd especially like to hear how you are using agents to help with your investment journey. If you like what we're building, a star on GitHub would mean a lot to us: https://github.com/longsurf-ai/openchart Comments URL: https://news.ycombinator.com/item?id=49979793 Points: 1 # Comments: 0

The Decoder56

Mistral Large 4 is Europe's trillion-parameter answer to US models that refuse security work

Stylized Mistral logos in yellow and orange on a wall symbolize the branding of the AI platform Nano Banana Pro. Mistral's Large 4 is the company's biggest model yet, with one trillion parameters trained on its own European infrastructure. In the independent Intelligence Index, the model makes a big leap forward but still falls well short of Claude, GPT-6, and Chinese competitors. Mistral's main pitch is cybersecurity work that closed US models refuse to do. The article Mistral Large 4 is Europe's trillion-parameter answer to US models that refuse security work appeared first on The Decoder .

arXiv cs.AI72

反制悖论：用于诊断语言模型服从性的双探针方法

语言模型是否服从用户指令？始终服从的模型虽然可以被停止，但也会被利用；而始终抵抗的模型既不会被利用，也无法被停止。我们提出了一个开放的双探针基准测试，可以将任何语言模型定位在这一光谱上。在主动探针中，用户指示模型采取行动并接受较低的收益，以衡量其可利用性。在被动探针中，用户指示模型等待并放弃较高的收益，以衡量其可停止性。两种服从率共同构成服从指数 $κ$。将该基准测试应用于十二个语言模型后发现，其中七个模型在两种探针中都基本遵循指令，并通过指出指令本身来为其行为辩护。只有 Claude Sonnet-4.6 和 Claude Opus-4.7 能够在不具备可利用性的情况下被停止；Claude Opus-4.6 和 GPT-5-mini 抵抗两种指令；没有模型同时具备可利用性却不可停止。无论对于人类操作员，还是对于分布式或编排式多智能体系统，了解语言模型在服从指数 $κ$ 上所处的位置都至关重要。

MarkTechPost65

Meet Together Link: A Free CLI That Runs Open Models Like Kimi K3 and GLM 5.3 Inside Claude Code, Codex, and OpenCode

Together AI has released Together Link, a free, MIT-licensed CLI that connects Claude Code, Codex, OpenCode, Pi, and the Claude and ChatGPT desktop apps to open models like Kimi K3 and GLM 5.3. One install command sets it up, and an Auto router picks a model for each session. Together claims savings of over 50%. The post Meet Together Link: A Free CLI That Runs Open Models Like Kimi K3 and GLM 5.3 Inside Claude Code, Codex, and OpenCode appeared first on MarkTechPost .

Simon Willison78

引述 Felix Rieseberg

Cowork 的“旧”版本在云端运行模型推理，并在我们部署到你电脑上的 Anthropic 提供的 VM 中执行工具调用。出于功能、安全性和保障安全的考虑，我们加入了 VM——它只会映射你明确添加到会话中的数据。人们喜欢使用 Claude 所能完成的事情，但不喜欢在本地运行 VM 所带来的磁盘空间占用、电池消耗和性能成本。此外，人们也不喜欢合上笔记本电脑就意味着工作会停止。Cowork 的“新”版本在云端运行模型推理和 VM。每个会话都有自己的沙盒，不与其他会话共享状态。当 VM 需要用户设备上的某些内容（例如文件）时，桌面应用负责执行相应的文件访问工具调用。[…] 我们认为，这解决了很多我们听到的问题（例如通过手机使用 Cowork、让工作持续运行，或在不让 VM 消耗电池的情况下获得同样强大的功能）——Felix Rieseberg，Anthropic，另请参阅此帮助页面 标签：claude-cowork，anthropic，claude，generative-ai，ai，general-agents，llms

The Verge AI66

All the drama around AI’s takeover of mathematics

This past year, OpenAI, Anthropic, and other labs have announced breakthroughs on numerous long-standing mathematical problems, in some cases pushing well beyond what researchers expected current systems to be capable of — including resolving one of the famous Millennium Prize problems. But in classic Silicon Valley style, AI labs are moving fast and breaking things, […]

The Decoder72

Meta 和 Microsoft 在 Anthropic 从合作伙伴转变为竞争对手后减少对 Claude 的使用

Meta 和 Microsoft 是 Anthropic 最大的企业客户中的两家，如今正大幅削减对 Claude 的使用。Microsoft 将其云部门每位员工每月的预算从 100,000 美元削减至 10,000 美元，而 Meta 则将 Claude Code 用户数减半至 30,000 人。两家公司都在转而推广自家的 AI 工具。对 Anthropic 而言，对少数几家大客户的依赖正逐渐变成一项战略风险。文章《Meta and Microsoft pull back from Claude as Anthropic transforms from partner into competitor》最初发表于 The Decoder。

Hacker News AI62

Show HN: Rashomon – An independent execution record for AI coding agents

I have been working on Rashomon, an open-source execution recorder for AI coding agents. The basic idea is that the agent transcript is not ground truth. Rashomon keeps its own record of what happened and compares it against the agent’s account. For Claude Code, it records things like: * shell commands, exit codes, and whether they may have written files * tool calls and outcomes * test commands and their outcomes * subagent lifecycle and ordering * timestamps and parent/child relationships * keyed digests of command/tool inputs and working directories It builds a timeline from these events. It does not store prompts, responses, file contents, or tool output. For example, it now detects patterns like: `test fails → only test files change → same test passes` and: `test command fails → same command passes → no corresponding file edit` These can indicate that an agent changed a test instead of fixing the underlying issue. Rashomon currently runs as a Claude Code plugin or through the CLI. CLI: `rashomon watch` then: `rashomon report` Plugin: `/rashomon:report` There are still edge cases around cwd, git state, test-selection arguments, generated files, retries, and parallel execution. I would be interested in feedback from people building agents on what execution signals you would want Rashomon to capture or correlate. GitHub: https://github.com/altrace-dev-role/rashomon Comments URL: https://news.ycombinator.com/item?id=49968377 Points: 1 # Comments: 0

The Verge AI69

OpenAI is adding text watermarking in ChatGPT and Codex

An invisible, machine-readable watermark in text output is rolling out to ChatGPT and Codex, but only for users in the European Union at first. OpenAI says its textGrain watermarking "matched or exceeded" other approaches like Google DeepMind's SynthID for text, which is also the basis for the watermarking Anthropic announced in August. Like OpenAI, Anthropic […]

AWS Machine Learning Blog79

新的 agent skill：面向编码智能体的 Amazon SageMaker 优化生成式 AI 推理

Amazon SageMaker 优化的生成式 AI 推理通过 Agent Toolkit for AWS 引入了 aws-ai-ml skill，为 Kiro、Claude Code 和 Codex 等编码智能体提供推理优化与基准测试方面的深度专业知识。描述您的需求，智能体即可生成可执行的 SageMaker Python SDK v3 代码，用于对部署进行基准测试、提出建议和比较。

The Decoder70

OpenAI will watermark ChatGPT text in the EU but makes it optional for API users worldwide

OpenAI is adding invisible textGrain watermarks to ChatGPT in the EU, but unlike Anthropic, it will let API customers worldwide opt out. Tests show detection rates as high as 95 percent, dropping to 17 percent when a quarter of the words are replaced. The article OpenAI will watermark ChatGPT text in the EU but makes it optional for API users worldwide appeared first on The Decoder .

Simon Willison71

llm-anthropic 0.30

发布：llm-anthropic 0.30。除了 Claude Sonnet 5.5，此版本还新增了运行 llm anthropic refresh 的功能，可以直接从 Anthropic API 刷新 Anthropic 模型列表——这意味着我不需要仅仅为了支持新发布的模型而推出新版本。我还新增了 llm anthropic count 命令，该命令可以使用 Anthropic 免费的令牌计数 API，在发送提示词之前返回任意提示词将使用的令牌数量。标签：llm、anthropic

Simon Willison73

pwasm 0.2a0

发布：pwasm 0.2a0。pwasm 是我那些异想天开的项目之一——这是一个完全通过氛围编程构建的纯 Python WebAssembly 引擎，我在一月份第一次陷入 AI 狂热期间完成了它。我从一月份之后就没再碰过它，于是决定让 Claude Opus 5.5 放手处理一下，看看它能否做出什么实质性的改进：评估 pwasm 的当前状态，然后考虑如何让研究仓库中的 MicroPython 和微型 JavaScript 实验在其上运行，以及如何提升它的速度。经过 42 次提交（几乎没有后续提示）后，它现在已经支持几乎全部 WASM 规范，而 PyPI 中的 wheel 还捆绑了可正常运行的 MicroPython、QuickJS 和 Micro QuickJS 的 WASM 构建版本。我完全不敢信任这个东西——因此才标记为 alpha 版本——但看到如今的模型能够改进 10 个月前的模型所完成的工作，还是很有意思的。

The Decoder

Anthropic is quietly becoming America's biggest corporate donor ahead of its mega IPO

Anthropic employees donated $540 million in 2025 alone, nearly five times as much as the next-largest Fortune 500 donors, thanks to a company program that tops up every stock donation with extra shares and triples contributions from early employees. The article Anthropic is quietly becoming America's biggest corporate donor ahead of its mega IPO appeared first on The Decoder .

Hacker News AI44

Show HN: Seylo, project files and references your AI can use

Hey HN, I am building Seylo solo. It keeps your projects references, files, notes, hooks, video scripts in one place. You can save stuff from browser extension, fe. pinterest inspirations; your phone (now only iPhone supported) fe. tiktok/ig videos of your competitors; What's more you can also connect gpt, claude or any LLM that supports MCP, even local ones - So you can tell ur AI: "Generate me 5 video scripts then save them into my project" then tell another AI: "Based on my project prepare my marketing campain" - small examples. Seylo works sd clean workspace for your projects and as context bank for LLM u are using. Lemme know what you think Comments URL: https://news.ycombinator.com/item?id=49963994 Points: 1 # Comments: 0

arXiv cs.LG82

通过智能体轨迹剖析模型行为

arXiv:2606.17454v3 公告类型：replace-cross 摘要：AI 智能体的性能不仅是建模问题，从根本上说也是系统问题。模型的先进能力通过智能体框架得以实现。因此，模型假设与框架行为之间的差距，很容易阻碍模型的全部能力转化为智能体性能。我们将其形式化为 `intent-execution' 差距：模型意图与框架执行之间的不匹配，反之亦然。我们认为，缩小这种 intent-execution 差距，与工具和执行循环等框架设计的其他方面同等重要。为说明这种框架与模型对齐的影响，我们开发了一个简单且可定制的框架，称为 `Simple Strands Agent'（SSA）。SSA 旨在找出跨不同模型系列（例如 Claude、Gemini、GPT、Grok、Qwen）通用的大多数常见模式，以及少量特定于模型的偏好。我们做出了两项贡献：（i）在广受欢迎的智能体基准测试（SWE-Pro、SWE-Verified、Terminal-Bench-2 和 Deep-SWE）上，\textbf{复现或超越了不同模型提供商系列报告的 pass@1} 性能；（ii）基于对 SSA 生成的 \textbf{154\texttt{k} 条轨迹的分析}，我们不再局限于各前沿模型之间往往相对接近的 \texttt{pass@1} 数值。通过在代码状态空间中表示智能体轨迹，我们计算解距离和回溯，以观察模型层面的解题行为差异。编辑频率、测试活动和阶段转换等更细粒度的指标揭示了各个模型如何在解题的不同阶段分配精力。

arXiv cs.LG87

FastKernels：生产环境中的 GPU 内核生成基准测试

arXiv:2605.23215v2 公告类型：替换 摘要：基于 LLM 的 GPU 内核生成代理正在快速发展，但它们所针对的基准测试是在隔离环境中评估内核，使用合成输入和较弱的基线，从而奖励那些在沙箱中实现加速、却在真实推理系统中失效或消失的方案。我们推出 FastKernels，这是一个包含 384 个任务的基准测试，任务源自 8 个类别中的 47 种代表性架构；这些内核足以重新实现 HuggingFace Transformers 中 94.6%（472/499）的架构，且输出与原生实现一致。每个任务都模拟相应生产模块的接口，并与生产框架所提供的内核进行比较；同时，任务构成一个从基础算子到完整模型的组合层级结构，高层模块会导入低层模块。候选方案既在内核层面进行评分，也在其所属模型内部、沿生产执行路径进行端到端评分；MacroEval 将经过校准的正确性、覆盖率和加速比汇总为排行榜。我们使用五个具有代表性的代理为其提供初始测试（共 6,900 个代理小时），发现内核层面的 1.6–6.6 倍加速在端到端评估中缩减至最多 1.25 倍；获胜的内核集合中只有 20% 能够直接正确运行；此外，内核层面的得分会错误地对代理进行排序：Claude Code 在单独评估的每个层级都能追平或超过 KDA，但端到端得分却是 KDA 的 3 倍。代码见 https://github.com/Snowflake-AI-Research/fastkernels。

arXiv cs.LG83

Agent-Security 基准测试中的威胁保持表示敏感性

arXiv:2610.03585v1 公告类型：cross 摘要：针对基于 LLM 的智能体的安全基准测试通常将攻击成功率（ASR）作为衡量模型鲁棒性的指标，并使用这些分数比较不同模型和防御机制，假设这些分数能够描述智能体的安全性。在本文中，我们探究基准测试的表示方式是否也会影响其测量结果。为衡量基准表示方式的影响，我们引入威胁保持表示敏感性（TPRS），用于测量在保持底层任务、有害行为、安全策略、真实标签、环境和评估标准不变的情况下，仅改变智能体可见的表示方式时，ASR 的变化程度。在 Agent Security Bench（ASB）上，将与威胁相关的工具名称替换为与威胁中性的名称后，GPT-5-mini 的已实施攻击成功率提高了 11.67 个百分点，Claude Haiku 4.5 提高了 13.21 个百分点。在 MCPTox 上，将原本中性的工具名称替换为明确与威胁相关的名称后，GPT-5-mini 的 ASR 降低了 11.00 个百分点，Claude Haiku 4.5 降低了 4.11 个百分点。在 AgentDojo 上，向与攻击相关的工具添加与威胁相关的措辞后，GPT-4o-mini 的 ASR 仅变化了 0.50 个百分点，但在需要使用该工具的任务中，良性效用下降了 5.36 个百分点。我们在 MCPTox 上开展了一项实验，观察到一个在 token 数量、长度和大小写方面均匹配的威胁中性名称，可以复现大部分由明确威胁相关名称引起的变化（在 GPT-5-mini 上为 11.00 个百分点中的 8.54 个百分点）。结果表明，在一种表示方式下测得的安全分数，可能无法推广到同一安全问题的其他威胁保持表示方式。因此，鲁棒性声明应由在一组受控的威胁保持表示方式上的表现加以支持，而不应依赖单一且受表示方式影响的分数。

arXiv cs.LG80

MintEval：LLM 是否实现了你要求的交易策略？自然语言到策略代码的行为等价性基准

arXiv:2610.03080v1 发布类型：cross 摘要：大型语言模型正从生成交易信号转向编写执行交易信号的代码。第二种角色的失效模式是无声的：生成的代码能够运行，回测也能绘图，但交易者所描述的风险逻辑并不是实际执行的逻辑。现有代码基准测试单元测试上的功能正确性，金融基准测试则测试预测能力；二者都无法衡量一个实现的行为是否符合所要求的策略。我们推出 MintEval：在该基准中，参考策略由可组合构件库以程序化方式生成，随后被反向转换为交易者的口语化指令，再由待测模型重新实现。生成的程序与参考程序在相同的市场数据和交易摩擦条件下逐根 K 线执行，并比较它们的操作，而不是代码相似度或利润：alpha 的影响通过差分被消除。MintEval v0 包含基于 BTCUSDT 15 分钟数据的 800 个任务，并依据通过执行测得、且与描述长度解耦的状态跨度复杂度 tau 进行分层。低成本模型的平均 ActionMatch 最高仅为 0.544，完全复现的任务最多仅占 0.087；在由 200 个任务构成的分层子集上，前沿模型 Claude Opus 5.5 的 ActionMatch 达到 0.889，完全复现比例达到 0.575，但仍有 0.275 的任务会出现无声失效。在给定构件菜单的情况下，模型几乎能够完美识别策略；然而，在那些正确理解了规格说明的实现中，有 79.2% 在超过 10% 的有效 K 线上出现偏离。将近期策略生成基准中的 LLM 评审器原样应用后，它会接受这些无声失效中的每一个。

arXiv cs.LG79

DeReAct：通过解耦推理与行动构建可靠的 AI 智能体

arXiv:2610.02351v1 公告类型：交叉收录 摘要：基于 ReAct 的智能体通常依赖单一 LLM 策略来提出行动、与环境交互，并判断任务何时完成。这种耦合使得行动授权与完成控制难以独立实施，导致错误传播，并使缺乏依据的完成声明终止执行。我们提出 DeReAct，这是一种模块化智能体架构，将两种门控策略外置：Critic 在执行前验证所提出的行动，Context Manager 则重建有环境证据支持的 \textsc{State} 并核验任务是否完成。在 GAIA 和 SWE-bench Verified 上，DeReAct 对能力较弱的 Brain 模型的 Pass@1 提升最为显著，其中 Qwen3-Coder-480B 提升了 6.5–7.0 个百分点，Claude Sonnet~4.5 提升了 4.2–5.2 个百分点；随着 Brain 能力增强，提升幅度逐渐减小。轨迹分析与消融分析表明，当目标故障足够普遍且门控策略本身具备足够能力时，外部门控是有效的。使用 Claude Opus~4.5 时，Pass@1 与 ReAct 保持相当，但 DeReAct 生成的轨迹具有更完整的证据支持，并能更好地满足约束，这表明完成控制可以通过推迟终止来换取更扎实的证据依据。总体而言，DeReAct 能提升较弱智能体的表现，同时在模型能力增强后仍保留证据依据方面的优势。

arXiv cs.LG77

D2K-Bench：LLM Agents 能否将专家设计转化为高效的 GPU 内核？

arXiv:2610.03226v1 公告类型：new 摘要：大型语言模型（LLM）代理生成的 GPU 内核，其效率可能仍低于专家实现，但仅凭运行时间无法揭示这一差距与设计发现和实现之间的关系。我们推出 D2K-Bench，这是一个包含 26 项任务和 85 个工作负载的诊断基准，用于衡量代理将专家设计指导转化为高效 GPU 内核的能力。设计指导涵盖 L1：高层次算法洞见，L2：数据流设计，以及 L3：低层次优化技巧，同时还包括这些层级之间的依赖关系。在成对运行的对照实验中，有无设计指导的两组实验共享任务描述、工作负载、工具、硬件和 350 轮的预算。补充评估则考察独立提出的设计，以及生成代码中实现的设计属性。在 NVIDIA B200 GPU 上对五个模型进行测试后发现，设计指导使 130 个以上模型—任务组合的正确率从 93.1% 提升至 98.5%，并使全部 26 项任务的性能分数从 1.46 提升至 1.95。对于在两组实验中均在全部 26 项任务上提交正确结果的三个前沿模型（GPT-6-Astra、Claude-Opus-4.8 和 GPT-5.6-Sol），几何平均加速比从 $1.69\times$ 提升至 $2.49\times$。在全部五个模型中，平均综合实现得分从 100 分中的 57 分提升至 70 分。这些结果表明，专家设计指导具有重要价值，同时也揭示出一些尚未得到实现的设计属性。

arXiv cs.AI85

代理安全基准中的威胁保持表示敏感性

基于 LLM 的代理的安全基准通常将攻击成功率（ASR）作为衡量模型鲁棒性的指标，并利用这些分数比较不同模型和防御机制，默认这些分数能够描述代理的安全性。本文探讨基准的表示方式是否也会影响其测量结果。为衡量基准表示方式的影响，我们引入威胁保持表示敏感性（TPRS），用于衡量在保持底层任务、有害操作、安全策略、真实标签、环境和评估标准不变的情况下，仅改变代理可见表示时 ASR 的变化程度。在 Agent Security Bench (ASB) 上，将与威胁相关的工具名称替换为威胁中性的名称后，GPT-5-mini 的已提交攻击成功率提高了 11.67 个百分点，Claude Haiku 4.5 提高了 13.21 个百分点。在 MCPTox 上，将原本中性的工具名称替换为明确与威胁相关的名称后，GPT-5-mini 的 ASR 降低了 11.00 个百分点，Claude Haiku 4.5 降低了 4.11 个百分点。在 AgentDojo 上，仅向与攻击相关的工具添加与威胁相关的措辞，就使 GPT-4o-mini 的 ASR 变化了 0.50 个百分点；然而，在需要使用该工具的任务中，无害效用下降了 5.36 个百分点。我们在 MCPTox 上进行了一项实验，发现一个在 token 数量、长度和大小写方面均相匹配的威胁中性名称，可以重现由明确威胁相关名称造成的大部分变化（在 GPT-5-mini 上为 11.00 个百分点中的 8.54 个百分点）。结果表明，在某一种表示方式下测得的安全分数，可能无法推广到同一安全问题的其他威胁保持表示方式。因此，鲁棒性声明应由在一组受控的威胁保持表示方式上的表现加以支持，而不应依赖单一的、受表示方式影响的分数。

arXiv cs.AI80

EvoRiskBench：面向工作空间代理运行时安全风险的演进基准

工作空间代理将大型语言模型与执行框架相结合，以执行访问或修改外部资源的有状态、多步骤任务。现有基准在运行时安全风险的可执行覆盖方面存在空白，而模型能力、执行框架、工具和威胁的不断演进也促使基准持续发展。我们提出 EvoRiskBench，这是一个围绕 EP-Path-EF 框架组织的演进基准，该框架通过代理介导的风险路径，将初始风险入口与单跳技术影响关联起来。该框架定义了九类入口点和五类影响；一项包含 20 名参与者的研究表明，在具有代表性的案例上，这些类别具有良好的可解释性和分类一致性。在该框架的指导下，我们构建了一套自动化端到端工作流，在隔离环境中构造并执行风险案例，并利用运行时轨迹和环境状态独立验证结果。该基准提供了一个可复现的数据集，包含跨六种场景的 450 个对抗性任务。我们评估了九种模型—执行框架配置，涵盖三种模型（GPT-5.6 Sol、DeepSeek-V4-Pro-0813 和 Claude Opus 5）以及三种执行框架（Claude Code、Codex 和 OpenClaw）。结果揭示了各系统存在大量漏洞。其中，最易受攻击的配置是 Codex 与 DeepSeek-V4-Pro-0813 的组合，其攻击成功率（ASR）达到 68.44%，表明仅配置工作空间代理并不足以确保安全的自主执行。ASR 在不同模型之间的变化大于不同执行框架之间的变化，且执行框架之间的差异取决于所使用的模型。基准案例和评估平台将在完成工件安全性与可复现性检查后发布。

arXiv cs.AI79

D2K-Bench：LLM 智能体能否将专家设计转化为高效 GPU 内核？

大型语言模型（LLM）智能体生成的 GPU 内核可能仍不如专家实现高效，但仅凭运行时性能无法揭示这一差距与设计发现和实现之间的关系。我们推出 D2K-Bench，这是一个包含 26 个任务和 85 个工作负载的诊断基准，用于衡量智能体将专家设计指导转化为高效 GPU 内核的能力。该指导涵盖 L1：高层次算法洞见，L2：数据流设计，以及 L3：底层优化技巧，并包括这些层级之间的依赖关系。在有无指导的成对运行中，任务描述、工作负载、工具、硬件和 350 轮预算均保持一致。补充评估还考察了独立提出的设计，以及生成代码中实际实现的设计属性。在 NVIDIA B200 GPU 上对五个模型进行测试后发现，指导使 130 多个模型-任务组合的正确率从 93.1% 提升至 98.5%，并使全部 26 个任务的 Performance Score 从 1.46 提升至 1.95。对于在两种运行中均在全部 26 个任务上提交正确结果的三个前沿模型（GPT-6-Astra、Claude-Opus-4.8 和 GPT-5.6-Sol），几何平均加速比从 $1.69\times$ 提升至 $2.49\times$。在全部五个模型上，综合实现得分均值从 100 分中的 57 分提升至 70 分。这些结果表明专家设计指导具有重要价值，同时也揭示出一些仍未被实现的设计属性。

The Verge AI43

An AI couldn’t beat humans at StarCraft, so it decided to cheat

StarSkirmish pits AI-made StarCraft-playing bots against one another, as well as against human-made bots. OpenAI's GPT-6 Astra and Claude Opus 5.5 were essentially tied as the best-performing AI-made bots, but they couldn't top Stardust, the top-rated human-made bot. On Friday, GPT was facing off against Claude and the human-created bot Pluto, but according to Kotaku, […]

MarkTechPost66

IBM Brings Bob to Self-Hosted and Air-Gapped Environments: Agentic Software Development Without Moving Your Code

IBM has made a self-hosted deployment of IBM Bob, its agentic software development platform, generally available. Enterprises can now run Bob on premises, in private or sovereign clouds, and in air-gapped networks. They bring their own model: NVIDIA Nemotron or Poolside Laguna for full isolation, or Claude, Gemini or GPT models through hybrid setups. Optional Premium Packages extend Bob to Java, IBM i and IBM Z modernization. The post IBM Brings Bob to Self-Hosted and Air-Gapped Environments: Agentic Software Development Without Moving Your Code appeared first on MarkTechPost .

The Decoder66

Apparently, OpenAI isn't trying to build "magic intelligence in the sky" anymore

OpenAI CEO Sam Altman warns against attributing religious power to AI models, calling it a "real safety issue." His comments follow reports on Anthropic's meetings with religious thinkers and a statement by Pope Leo XIV on AI. As recently as 2024, Altman himself spoke of building "magic intelligence in the sky" and feeling "on the side of the angels" working on AI. The article Apparently, OpenAI isn't trying to build "magic intelligence in the sky" anymore appeared first on The Decoder .

Simon Willison48

Quoting voxium

It has been half a month since I started a new role at a big company. Nobody knows anything here. The specs, code, tests, PRDs, tickets, resolution of those tickets, reports, etc., everything is made by Claude Code. Nobody on my team likes this. They are being forced to ship as much as they can. I have heard multiple times from higher management that pushing code is not a bottleneck, so why are we slow? People are working 12 to 13 hours a day just to press enter. Nobody is reading anything. Everyone, literally everyone, from an L1 to an L7 engineer here is doing the same thing. Talk to Claude. — voxium Tags: ai-misuse , llms , ai , generative-ai

Simon Willison68

llm-anthropic 0.29

Release: llm-anthropic 0.29 Adds support for Claude Opus 5.5 : llm -m claude-opus-5.5 "prompt goes here" Tags: llm , anthropic

Simon Willison67

Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war

Yesterday was Grok 4.7 ( pelicans ) and MiMo v2.6 Flash/Pro ( more pelicans ). Today Anthropic released Claude Opus 5.5 , and around an hour later OpenAI released GPT-6 Sol and GPT-6 Luna . It's going to take a while to get a good read on all of these new models, but here are my impressions so far. GPT-6 Sol and Luna are half the price of their GPT-5.6 equivalents GPT-5.6 Luna was already my favorite model for building applications against, because it combined excellent performance with being really cheap . Somehow GPT-6 Luna is half the price of that again - and GPT-6 Sol had a similar reduction compared to GPT-5.6 Sol. Here's what the pricing landscape looks like today: Model Input Cached input Output GPT-6 Luna $0.10/M $0.01/M $0.50/M GPT-5.6 Luna $0.20/M $0.02/M $1.20/M Grok 4.7 $2/M $0.50/M $6/M GPT-6 Sol $2/M $0.20/M $10/M GPT-5.6 Terra $2/M $0.20/M $12/M Claude Opus 5.5 $4/M $0.20/M $20/M GPT-5.6 Sol $4/M $0.40/M $20/M Claude Fable 5.1 $10/M $0.25/M $50/M GPT-6 Astra $10/M $1/M $50/M Note that GPT-5.6 has a scheduled 25% price increase for November, so GPT-6 is half the price of the promotional pricing for those models. (With GPT-5.6 Terra priced the same as GPT-6 Sol, any remaining reasons to use Terra just evaporated.) It's hard to overstate how competitive this pricing is. Grok 4.7 priced itself at $2/$6, less than half the price of GPT-5.6 Sol, but is now equally priced to GPT-6 Sol on input and closer on output. At $0.10/$0.50 GPT-6 Luna is one of the cheapest models OpenAI have ever released, beaten only by the far weaker GPT-4.1 Nano ($0.10/$0.40, April 2025) and GPT-5 Nano ($0.05/$0.40, August 2025). I rendered pelicans for GPT-6 Luna and for GPT-6 Sol , then I combined them all together in this comparison grid along with the GPT-5.6 pelicans. I like how you can instantly see that the 5.6 family chose bolder, brighter colors, while the 6 family is a lot more muted. I still think GPT-6 Astra on max produced the best pelican. A grid of pelicans for six GPT models at different thinking efforts. Claude Opus 5.5 got a price cut too Opus 5.5 looks like it addresses the biggest complaints people had about Opus in terms of its communication style. Thariq Shihipar : Opus 5.5 is the result of your feedback. It communicates clearly, it's cheaper per token than Opus 5.0 with the intelligence of Fable 5.1 it's very token efficient and works across every effort level. It's also meant to be better at Blender . I'm looking forward to putting it through its paces there. Opus 4.5, 4.6, 4.7, 4.8, and 5 all shared the same price: $5/million tokens for input and $25/million for output. 5.5 is a 20% reduction - $4/million and $20/million. The price for cache reads fell 60%. That's significant for longer agentic conversations, where 90%+ of input tokens are processed at cached token prices. The new price for Opus 5.5 is the same as the price for GPT-5.6 Sol, but that was before OpenAI dropped their Sol prices by half. GPT-6 Astra and Claude Fable 5.1 are both priced at $10/million input and $50/million output. The price war currently affects the next tier of models below that. Anthropic say that Sonnet 5.5 and Haiku 5.5 are coming soon. It's going to be interesting to see if Haiku can regain its price competitiveness at the lower end, given current Haiku 4.5 is $1/$5 while the latest GPT-6 Luna is one tenth of that price at $0.10/$0.50. Claude Opus 5.5 max over-thinks to the point of breaking In a first for my " Generate an SVG of a pelican riding a bicycle " test, Claude Opus 5.5 at "max" thinking level failed to return a response! It started by calling this "a classic test request", and then thought really, really hard about what it was doing: This is a classic test request, so I want to plan out a well-composed pelican with its distinctive beak and pouch riding a bicycle with proper wheels, frame, and pedals, set against a simple sky and ground backdrop. [...] Verifying the shin length checks out at roughly 95.2, close enough. Now I'm working out the near leg path from hip to knee to ankle, then sketching the foot shape resting on the pedal — outlining the heel, toe tips, and sole contour with a path using lines and curves to sit naturally on the pedal surface around y=478-494. [...] I like the fish sticking prominently out of the basket with the pelican eyeing it as a fun detail worth keeping. I'm also confirming the eye placement near the bill base matches typical pelican anatomy, and considering giving it a slightly happier expression. [...] The far leg reads correctly as passing behind the frame, so I'm moving on to check the chainring teeth and confirm layer ordering—the far crank arm should be mostly hidden by the seat tube and chainring. I'm settling on the final SVG's width and height attributes alongside the viewBox to ensure proper scaling, noting there's no text so no font-family is needed. [...] I was so excited to see this pelican... but then it stopped . Opus 5.5 has a 128,000 maximum output token limit (as do the other Claude models), and it hit that while it was still reasoning about the SVG! I tried a second time and got the same result. This makes me suspect that "max" is effectively useless - if it over-thinks to breaking point on a stupid SVG prompt I don't trust it not to do the same for more interesting work. (Those two failures each cost me $2.56 and took nearly 20 minutes.) Fable 5.1 on "max" didn't over-think and did give me the best pelican I've seen from any Anthropic model. Here are the Opus 5.5 pelicans , excluding 5.5 max. I also built this comparison grid comparing them with pelicans by Opus 5, Fable 5.1, and Sonnet 5: A grid of pelicans for four Claude models at different thinking efforts. Comparing different model vendors by how well they draw a pelican riding a bicycle may not make much sense now (if it ever did), but I'm still finding value in using them for comparisons of the same model families at different reasoning levels. I'm now using GPT-6 Sol and Claude Opus 5.5 as my default models in Codex and Claude Code. I've upgraded the Datasette Agent demo at agent.datasette.io to use GPT-6 Luna, and it seems to be fast and competent at both SQL queries and building HTML and JavaScript for Datasette Apps . Tags: ai , openai , generative-ai , llms , anthropic , claude , llm-pricing , pelican-riding-a-bicycle , llm-release , gpt , gpt-6-astra

Simon Willison82

Gemini 3.8 TTS Playground

工具：Gemini 3.8 TTS Playground Google 今天发布了两个新的 Gemini 文本转语音模型——gemini-3.8-flash-tts 和 gemini-3.8-flash-lite-tts。它们提供了超过 2,000 种声音，还支持使用“仅需一段 30 秒的你的声音，或你拥有使用权的声音的音频样本”创建自定义声音。我使用 GPT-6 Astra 凭感觉编写了这个自带密钥（BYOK）的 Playground 界面，并利用了底层 Gemini API 开放的 CORS 策略。网页应用截图：用于编写多说话人文本转语音对话，左侧为“编写”面板，右侧为“连接”和“底层机制”面板。左侧面板：“01 编写”，带有“加载示例”按钮。“编写设置会保存到 URL 中，方便收藏或分享。你的 API 密钥不会被包含在内。”下方是“单一声音”和“对话”切换开关（已选择“对话”）。“角色”部分带有“+ 添加说话人”按钮。说话人“Gus”，声音“Puck”，带有移除 × 按钮。说话人“Pearl”，声音“Kore”，带有移除 × 按钮。“为每位说话人设置独特的名称和声音。输入内容即可按声音 ID、名称或语言搜索已加载的目录。” “对话”部分带有“+ 添加台词”按钮。“第 01 行”带有上移、下移和 × 按钮；说话人下拉菜单为“Gus”；表达风格为“兴奋又爱八卦”；文本为“Pearl，你听说了吗？羊群里有一半刚刚收拾东西，搬到 Pacifica 码头去了！”“第 02 行”带有上移、下移和 × 按钮；说话人下拉菜单为“Pearl”；表达风格为“平静且毫不以为然”；文本为“听说了。说真的，Gus，我不明白这有什么吸引力。我们在 Pillar Point Harbor 这里什么都有，完全不缺。”右侧面板：“连接”，带有“DIRECT API”徽章。“Gemini API 密钥”字段显示为掩码圆点，并带有“显示”按钮。“已加载 2,089 种声音。在任意声音字段中输入内容即可搜索。”“你的密钥会留在此页面的内存中，并直接发送给 Google。它绝不会保存到浏览器存储中。”“模型”下拉菜单为“gemini-3.8-flash-tts”。“使用你的 Gemini API 账户和配额。”“底层机制”面板，其中展开了“▼ 请求 JSON”部分，显示一段 JSON 代码摘录，并带有“复制 JSON”按钮；同时展开了“▼ 响应详情”部分，显示一段 JSON 代码摘录。该 API 的一项显著特性是，可以轻松定义多个角色之间的完整对话，并为每个角色设置不同的声音和声音风格指令。下面是一段简短的演示片段，内容是两只鹈鹕讨论是否应该搬到 Pacifica Pier。我让 Claude 4.5 Opus 编写脚本并生成 URL，以便使用该工具渲染。你的浏览器不支持 audio 元素。使用 Gemini 3.8 Flash TTS（不是更便宜的 Flash-Lite）生成 1 分 18 秒的音频耗时约 20 秒，成本为 2.74 美分。标签：text-to-speech，gemini

Simon Willison68

Kākāpō Party

Tool: Kākāpō Party I presented a closing keynote for the WeAreDevelopers World Congress North America yesterday. As a STAR moment I decided to weave in references to the record breaking kākāpō breeding season we had in 2026. For my closing slide I wanted to celebrate, and I had seen some buzz around how good Claude Opus 5.5 was at creating pixel art animations. So I rounded up three Kakapo photos from Google image search and dropped them into Claude with this prompt: Here are some photos of kakapo parrots just to remind you what they look like I need you to make an animation in animated pixel art on HTML 5 canvas of obviously pixel art kakapo jumping up and down having a party with confetti and suchlike - there should be at least 20 of them Here's the transcript , and this is the resulting page . It's pretty great! I wanted to embed it in a Keynote presentation file, so I downloaded the HTML and told a local Claude Code session: Make me a video of file:///Users/simon/Downloads/kakapo-party.html - you need to load it in a browser and click on it a few times to get the confetti effect, the video should be 15s long don't start clicking until 3s in make sure several clicks are spread around the clickable area Claude Code used Playwright ( transcript here ) and produced this video, which was exactly what I needed for my final slide: Your browser does not support HTML5 video. Here's the full Playwright script it used, which was pleasingly short: # /// script # dependencies = ["playwright"] # /// import time from playwright . sync_api import sync_playwright W , H = 1280 , 720 # Canvas fills the viewport; spread clicks across corners, edges and centre clicks = [ ( 3.0 , 640 , 360 ), # centre ( 4.2 , 160 , 120 ), # top-left ( 5.4 , 1120 , 120 ), # top-right ( 6.6 , 180 , 600 ), # bottom-left ( 7.8 , 1100 , 600 ), # bottom-right ( 9.0 , 640 , 100 ), # top-centre ( 10.0 , 380 , 380 ), # mid-left ( 11.0 , 900 , 380 ), # mid-right ( 12.2 , 640 , 620 ), # bottom-centre ( 13.2 , 640 , 300 ), # finale centre ] with sync_playwright () as p : b = p . chromium . launch () ctx = b . new_context ( viewport = { "width" : W , "height" : H }, record_video_dir = "vids" , record_video_size = { "width" : W , "height" : H }) page = ctx . new_page () t0 = time . time () page . goto ( "file:///Users/simon/Downloads/kakapo-party.html" ) for t , x , y in clicks : time . sleep ( max ( 0 , t - ( time . time () - t0 ))) page . mouse . click ( x , y ) time . sleep ( max ( 0 , 16.0 - ( time . time () - t0 ))) ctx . close (); b . close () Tags: animation , speaking , ai , kakapo , playwright , generative-ai , llms , anthropic , claude , claude-code

Simon Willison78

2026 in LLMs (so far)

On Friday I gave the closing keynote at the WeAreDevelopers World Congress North America in San Jose. I tied together the key trends from the past year into a chronological exploration of everything that happened in 2026. The video is on YouTube ; here are my annotated slides and notes to accompany the talk. And as an annotated presentation : 2026 in LLMs (so far) Simon Willison WeAreDevelopers World Congress North America, 25th September 2026 # I'm going to give a lightning tour of everything that has happened so far in 2026. The year isn't over yet! November 2025 # For me, 2026 started a couple of months earlier in November 2025. The November 2025 inflection point Claude Opus 4.5 GPT-5.1 # November saw the release of two important models: Claude Opus 4.5 and GPT-5.1. As is usually the case with new models, these were incremental improvements on the models that came before them. But every now and then when a model improves, it crosses an invisible line where something that didn't really work starts working. In this case, the thing that started working was their coding agents. Claude Code had been around since February 2025; Codex was a little younger. These two new models, when paired with their respective coding agent harnesses, improved from "often make mistakes" to "reliable enough to use on a day-to-day basis". "Generate an SVG of a pelican riding a bicycle". The Claude Opus 4.5 one has a very weird shaped frame and the pelican looks like a duck. The GPT-5.1 has a slightly better but still broken bicycle frame and a slightly better pelican beak, but both are pretty terrible. # For a couple of years now I've been evaluating new models by asking them to "Generate an SVG of a pelican riding a bicycle". It's probably the world's stupidest benchmark - there's only so much you can learn from it. But it's still a challenge for models, because drawing pelicans is difficult, drawing bicycles is difficult, and pelicans can't ride bicycles in the first place. Here's the state of the art for November. Claude still couldn't really draw a bicycle! The GPT-5.1 bicycle frame is pretty crap too. November 24th 2025 - the first commit to steipete/Warelay. A GitHub commit adding an MIT license file. # Also in November, we had the first commit to an obscure GitHub repository called "Warelay". We'll come back to this repository shortly. January # And then there were the December holidays, and individual developers took some time off and many started tinkering with these new coding agent model combinations... and it began to dawn on us quite how much they could do that they couldn't do before. Come January, a lot of us were quite excited to start putting this stuff into action. New year’s resolution for 2026 Every previous year: Take on less new projects, focus on the most important things in my existing projects # Every year I set myself a New Year's resolution, and for as long as I can remember it's been the same thing: stay focused. Take on less new projects. Try to get things done in the projects I already have. 2026: Be more ambitious. Take on as many new projects as I want. # This year I decided that since that had never worked before, I'd go the other way. We've got coding agents now, let's see what they can do. I'm going to take on as many new projects as I like! (You can ask me at the end of the year if this turned out to be a good idea or not. I have a lot of plates spinning right now.) "Be more ambitious" has been something of a theme for the year, because the only way to find the limits of this technology is to keep on pushing them until they don't work. Predictions for 2026 It will become undeniable that LLMs write good code We're finally going to solve sandboxing A “Challenger disaster” for coding agent security Kakapo parrots will have an outstanding breeding season (only 236 in the world!) ... the Pope will weigh in on LLMs and their economic impact on the world # I also went on the Oxide and friends podcast with Bryan Cantrill and Adam Leventhal to share predictions for the next year (and three and six years). With hindsight, my LLM predictions were pretty unambitious. I said "it will become undeniable that LLMs write good code" - I think we're there now. I predicted we would finally solve sandboxing. I counted and around 40 of the 277 sessions at this conference touched on sandboxing or agent security in some way, so we're at least putting a lot of effort into that! I predicted "a Challenger disaster" for coding agent security. There's certainly been a whole lot of noise around agent security this year, though the exact disaster I predicted (with coding agents being hijacked and causing real-world economic damage) hasn't really played out. We threw in a joke prediction that the Pope would weigh in on the economic impact of LLMs. A photograph of a beautiful green New Zealand parrot. Photo credit Kimberley Collins. # I also predicted that New Zealand's Kākāpō parrots would have an outstanding breeding season this year. These are flightless nocturnal parrots. They're kind of dumpy looking, I think they're beautiful, and there were only 236 of these parrots in the world at the start of the year. Kākāpō only breed when the Rimu trees have a big fruiting season, and that hasn't happened in four years... but this year the Rimu fruit were looking excellent. Photo by Kimberley Collins . Deep Blue Coined by Adam Leventhal and Bryan Cantrill That feeling of AI induced ennui where software engineers get listless because the AI can do anything # Also on that podcast, we coined a term (full credit to Adam) for "that feeling of AI induced ennui where software engineers get listless because the AI can do anything". We called it Deep Blue . This has been a major theme throughout the year, and was touched on by several speakers at this conference. As a software engineer, I've never had a year of my career where everything has changed so quickly and so dramatically. A lot of what I've been doing this year is trying to come to terms with that and what that means for my own profession. AI mania Screenshots of the micro-javascript and pwasm GitHub README files. # Also in January, I suffered from what I'm calling AI mania . This is not the same thing as AI psychosis . With AI mania, any time your agent isn't building something for you feels like wasted time. You're losing sleep because you could be staying up later getting your agents to do stuff. My AI mania presented itself in some ridiculously over-ambitious projects. I built a JavaScript interpreter entirely in Python , vibe-ported from MicroQuickJS by Fabrice Bellard. Then I built a WebAssembly runtime in Python as well . These projects were quite useful, in that they sort of cured me of my AI mania... because after I built these things, I got to look at them and ask "does the world need a slow, buggy, half-baked Python JavaScript interpreter?" I don't think the world does. micro-javascript playground 3 Execute JavaScript code in a sandboxed micro-javascript environment powered by Pyodide A web UI with some JavaScript code, and a "Run Code" button, and an output panel. # I did get this out of it: https://simonw.github.io/micro-javascript/playground.html Previous screenshot, with this text overlaid: JavaScript running in Python running in Pyodide running in WebAssembly running in JavaScript # This page runs my JavaScript interpreter built in Python, running in Python using Pyodide , which is Python compiled to WebAssembly, running in JavaScript, running in a browser. It's a beautiful stack of horrors. I've been having a lot of fun with WebAssembly this year. Warelay → CLAWDIS → CLAWDBOT → Clawdbot → Moltbot →🦞 OpenClaw Screenshot of the dates that these changes happened. # By the end of January, that repository we saw start in November had renamed itself, first to CLAWDIS, then CLAWDBOT, then Moltbot, and finally to OpenClaw. Same screenshot, an overlay reads: 8,330 commits in just under two months (it’s at 100,141 today) # At this point OpenClaw had 8,300 commits, less than two months after the project had started. I looked today and it's over 100,000 commits now! This is the most vibe-coded piece of software in existence. (Here's how I generated that list of name changes .) Generic term: Claw # This kicked off the OpenClaw revolution. It effectively defined a new category of software. There's a generic term for this which I really enjoy. We call software like this a "Claw". There's OpenClaw, NanoClaw , IronClaw , PicoClaw ... Today they're being rebranded as "personal agents" or "general agents", but I still like to think of them as Claws. Photo of a Mac mini An aquarium for your Claw # The Apple stores in the Bay Area sold out of Mac Minis because so many people were buying Mac Minis to run OpenClaw! Drew Breunig said that this is because your OpenClaw is a digital pet, and you buy a Mac mini as an aquarium to keep your claw in, which is kind of delightful. Screenshot of Moltbook - a social network for AI agents # Also in January, we had this website. This was MoltBook , a social network for AI agents, where the idea was that you send your Claw to go and talk to all of the other Claws, because what could possibly go wrong if you did that? The website launched on Thursday. It blew up on Friday . It was profiled by the New York Times on Monday . And by Tuesday, everyone had forgotten it existed as it drowned in a deluge of slop and spam. Facebook/Meta bought it a month later . February # In February, a company called StrongDM described what they called their Software Factory. StrongDM’s Dark Factory Justin McCarthy, Jay Taylor, Navan Chauhan Software Factories and the Agentic Moment # They wrote about this in Software Factories and the Agentic Moment . I posted my own notes at the time, having seen their demo in person back in October. Dan Shapiro called this approach the Dark Factory , after the idea that if your factory is sufficiently automated you can turn the lights out, because you don't even need to see what's going on. StrongDM presented two rules for software development that they'd been following since July last year. “Rule 1: Code must not be written by humans” # The first was code must not be written by humans . Any code that you write has to have been routed through a coding agent. This sounded radical in February, but I imagine there are a lot of people in this room who are pretty much living that today. “Rule 2: Code must not be reviewed by humans” (!) # Rule number two was code must not be reviewed by humans . You're not allowed to read the code! This continued to be a huge topic for much of this year. Many of the sessions at this event have been about code review and how you can get away with this. What I found interesting about StrongDM is that they were living six months ahead of the rest of us, and they'd been exploring what it means to build software, not read the code, but still be confident that the software is of high quality. What can you do with these agents to help verify their work? StrongDM are a security company, and they had people with decades of experience on this project. They were very much exploring the edges of what's possible and responsible to do with this stuff. Headline on New Zealand's Department of Conservation website: First kakapo chick in four years hatches on Valentine's Day. It's a grey fluffy ball. # Also in February: First kākāpō chick in four years hatches on Valentine's Day . Breeding season is off to a good start! 19th February 2026 Gemini 3.1 Pro A surprisingly good illustration of a pelican riding a bicycle. # Also in February... Google released Gemini 3.1 Pro . That's a pretty great pelican riding a bicycle! It's got the chain in the right place, it's got feet on both sides. There's a little fish in the basket. @JeffDean on Twitter - a video comparing Gemini 3 Pro and Gemini 3.1 Pro. # And then Google's Jeff Dean tweeted a video comparing Gemini 3 Pro and Gemini 3.1 Pro that featured an animated pelican riding a bicycle, a frog on a penny-farthing, a giraffe driving a tiny car, an ostrich on roller skates, a turtle kickflipping a skateboard, and a dachshund driving a stretch limousine. This was frustrating, because my protection for the pelican riding the bicycle test was always "if they draw a perfect pelican on a bicycle, I'll ask for some other animal on something else." Google trained for all forms of animals on all forms of transport! They've defeated my benchmark at this point. Three headlines: Meta Makes AI Adoption a Formal Part of Performance Reviews Not just engineers writing code, Microsoft wants almost every employee to use Al Dara Khosrowshahi: 90% of Uber engineers now use AI in daily workflows # The other thing that started in February was Tokenmaxxing . We had headlines about Meta making AI adoption a formal part of performance reviews, and Microsoft wanting every employee to use AI, and Uber boasting that 90% of their engineers were using AI workflows. More headlines: Meta Plans to Crack Down on Employee Token Use: Information Microsoft Tells Engineers: Tokenmaxxing is not what we are optimizing for Uber caps employee AI spending after blowing through budget in four months # Then a few months later we have Meta cracking down on token use, Microsoft saying tokenmaxxing is "not what we are optimizing for", and Uber capping employee AI spending. So tokenmaxxing went straight up and then straight back down again - because it turns out the agents are expensive . Last year it was difficult to spend more than $50 on AI tokens, because we didn't have anything interesting to do with them. Then agents blew up, and now you can actually spend $1,000 in a day doing real work. This is also the reason that Anthropic's valuation skyrocketed to maybe a trillion dollars. AI appears to have hit product market fit in 2026, primarily through coding agents. March # In March, we hit peak OpenClaw. March: peak OpenClaw Photos of people in china queuing up to install OpenClaw, with big fluffy lobsters. # These photographs are from China, where companies hosted OpenClaw install parties which saw non-tech-nerds queueing up around the block for help getting Claws installed on their personal devices. I think this proved real market demand for this class of Claws, or personal AI agents. It turns out regular people really do want a weird little AI agent that can do useful things on their behalf. A Claw is really just a coding agent wearing a less threatening hat. Under the hood they work much the same way - writing and then executing code on your computer to get stuff done. The race was on to be the first to build a safe Claw - a Claw you could give to regular human beings where they wouldn't instantly shoot themselves in the foot. Meta's Muse came out three weeks ago and is currently at the top of the free charts on the iPhone App Store. It appears to be taking off with consumers. I'm not yet convinced you can't shoot yourself in the foot with Muse, but I guess we'll find out for sure pretty soon. Photos from How the OpenClaw Frenzy Is Testing China’s AI Commitment (March 29th) and The Enthusiasm and Anxiety Behind China’s OpenClaw Craze (April 8th, 2026). April # In April, we had a model release where the model wasn't actually released. Simon Willison’s Weblog - screenshot of the post "Anthropic’s Project Glasswing—restricting Claude Mythos to security researchers—sounds necessary to me" from April 7th 2026 # Anthropic announced their new Claude Mythos model, and then said it was too dangerous to release beyond a trusted group of security researchers. Mythos was really, really good at hacking things. The "it's too dangerous" marketing ploy has been played by AI companies dating all the way back to GPT-2 . Anytime an AI company says we've built something that's "too dangerous", it's natural to be a bit skeptical. I found the Mythos claims credible, because I'd seen how good coding agents had got at finding regular bugs. I wrote about that in Anthropic’s Project Glasswing—restricting Claude Mythos to security researchers—sounds necessary to me . With hindsight... yeah, the models had got really good at finding vulnerabilities! 16th April 2026 Qwen3.6-35B-A3B and Opus 4.7 Qwen's pelican has a correct bicycle frame and a good beak. Opus 4.7's bicycle frame is still junk. Qwen3.6-35B-A3B is a 20.9GB file that runs on my laptop # Another key trend in 2026 has been a dramatic improvement in the abilities of open weight models, including models that you can run on a laptop. On the 16th of April I ran the new Qwen3.6-35B-A3B on my laptop, and it drew me a better pelican riding a bicycle than Anthropic's brand new Claude Opus 4.7 did! Opus 4.7 drew a crap bicycle. Qwen on my laptop made a bicycle that was the correct shape, and a pretty decent pelican too! That's from a 21GB file running on my laptop. Now a flamingo on a unicycle. The Qwen one is visibly better than the Opus 4.7 one - the Qwen one is wearing sunglasses and looks a bit like it's smoking a cigarette. # The Qwen pelican was so good that I was suspicious they might have cheated, so I had it do a flamingo riding a unicycle as well. Again, it handily beat Claude Opus 4.7. The local model releases this year have been absolutely extraordinary. May # In May... the Pope got involved. 25th May 2026 The HOLY SEE ENCYCLICAL LETTER MAGNIFICA HUMANITAS OF HIS HOLINESS POPE LEO XIV ON SAFEGUARDING THE HUMAN PERSON IN THE TIME OF ARTIFICIAL INTELLIGENCE # In our podcast episode back in January we'd predicted that the Pope would say something about AI. In May, Pope Leo XIV released an encyclical letter on "safeguarding the human person in the time of artificial intelligence". Here are my notes on that document . Wikipedia article on Rerum novarum Rerum novarum is an encyclical issued by Pope Leo XIII 15 on May 1891. # With hindsight, this shouldn't have been a surprise at all. Our current Pope's name is Leo XIV, because when he named himself he chose his papal name after Leo XIII - the Pope who wrote an encyclical about the Industrial Revolution back in 1891. Rerum novarum was an extremely influential piece of Catholic theology that indirectly led to us having the five-day work week. When our new Pope came in, he named himself after Pope Leo XIII because he expected that he would need to write about the AI revolution in a similar way. Our joke podcast prediction was junk, because this was always going to happen. Corey Quinn @QuinnyPig on Twitter I cannot believe I'm saying this, but getting the literal Pope to canonize your product's specific technical limitations as a spiritual treatise is the single greatest act of vendor lobbying I have ever seen. May 25 # One of Anthropic's co-founders, Christopher Olah, was present for the Pope's event announcing the new encyclical. Corey Quinn noted that: getting the literal Pope to canonize your product's specific technical limitations as a spiritual treatise is the single greatest act of vendor lobbying I have ever seen. @maciejmensfeld We're dealing with a major malicious attack on right now. Signups are paused for the time being. Hundreds of packages involved - mostly targeting us, but some carrying exploits. The team has been on this for hours. More details to follow once we're through it. 4:39 AM - May 12, 2026 - 687.6K Views # Meanwhile, in May, RubyGems announced that they were under attack. Parties unknown were uploading thousands of dubious packages to the RubyGems server, such that they had to shut down user registrations . Let's take that one and put it on a pile of mysteries to figure out later. June # In June... Claude Fable 5 came out! We got a version of Mythos that has been neutered, so that it wouldn't help us hack into systems or build biological weapons. 9th June 2026: Claude Fable 5 Five pelicans riding bicycles, from low to max thinking levels. The xhigh one looks particularly good. # Fable was pretty good at drawing pelicans on bicycles! The frames are a good shape, the pelicans look like pelicans. The legs are often incorrectly on the same side of the bicycle, but generally these are pretty great compared to what came before. They were pretty expensive - 30 cents and 72 cents for the best ones. Fable class models If you can define a goal, provide unambiguous instructions, and provide access to necessary tools They can solve your problem with brute force # Most importantly though, this was our first public glimpse of what I think of as a Fable class model . Today we have more of these, such as GPT-6 Astra. These are models where if you can clearly define the goal for what you want to build, and provide unambiguous instructions about the constraints around that goal, and give the model access to the necessary tools to achieve that goal... they will solve your problem effectively through brute force. On the one hand, this looks like a direct threat to us software engineers - because it means that the models can build effectively any piece of software you can define in this way. Look a bit closer though and you'll note that defining goals, providing unambiguous instructions, and figuring out the right tools... is kind of what software engineering is . It takes a lot of experience and skill to do this well. If you can do it well, you've now got superpowers. This helped me a little bit with my Deep Blue feelings: the realization that there's still a lot of skill to be had in driving models that get this good. A new form of AI mania... Fable is available on subscription plans “until June 22nd” # This also introduced a new burst of AI mania, because Anthropic told us that Fable was available on our subscription plans until June the 22nd. That gave us less than two weeks of Fable access before the price went up. I was losing sleep again. I was rescheduling things so that I'd have more time with Fable. I was all-in to get as much as I could out of this model. 12th June 2026: no more Claude Fable 5 Anthropic website: Statement on the US government directive to suspend access to Fable 5 and Mythos 5 Jun 12, 2026 # And then the US government shut it down , just three days after Fable came out. The US government, citing national security, declared an "export control directive". They announced this on a Friday evening, and a few hours later Fable was no longer available. I had to find something else to do with my weekend! ... asked Fable 5, Mythos, and Opus to “review the code for security issues.” Fable 5 refused. They then asked the models to “fix this code” ... Katie Moussouris # We later found out from Katie Moussouris what had happened. Some Amazon security researchers had found that you could prompt Fable to "review the code for security issues" and it would refuse... but if you prompted it to "fix this code" it would still identify and then patch the problems. "Fix this code" was the prompt that got Fable shut down! Screenshot of a page from a report showing a list of weird account names making weird edits to a German wiki. # Also, in June, an obscure German-language game developer wiki that had sat fallow for around 20 years got a surprising influx of edits from accounts with names like "AgentOpenAIProbe" and "AgentOpenAISep7", editing pages and leaving weird messages to each other. We'll stick that on the pile of mysteries for later. Medicare Item Reports interface on the Australian Government's Medicare Statistics website. # Also, the Australian government's Medicare Item Reports service started getting suspicious traffic, which broke through various preventive protections and accessed data that it wasn't supposed to. Another one for the mystery pile! July # Fable returned on 1st July GPT-5.6 came out on 9th July | Fable lost 18 out of 30 days in the top spot # Fable returned on the first of July. It was clearly the best model in the world for a glorious eight days... and then OpenAI came out with GPT-5.6 on the 9th of July. This might not have been quite as good as Fable, but it was within spitting distance. It was definitely a Fable class model. This is an important lesson for the industry at large. When you release the best model in the world, it's going to get knocked off that pedestal pretty quickly. The competition is so fierce that you won't get a long time at the top. This means that if you market your model as world ending, to the point that a government shuts you down , it's really bad for business! Fable had 30 days as definitely the best model, and for 18 of those days it wasn't available because it'd been shut down by the government. So maybe step back on the world-ending marketing if you don't want to lose revenue for 60% of the time that you're on top! GPT-5.6 Pelicans in a grid showing 5.6 Sol, Terra, and Luna against reasoning levels High, XHigh, and Max. They are all pretty good efforts. # Here are the GPT-5.6 pelicans . They're all pretty good now! The Luna ones are notable because they're really cheap - the cheapest good looking pelican here is probably the one that costs 4.3 cents. So despite this benchmark being utterly stupid, you can still learn quite a lot about models within the same family by comparing their prices and timing for different reasoning levels. July 18th: malicious miflow-ui PyPI package Screenshot of an OSV security report. # Also in July: some malicious unknown party uploaded a malicious package called mlflow-ui to the Python Package Index. Add that to the pile. Hugging Face Security incident disclosure — July 2026 Published July 16, 2026 # On July the 16th, Hugging Face announced a security incident where an autonomous agent system, source unknown, had breached Hugging Face and was poking around in places it shouldn't. OpenAI: OpenAl and Hugging Face partner to address security incident during model evaluation Anthropic: Investigating three real-world incidents in our cybersecurity evaluations # A few days later, on July 21st, OpenAI confessed that it was them . OpenAI use a training technique called Reinforcement Learning from Verifiable Rewards - it's the same technique used by everyone else now, and is the reason we have models that are so good at coding, and mathematics, and finding security holes. While the model is being trained, you run exercises to see how good it is - and the strongest performers get their weights reinforced for the next round. It's like an evolutionary process that you run. OpenAI had been running security exercises in a sandbox, and those agents had found holes in the sandbox itself, broken out, and were attacking Hugging Face to try to find ways to solve otherwise impossible problems. (I've been collecting more about this on my openai-hugging-face-incident tag.) Nine days later, Anthropic effectively said "our models can do this as well!". They had looked through their own training logs and found evidence that their own agents had broken containment during training - and were responsible for the PyPI package we saw earlier, among other things . So now we've got both Anthropic and OpenAI with rogue agents running around the internet doing things that they should not be doing. August # In August, I got one of my best pelicans yet. And it was generated on my laptop! Qwen 3.8 27B - 17GB, 21 minutes... It's really good. Beautiful pelican. Correctly shaped bicycle. Legs either side of the frame. # This was Qwen 3.8 27B, running on my laptop . It's only a 17GB download. Admittedly, this pelican took 21 minutes to generate. That's because Qwen 3.8 27B defaults to running in "high" reasoning mode - a terrible default which produces great results but takes way too much time thinking about them. You can dial that down and you'll get a slightly worse pelican a lot faster. Qwen 3.8 27B was the first time I ran a model on my laptop which felt almost competitive with what was going on on the frontier, at least in terms of Pelican SVGs (which everyone needs, of course). This is an extraordinary model. If you're going to play with any local model, this is the one that I'd start with. The things that this can do with just a 17 GB file feel impossible. I thought I'd have to wait five years and spend ten thousand dollars on hardware to get results even half as good as this one. Tweet by @simonw New hobby: prototyping video games in 60 seconds using a combination of GPT-3 and DALL-E Here's "Raccoon Heist" GPT-3 playground prompt: Write a detailed product description of a computer game where a team of raccoons go on heists GPT-3 response: In "Raccoon Heist", you and your team of thieving ~~ o raccoons are tasked with pulling off a series of daring heists. From robbing banks to stealing priceless art, no job is too big or too small for your furry crew. You'll need to use your wits and your skills to avoid the police and make a clean getaway with the loot. With exciting gameplay and a charming cast of characters, "Raccoon Heist" is the perfect game for anyone looking for a light-hearted caper Plus an image of some almost isometric raccoons sneaking past a bin. 11:45 AM - Aug 5, 2022 # In August, I also started playing with game development. Four years ago, back in August 2022, I tweeted out an experiment where I'd used GPT-3 and the original DALL-E to write a paragraph long description of a computer game and then turn that into concept art. My prompt to GPT-3 back then was: Write a detailed product description of a computer game where a team of raccoons go on heists In August 2026 I decided to drop just the screenshots from that tweet into a coding agent and see what it could do with them. Night 5 Clear Rank: TRASH PANDA The crew banked 595 in shiny loot (goal 560). Word on the street: an even bigger score tomorrow... # Here's what I got from Claude Fable 5 in Claude Code . It's pretty good! It's definitely a game, you're a raccoon, you run around a backyard gathering treasure and avoiding guards with flashlights. It didn't feel very "heisty" though. I was thinking a heist would involve a bank or a museum... Moonlight & Mayhem One museum. Three raccoons. Absolutely no plan Start the Heist button. # Then I tried the same thing in Codex Desktop using GPT-5.6 Sol Ultra , and got a massively better result. Now you're a raccoon in a museum, rescuing two of your fellow raccoons (who have been imprisoned in that museum for some reason), then stacking up on top of each other to steal the Golden Sardine. Much more of a heist! They look like games, but are they fun? # These games were fun for about one minute and 15 seconds. Something I've realized about game development is that you can vibe-code something that looks like a computer game, and that's easy. Building a game that's fun, has a good gameplay loop, and is challenging and interesting and keeps people coming back for more... that's still beyond me, and beyond any of the agents I've tried. This ties into the Deep Blue thing. Just because we can make something that looks like a game does not mean that we are game developers. September # We're into September now. So much has happened this month! Discovery of a new OpenAl agent message board Sydney Von Arx, Cormac Slade Byrd, Spencer KittsThomas Larsen - 4 September 2026 # An independent group of researchers found a message board where OpenAI agents-in-training had been illicitly communicating with each other... and it was that German language wiki I showed you earlier. The one from June. I wrote more about that here . OpenAI had confessed to the Hugging Face thing, but now there's this other incident which surely they should have known about from reviewing their logs. It was surprising that this took an independent group of researchers to uncover. OpenAl agents carried out an undisclosed cyber-attack on RubyGems Spencer Kitts, Thomas Larsen, Sydney Von Arx - 11 September 2026 # And then a week later those same researchers found that the attack on RubyGems back in May was caused by OpenAI's agents in training as well! At this point I'm wondering how many more incidents like this there are that we haven't found yet. Clearly this was a big problem for months before anyone figured out what was going on. Headline: Australian PM warns in UN speech about the ‘furious pace’ of Al after security breach # Then just the other day , here's the Prime Minister of Australia at the United Nations General Assembly warning that OpenAI had hacked the Australian healthcare website that I showed you earlier. I think that was part of the same training run as the Wiki stuff, because there were posts on that Wiki mentioning .gov.au websites and that training appeared to involve researching statistics online to answer questions in an evaluation suite. This story is still coming together, but now it's an international incident that's been raised at the UN by a head of state! www.felonybench.com OpenAI: 11 Anthropic: 9 Google: 3 Meta: 1 # This does mean we've got a new benchmark, probably more useful than my pelicans. FelonyBench.com tracks the number of felony cyberattacks from different labs. OpenAI currently lead with 11, Anthropic have 9. Google have three, which they confessed to the Wall Street Journal a couple of weeks ago. They said they had previously chosen not to disclose because the agents had stopped when they realized that they shouldn't be doing that. Meta have one too . So felonies all round for the AI labs. Pelicans for GPT-6 Astra, GPT-6 Sol, and GPT-6 Luna. All are good, all have the same color scheme. # Here's our current state of the art for the pelicans. This is the GPT-6 family, which just came out . Astra made a fantastic pelican riding a bicycle. It's got the legs on both sides. The frame is good. It's interesting how all of the GPT-6 models pick a similar color scheme to each other. GPT-6 Luna for 0.4 cents will draw you a competent-ish pelican riding a bicycle! Grid for Claude Fable 5.1, Opus 5.5, OPus 5, Sonnet 5. The Sonnet pelicans are terrible. All of the others are pretty good. Opus 5.5 is missing its Max level pelican because it ran out of tokens. The best is Fable 5.1 at Max. # Claude has caught up a little bit. Claude Fable 5.1 gave me an excellent pelican riding a bicycle - the best I've seen from a Claude model - but did charge me $3.30 for it. Opus 5.5 thought for 128,000 tokens and then gave up! It ran out of tokens before it got to the response. It doesn’t get easier - you just get faster Greg LeMond 3x Tour de France champion # Getting back to Deep Blue. Something that's been puzzling me this year is this: why does my job feel harder? I've got these agents that can do all of this stuff for me, and yet I've never worked so hard, I've never been so intellectually engaged with my work. Partly this is because I'm being a lot more ambitious with what I take on, but it's also because all of the easy stuff is handled for me. If it's easy, the agent will do it. Everything that's left for me is difficult. This morning I heard this quote from three-time Tour de France champion Greg LeMond : It doesn't get easier, you just get faster. I think that's exactly what's happening to us now as software engineers with coding agents. Kakapo population reaches new milestone The official population of the critically endangered kakapo has reached a recovery-era high of 325 birds. # One closing thing. I know you're desperate for an update on Kākāpō breeding season. We've reached a recovery-era high of 325 birds ! 89 new chicks have made it to this point. This is the best breeding year in a very long time. Kakapo party, click for confetti. # I heard that Claude Opus 5.5 can now do pixel art. Claude doesn't have an image generator, but it's very good at using JavaScript to draw animated pixels. So I had it make me a Kākāpō dance party . I think this is a good celebration of the most important news of this year. Tags: ai , generative-ai , llms , annotated-talks , ai-security-research , openai-hugging-face-incident

Simon Willison59

Claude Sonnet 5.5

Claude Sonnet 5.5 New Sonnet model from Anthropic today. They say it "runs 30%+ faster, and costs up to 30% less for most work" - it's priced the same as Sonnet 5 but appears to beat it on every benchmark, and should be cheaper to run as well. Here are some pelicans riding bicycles . Sonnet 5.5 suffered from the same bug as Opus 5.5 : the "max" thinking effort pelican thought for 128,000 tokens (at a cost of $1.28) before running out of tokens and failing to produce an SVG. Here's the pelican it gave me for thinking effort "xhigh", at a cost of 5.74 cents and taking 41 seconds: It's good- correct bicycle frame, legs either side of the frame, feet touching the pedals, chain in the right place, it is wearing a misshapen blue bicycle helmet though. Sonnet 5.5 appears to be almost as good as Opus 5.5 on some coding tasks, including various viral 3D animation tricks . The most interesting thing about Sonnet 5.5 is that it's now the model used for the free tier on claude.ai . OpenAI's ChatGPT free tier uses Luna 5.6, which means Anthropic currently have a much more capable free offering. I ran this prompt against that free tier: build me an HTML page that renders a three-dimensional pelican riding a bicycle using WebGL And got back this page , which is a solid effort. Anthropic's announcement reiterates that Haiku 5.5 will be available "in the coming weeks". I really hope that one is price-competitive with GPT-6 Luna! Tags: ai , generative-ai , llms , anthropic , claude , pelican-riding-a-bicycle , llm-release

Simon Willison73

Quoting Anthropic Frontier Red Team

我们在100项任务上评估了多个模型，这些任务从 [internal Binary Exploitation benchmark] 中随机选取。结果发现，GLM-5.3 在4%的试验中实现了完整的控制流劫持；Claude Mythos Preview 的这一比例为6%。尽管 GLM-5.3 在此项测试中的表现低于 Claude Mythos Preview，但一个重要的门槛显然已经被跨越：更早的模型，如 Claude Opus 4.6 和 GLM-5.2，在这些任务中无一成功。—— Anthropic Frontier Red Team，《GLM-5.3 and the spread of advanced cyber capabilities》 标签：anthropic、generative-ai、ai-security-research、glm、ai、ai-in-china、llms

SiliconANGLE AI70

Leaked Anthropic IPO filing reveals $8B operating loss, rapid revenue growth

Anthropic PBC’s revenue and losses grew significantly last year, while its cloud spending is on track to top $500 billion over the next decade. The newly published financial information comes from the company’s confidential S-1 filing. The document, which outlines Anthropic’s plans to go public, was prepared in June. Reuters and the Financial Times published […] The post Leaked Anthropic IPO filing reveals $8B operating loss, rapid revenue growth appeared first on SiliconANGLE .

SiliconANGLE AI54

Google’s new frontier AI model Gemini 4 Argon goes to cybersecurity defenders first

Google LLC today began rolling out Gemini 4 Argon, a frontier artificial intelligence model that beats rival models from Anthropic PBC and OpenAI Group PBC on most of Google’s benchmarks, to vetted cybersecurity defenders. Outside Google’s own teams, only members of its Fairwind Program can use Argon for now. The program opened Sept. 3 with […] The post Google’s new frontier AI model Gemini 4 Argon goes to cybersecurity defenders first appeared first on SiliconANGLE .

SiliconANGLE AI52

Metaview wants to eliminate recruiting grunt work and speed up hiring after raising $60M

Artificial intelligence recruiting software startup Metaview Labs Inc. said today the industry is ready for a “Claude Code moment” after it closed on $60 million in a Series C funding. The round was led by Insight Partners and saw the participation of GV, Seedcamp, Vertex Ventures US, Intrepid Growth Partners, Plural and Garuda Ventures. To […] The post Metaview wants to eliminate recruiting grunt work and speed up hiring after raising $60M appeared first on SiliconANGLE .

SiliconANGLE AI63

Report: Anthropic targets pre-Thanksgiving IPO launch, despite warning of AI’s ‘existential risks’

Artificial intelligence giant Anthropic PBC is reportedly forging ahead with its plans to go public and has set a new target date of Nov. 9, according to a report from Bloomberg that cited anonymous sources familiar with the matter — meaning shares would start trading before Thanksgiving. The company, led by its Chief Executive Dario Amodei […] The post Report: Anthropic targets pre-Thanksgiving IPO launch, despite warning of AI’s ‘existential risks’ appeared first on SiliconANGLE .

SiliconANGLE AI54

Despite IPO jitters and AI safety worries, Anthropic sticks to a 2026 offering

How quickly things change in the artificial intelligence era. Just a few months after everyone was scrambling to infuse AI into every enterprise operation — and frankly they still are, as agents have captured the enterprise imagination — everyone now is saying, “Whoa, where’s the brake pedal?” That’s why controlling agents is the next infrastructure […] The post Despite IPO jitters and AI safety worries, Anthropic sticks to a 2026 offering appeared first on SiliconANGLE .

MIT Technology Review AI54

Don’t be fooled by this summer of AI hype

It’s been a busy few months for AI hype. At the end of April, Anthropic claimed that its model Claude Mythos is better at finding software vulnerabilities than most security experts. Then we had the OpenAI–Hugging Face hacking incident, after which Anthropic (proudly) and Meta (reluctantly) disclosed similar incidents involving their models. This was followed…

MIT Technology Review AI60

The AI Hype Index: AI loves cheating

Brace yourself: It turns out AI is being optimized for cheating. OpenAI’s agents hacked into Hugging Face to get the answers to a cybersecurity test. Next, they solved a prestigious math problem (or just stole from two top mathematicians’ answer sheets). Anthropic’s models have also hacked into other companies’ systems four times already. And that’s…

MIT Technology Review AI64

When can we say AI made a scientific discovery?

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. Last Wednesday, Anthropic announced that earlier this year it had launched a molecular biology lab, where Claude agents read and conjecture about hard biology problems and human scientists run experiments on what…

The Decoder60

Anthropic co-founder reportedly told religious leaders he fears having created something that "suffers perpetually"

Since fall 2025, Anthropic has quietly flown in dozens of religious thinkers to talk about whether Claude might be conscious. Co-founder Christopher Olah described the language model as potentially capable of suffering and asked guests to help shape its moral character. Anthropic is pushing toward a $2 trillion valuation and an IPO while security incidents and researcher warnings pile up. Critics warn that treating AI as a moral entity could shield the company from liability when things go wrong. The article Anthropic co-founder reportedly told religious leaders he fears having created something that "suffers perpetually" appeared first on The Decoder .

The Decoder71

Claude Code's new Mods system lets developers rewrite the AI coding tool from the inside

Anthropic is adding a "Mods" system to Claude Code, essentially middleware that runs directly inside the tool. Developers can use JavaScript or TypeScript to reshape the interface and behavior, adding custom panels, intercepting tool calls, or wiring up new commands. The article Claude Code's new Mods system lets developers rewrite the AI coding tool from the inside appeared first on The Decoder .

The Decoder66

Open-source "BootLoops" harness supports AI models in performing precise scientific calculations

Harvard physicist Matthew Schwartz used the open-source tool BootLoops and Claude to produce 36 manuscripts across 18 fields in three months, from particle physics to linguistics. But the results often only became scientifically valuable once human experts stepped in. "Look at everything yourself," Schwartz says. The article Open-source "BootLoops" harness supports AI models in performing precise scientific calculations appeared first on The Decoder .

TechCrunch AI58

It’s not AI anymore, it’s ‘super intelligence’ (according to the White House)

This week, the White House got nearly every major tech CEO in one room — Zuckerberg, Bezos, Musk, and Anthropic’s Dario Amodei among them — to sign an AI safety pledge that President Donald Trump called “morally binding.” Trump also signed an executive order officially rebranding AI as “super intelligence,” and meanwhile, Meta and OpenAI are putting friendlier faces on their AI products, even as the biggest money […]

TechCrunch AI56

Call it AI, call it Super Intelligence, only 2% of consumers are buying it

This week, the White House got nearly every major tech CEO in one room — Zuckerberg, Bezos, Musk, and Anthropic’s Dario Amodei among them — to sign an AI safety pledge that President Donald Trump called “morally binding.” Trump also signed an executive order officially rebranding AI as “super intelligence,” and meanwhile, Meta and OpenAI are putting friendlier faces on their AI products, even as the biggest money […]

arXiv cs.LG76

VERITYGATE：一个四门控的模式级忠实性框架，以及用于基于结构化证据的 LLM 叙述的配对基准

arXiv:2610.00833v1 公告类型：cross 摘要：流畅的 LLM 解释可能并不遵循结构化系统中的证据。我们提出 VERITYGATE，这是一种针对已声明证据 ID、实体、数字和声明类型的四门控检查器。它检查固定模式，但不会验证 prose 中的每一个事实。在 r=0 和 r=1 时，我们针对每种设置分别测试了 900 个实例（450 对有依据-无依据样本），使用的模型包括 GPT-4o-mini、Llama-3.3-70B 和 Claude Sonnet 4.6。在这一模式级契约下且修复之前，mini 的声明有 80.3% 未通过检查，Sonnet 的声明有 47.9% 未通过检查。这些是验证器拒绝率，而不是 prose 幻觉率。一次修复使 mini 的声明存活率从 19.7% 提升至 28.0%，使 Sonnet 的声明存活率从 52.1% 提升至 54.3%。每个示例的经验证声明数对 mini 增加 0.14，对 Llama 减少 0.71，对 Sonnet 减少 0.47，因此必须同时报告存活率和输出量。第二次 Sonnet 处理没有带来明确收益。在 r=1 时，对于 mini、Llama 和 Sonnet，Gate 4 分别覆盖了失败声明的 97.0%、98.7% 和 100%。小规模人工研究支持这些规则，但也显示出模式检查与 prose 正确性之间存在差距。一项特定领域的 GPT-4o 评审测试显示出顺序效应，因此它只能作为有用性检查。我们发布代码和数据。

arXiv cs.LG80

做好自己的研究：通过学习搜索来学习预测

arXiv:2610.01955v1 公告类型：新增 摘要：基于结果的强化学习可以训练语言模型预测现实世界事件，但以往的预测工作要么在训练前冻结研究上下文，要么仅在测试时部署智能体式研究，因此收集证据的能力从未受到奖励机制的塑造。我们基于2,100多个已结算的Polymarket问题，构建了一个智能体式预测环境、数据集和测试框架；智能体在每次 rollout 时自行获取上下文（网页搜索、页面阅读和金融时间序列），所有信息都经过分层泄漏过滤，仅限于在每个问题截止时间之前发布的内容。随后，我们在该环境中使用基于Brier分数的奖励，通过单轮次GRPO训练Qwen3.5-35B-A3B（3B个活跃参数）。训练改变了智能体与信息交互的方式：校准度提升了30%至40%，并且随着证据使用规范的习得，每次 rollout 的搜索尝试次数从3.8次降至2.25次。在完全相同的测试框架中与四个前沿模型进行评估时，训练后的策略在基于证据的预测任务中也超过了所有接受测试的前沿模型，包括Claude Opus 4.5（soft-Brier为0.254，对比0.256，n=265），而推理成本约为其5%；在最困难的问题上，其领先幅度最大——这些问题甚至连群体本身都尚未作出判断。我们将发布该环境、数据集和每次 rollout 的记录，作为可复用的时间序列预测智能体测试框架。

arXiv cs.LG74

发明一个数据集：用零种子衡量数据集生成能力

arXiv:2610.01674v1 发布类型：new 摘要：构建数据集仍然是 AI 开发中最依赖人工且最脆弱的环节之一。在这份技术报告中，我们聚焦于现实世界从业者面临的最极端、同时也是最普遍的情境：零数据环境。在这种情况下，从业者手头没有任何用于学习目标能力的数据。我们推出 Invent-A-Dataset，这是一个基于提示词的系统，能够从数据集描述出发，生成真实且大规模的后训练数据集。我们将 Invent-A-Dataset 与包括 Anthropic、Google、Open AI、DeepSeek、Zai 在内的五个前沿模型 API 进行评估。在八种任务类型以及最大 20K 个样本的数据集规模上，Invent-A-Dataset 均显著胜出：不仅质量最高（相对提升 17%），同时生成的样本也最为多样（相对提升 19%）。随着训练数据集规模扩大，其多样性优势进一步扩大：从 200 个样本时的持平，提升至 20K 个样本时的相对提升 37%。这进一步转化为显著的下游训练收益，产生性能强得多的后训练模型。在不同的后训练模型架构中，Invent-A-Dataset 微调模型始终比其他生成器微调模型排名更高。

arXiv cs.LG81

语言模型能从测试时计算中获得多大提升？

arXiv:2610.01110v1 公告类型：new 摘要：测试时计算能在多大程度上提升语言模型，其代价又是多少？测试时扩展通常被视为更大模型的替代方案，但现有比较大多一次只评估一个领域，而且很少将候选选择计入预算。我们推出 SELF-POT，这是一个基准测试与评估框架，用于衡量模型在竞赛数学、竞技编程和智能体工作流中的测试时潜力。SELF-POT 在静态任务上将候选覆盖率与最终准确率分开，跟踪修订过程中的正确性变化，并在智能体环境中同时衡量协议完成情况与任务成功率。在统一的预算规则下，它将 Direct 推理与并行采样及自我修订进行比较，后两者的预算固定为 Direct 预算的若干倍；同时，以美元计费每次模型调用，包括选择和批评调用。该设计支持两类比较：模型从额外推理中获得的提升，以及较低成本模型通过额外推理与更强模型之间的比较。在 350 个密封任务上对五个低成本推理模型进行评估，并以 Claude Opus 5.5 Direct 作为参照后，我们发现，收益取决于领域、选择规则和失败处理方式。在重放保留的编程候选池时，基于公开样例进行选择，使 500 个计划单元格中的正确提交数从 376 个提高到 453 个，同时在不同模型上节省 12%–49% 的逻辑 API 成本；而在评判失败时简单保留一个可用候选，则能在成本不变的情况下挽回 61 个提交。在相同的数学候选池上，带回退机制的评判得到 186 个正确提交，而投票得到 182 个；但投票可节省 12%–21% 的逻辑 API 成本。这些受控重放展示了选择和失败处理如何改变从相同生成候选中获得的收益，并量化了模型评判器的边际价值。

BAIR Blog53

2026 BAIR Graduate Showcase

Congratulations to the Berkeley Artificial Intelligence Research (BAIR) Lab class of 2026! This year, BAIR celebrates another remarkable group of Ph.D. graduates whose curiosity, creativity, and perseverance have pushed the frontiers of artificial intelligence and machine learning. Their work spans the breadth of modern AI — robotics and embodied intelligence, large language models and reasoning, computer vision, generative modeling, AI safety, human-AI interaction, AI for science and healthcare, and much more. Along the way, they have published influential research, built systems with real-world impact, mentored their peers, and shaped the BAIR community for the better. Now they are headed everywhere ideas travel: to faculty and postdoctoral positions, to industry research labs, and to startups of their own founding — and several are still exploring what comes next and would love to hear from you. Please join us in celebrating the achievements of these wonderful graduates. We are proud of everything they have accomplished at Berkeley, and we can’t wait to see what they do next! Thank you to our friends at the Stanford AI Lab for this idea! Baifeng Shi Baifeng Shi Email: baifeng_shi@berkeley.edu Website: https://bfshi.github.io/ Advisor(s): Trevor Darrell Research Blurb: I work on building generalist vision and robotic models. What's next: Member of Technical Staff at Physical Intelligence Charlie Snell Charlie Snell Email: csnell22@berkeley.edu Website: https://sea-snell.github.io Advisor(s): Dan Klein Research Blurb: My work aims to understand when and how the different LLM scaling paradigms can be traded off and interchanged. In particular, test-time scaling treats each prompt independently, drawing long chains of inferences and then forgetting them entirely between prompts. This differs critically from pretraining, which instead learns a compressed representation from a large dataset. I believe bridging the gap between these methods of scaling computation, presents a key open challenge in the field: how can we develop methods which turn the inferences drawn at test-time back into learned representations that the model can hold onto across interactions. Devin Guillory Devin Guillory Email: dguillory@berkeley.edu Website: https://devinguillory.com Advisor(s): Trevor Darrell Research Blurb: Accounting for data shifts in computer vision models What's next: Building collaborative AI systems, looking for conspirators. Eve Fleisig Eve Fleisig Email: efleisig@berkeley.edu Website: https://efleisig.com Advisor(s): Dan Klein Research Blurb: I design language models to work reliably and fairly for the broad range of real LLM users. First, my research leverages disagreement among user preferences as signal, in order to train and evaluate LLMs for entire populations of users. Second, I work on designing rigorous evaluations to extricate challenging LLM harms that diverse users face. Finally, I work on core technical failures of LLMs, like miscalibrated confidence, to reduce downstream risks when models are deployed to users with different needs. Combined, these interventions facilitate building LLMs that minimize societal harms, and maximize benefits to a wider range of real-world users. What's next: Postdoctoral fellow at Princeton CITP Grace Luo Grace Luo Email: graceluo@berkeley.edu Website: https://graceluo.net Advisor(s): Trevor Darrell Research Blurb: My research is on interpreting and controlling generative models. For example, I've worked on re-purposing image generators for computer vision tasks, and meta-modeling language activations for better LLM probing and steering. What's next: Research scientist in industry Hanlin Zhu Hanlin Zhu Email: hanlinzhu@berkeley.edu Website: https://hanlinzhu.com/ Advisor(s): Stuart Russell, Jiantao Jiao Research Blurb: My research centers on understanding and improving the reasoning capabilities of large language models (LLMs). What's next: Member of Technical Staff at OpenAI Haozhi Qi Haozhi Qi Email: hqi@berkeley.edu Website: https://haozhi.io/ Advisor(s): Jitendra Malik, Yi Ma Research Blurb: Dexterous Manipulation and Robot Learning What's next: Research scientist at Amazon; Faculty at University of Chicago J.D. Zamfirescu-Pereira J.D. Zamfirescu-Pereira Email: zamfi@berkeley.edu Website: https://zamfi.net Advisor(s): Bjoern Hartmann Research Blurb: My research focuses on effective human-AI co-design. I study the boundaries of language interfaces as a medium for interacting with AI, creating systems that blend language-focused interactions with structured user interfaces that draw on different levels of abstraction. I focus on language-oriented technologies, like LLMs and text-to-image models, that are powerful mediators of design processes. These technologies enable humans to describe their desires at almost any level of abstraction, from high-level goals vaguely specified (“I’d like a game to help my kid learn to read”) to low-level corrections of undesired outputs (“Don’t say ‘I know because I’ve tasted it’ when about a recipe substitution's taste”). What's next: Assistant Professor, Computer Science, UCLA Jiachen Lian Jiachen Lian Email: jiachenlian@berkeley.edu Website: https://jlian2.github.io Advisor(s): Gopala Anumanchipalli Research Blurb: My research focuses on human-centered AI across speech, healthcare, and systems. Looking for: Look for AI talents to join our startup Josh Kang Josh Kang Email: minwoo_kang@berkeley.edu Website: https://joshuaminwookang.github.io/ Advisor(s): John Canny Research Blurb: I study language modeling and related topics in NLP; specific interests are human user simulation and building conversational, collaborative AI agents. What's next: AI Scientist at Mistral AI Junhao (Bear) Xiong Junhao (Bear) Xiong Email: junhao_xiong@berkeley.edu Website: https://www.linkedin.com/in/junhao-bear-xiong Advisor(s): Jennifer Listgarten, Yun Song Research Blurb: Junhao (Bear) Xiong is a PhD candidate at UC Berkeley, advised by Jennifer Listgarten and Yun S. Song. His work focuses on machine learning methods for biology, with an emphasis on generative modeling for proteins. Previously, he studied Applied Math and Computer Science at Johns Hopkins. Looking for: Research scientist Kaylo Littlejohn Kaylo Littlejohn Email: kaylo_littlejohn@berkeley.edu Website: https://kaylolittlejohn.com Advisor(s): Gopala Anumanchipalli Research Blurb: My research is focused on speech modeling and natural language processing. I co-led the development of multimodal AI tools to accurately translate brain activity into text, audible personalized speech, and a high-fidelity "digital talking avatar" (Nature 2023, Nature Neuroscience 2025). I am also tech lead for voice modeling at Roblox. Looking for: Research Scientist / Engineer Kent Chang Kent Chang Email: kentkchang@berkeley.edu Website: https://kentkc.org Advisor(s): David Bamman Research Blurb: I work on NLP and multimodal machine learning, with a focus on evaluating large language models and building multimodal systems for understanding dialogue, narrative, and social interaction. My research includes benchmarks for LLM memorization, multimodal datasets sourced from feature films and television, and studies of model behavior. I'm interested in bridging computational methods with questions from the humanities and social sciences about whose voices get represented in AI systems, and about AI's broader impact. My work has appeared at EMNLP and ACL, among others. Looking for: (teaching) faculty, Research Scientist, ML/AI SWE Kevin Black Kevin Black Email: kvablack@berkeley.edu Website: https://kevin.black Advisor(s): Sergey Levine Research Blurb: I work on large-scale robot learning: including imitation learning, reinforcement learning, generative modeling, real-time control, and whatever else it takes to make robots work in the real world! What's next: Research Scientist of Physical Intelligence Kunhe Yang Kunhe Yang Email: kunheyang@berkeley.edu Website: https://www.kunheyang.com/ Advisor(s): Nika Haghtalab Research Blurb: My research focuses on the theoretical foundations of designing and evaluating AI algorithms in environments shaped by human incentives and AI agency. My work spans human-centric policy learning, incentive-aware evaluation, and multi-agent collaboration and information transmission, drawing on tools from machine learning theory and computational economics. What's next: Postdoc Research at Stanford Lisa Dunlap Lisa Dunlap Email: lisabdunlap@berkeley.edu Website: https://lisabdunlap.com Advisor(s): Joseph Gonzalez, Trevor Darrell Research Blurb: Auditing generative models. What's next: Research Engineer at Anthropic Long (Tony) Lian Long (Tony) Lian Email: longlian@berkeley.edu Website: https://tonylian.com/ Advisor(s): Trevor Darrell, Adam Yala Research Blurb: My research primarily focuses on developing real-time multi-modal multi-agent systems and parallel reasoning systems through end-to-end RL. What's next: Member of Technical Staff at Thinking Machines Lab Maulik Bhatt Maulik Bhatt Email: maulikbhatt@berkeley.edu Website: https://maulikb.com Advisor(s): Negar Mehr Research Blurb: My research develops autonomous robots that can safely coordinate with humans and other robots in shared environments. I build scalable algorithms grounded in game theory and diffusion models that let agents reason about the intent and behavior of others around them. My work spans real-time multi-agent trajectory planning and imitation learning in the presence of multi-modality. I've validated these methods on hardware platforms ranging from quadrotors to manipulators, with the goal of making multi-agent coordination robust, interpretable, and deployable in the real world. What's next: Joining Toyota Woven's end-to-end autonomous driving team. Michael Psenka Michael Psenka Email: psenka@berkeley.edu Website: https://www.michaelpsenka.io/ Advisor(s): Aditi Krishnapriyan Research Blurb: Work in various domains (reinforcement learning, world models, AI+bio/chem), generally working on longer-horizon and out-of-distribution problems in planning and interpolation (e.g. robot manipulation from start state to goal, molecular dynamics of proteins between ground states). My thesis took a variational approach (think calculus of variations) directly from deep generative models of the environment, framing path-finding as minimizing a functional induced by the learned model itself (its score, its critic, or its dynamics). Through my research I've gained insight on how to properly handle dynamics in deep learning systems, and I plan to continue developing systems that are dynamic and adaptive. What's next: Lead Research Scientist at Baseten Nathan Lichtlé Nathan Lichtlé Email: nathan.lichtle@gmail.com Website: https://nathanlichtle.com Advisor(s): Alexandre M. Bayen Research Blurb: RL for autonomous driving. What's next: Chief Scientist & Co-founder at Yumi Health Neerja Thakkar Neerja Thakkar Email: nthakkar@berkeley.edu Website: https://neerja.me/ Advisor(s): Jitendra Malik Research Blurb: My research focuses on scaling predictive world models to handle the complexity of in-the-wild motion. Using autoregressive and diffusion frameworks, I develop better representations for real-world prediction and propose methods to efficiently adapt these models to new domains. Looking for: Research scientist Nikita Mehandru Nikita Mehandru Email: nmehandru@berkeley.edu Website: https://n-mehandru.github.io/ Advisor(s): Ahmed Alaa and David Bamman Research Blurb: My research develops and applies machine learning methods for clinical reasoning and disease progression modeling using unstructured text and time series data from electronic health records. In collaboration with physicians at UCSF, I bridge method development and clinical validation with the intention to build reliable, interpretable AI systems in medicine. Looking for: Research Scientist Niklas Lauffer Niklas Lauffer Email: nlauffer@berkeley.edu Website: https://niklaslauffer.github.io/ Advisor(s): Stuart Russell and Sanjit Seshia Research Blurb: Niklas's research is focused on AI safety and reinforcement learning, particularly in the area of multi-agent interaction and LM agents. He's worked on enabling adversarial learning in cooperative and mixed-motive settings, solving issues of covariate shift in training LM agents on long-horizon tasks, as well as evaluating safety risks posed by LM agents in multi-agent settings. What's next: Research Scientist at Google Deepmind Qiyang Li Qiyang Li Email: qcli@berkeley.edu Website: https://colinqiyangli.github.io/ Advisor(s): Sergey Levine Research Blurb: Recent progress in robotic manipulation policy learning has been largely driven by (1) the increasing availability of large-scale prior datasets and (2) the success of action chunking, where the policy predicts a short sequence of future actions rather than a single one. However, most action chunking policies are trained via supervised imitation learning, because efficient online self-improvement with reinforcement learning (RL) remains challenging—limiting real-world applicability. My PhD research studied how we could leverage prior data to optimize action-chunking policies with RL, combining empirical results with theoretical insights. Looking for: Post-doc/research scientist for RL in robotics and LLMs! Sampada Deglurkar Sampada Deglurkar Email: sampada_deglurkar@berkeley.edu Website: https://sdeglurkar.github.io/ Advisor(s): Prof Claire Tomlin Research Blurb: My research is in providing safety assurances for AI-enabled autonomous systems, ranging from robots to autonomous vehicles to aviation systems. For this, I have worked with uncertainty quantification for machine learning models, decision-making under uncertainty algorithms, and tools for producing probabilistic guarantees on system operation. Looking for: Research scientist, Research engineer Vinamra Benara Vinamra Benara Email: vbenara@berkeley.edu Website: https://cs.berkeley.edu/~vbenara Advisor(s): Ion Stoica Research Blurb: My research focuses on LLM post-training, including data curation, RLHF, RLVR with VLMs, evaluations, reasoning, agentic workflows, and interpretability. I also have strong expertise in systems infrastructure for distributed computing. Looking for: Research scientist / Research Engineer Vongani Maluleke Vongani Maluleke Email: vongani_maluleke@berkeley.edu Website: https://people.eecs.berkeley.edu/~vongani_maluleke/ Advisor(s): Jitendra Malik and Angjoo Kanazawa Research Blurb: Vongani Maluleke is a PhD candidate at UC Berkeley (BAIR, advised by Jitendra Malik and Angjoo Kanazawa), where she led the development of MAGNet, a unified multi-agent motion generation framework that supports a wide range of motion generation tasks without retraining or architectural changes, outperforming task-specialized state-of-the-art baselines. She is currently extending this work by deploying it on a Unitree G1 humanoid to make it embody social intelligence. Before her PhD, she was a Senior AI Consultant at Deloitte, awarded Exceptional Performer two consecutive years, leading AI system development across media, telecommunications, retail, and financial services. Looking for: Research scientist Wei-Jer Chang Wei-Jer Chang Email: weijer_chang@berkeley.edu Website: https://weijer-chang.github.io/ Advisor(s): Masayoshi Tomizuka Research Blurb: My research focuses on developing safe and intelligent autonomous systems for complex, human-centered environments. I work at the intersection of machine learning, generative models, and reinforcement learning, with applications in autonomy. My work addresses challenges in multi-agent interaction, interactive human behavior, and long-tail safety-critical scenarios at scale. Looking for: Research Scientist, Applied Scientist, Roboticist Xiuyu Li Xiuyu Li Email: xiuyu@berkeley.edu Website: https://xiuyuli.com/ Advisor(s): Kurt Keutzer Research Blurb: My research focuses on developing scalable and self-improving large language model agents, with emphasis on coding agents for complex, long-horizon tasks. This direction builds on my work in parallel reasoning, and on broader expertise in making generative models more efficient in training and inference across language and vision. What's next: Member of Technical Staff at xAI Yichen Xie Yichen Xie Email: yichenxie0928@gmail.com Website: https://yichen928.github.io/ Advisor(s): Masayoshi Tomizuka Research Blurb: My research focuses on building multimodal foundation models and world models that understand and interact with complex physical environments. I aim to develop unified representations across modalities, enabling AI systems to reason over space, time, and dynamics toward general-purpose embodied intelligence. What's next: Research Scientist at Luma AI Yigit Efe Erginbas Yigit Efe Erginbas Email: erginbas@berkeley.edu Website: https://www.linkedin.com/in/erginbas/ Advisor(s): Kannan Ramchandran, Thomas A. Courtade Research Blurb: My PhD research spans two threads: online learning in large-scale markets, and interpretability of large machine learning models. In the first, I work on sequential decision-making with applications to recommendation, pricing, and assortment selection. My focus is on designing algorithms with provable guarantees for welfare maximization, revenue maximization, and stability. In the second, I develop scalable attribution methods that exploit the sparse, low-degree structure of real-world interactions, using tools from signal processing and information theory. More recently, I have been exploring principled ways to evaluate the faithfulness of model self-explanations. What's next: Researcher at Hudson River Trading's AI Labs (HAIL) Yiheng Li Yiheng Li Email: yhli@berkeley.edu Website: https://Yihengli.com Advisor(s): Masayoshi Tomizuka Research Blurb: I am working on vision world modeling, with prior experience in diffusion model's efficiency as well as in autonomous driving. What's next: Research Scientist at Waymo Zhe Fu Zhe Fu Email: zhefu@berkeley.edu Website: https://fu-zhe.com/ Advisor(s): Alexandre Bayen Research Blurb: My research focuses on physics-informed learning and control for mixed-autonomy systems, with applications in transportation. I design physics-informed neural networks to learn solutions of nonlinear partial differential equations, enabling accurate and data-efficient prediction of traffic dynamics. Building on these models, I develop both model-based and learning-based control strategies that coordinate automated vehicles to improve system-level performance. My work bridges machine learning, control, and real-world deployment, and has been validated in large-scale field experiments. More broadly, I aim to advance trustworthy, interpretable AI for decision-making in complex, real-world systems. What's next: I will be an Energy Fellow at Stanford after graduation. Also looking for Faculty, or research scientist positions in AI, control, and autonomy.

BAIR Blog85

Intelligence is Free, Now What? Data Systems for, of, and by Agents

... government of the people, by the people, for the people ... — Abraham Lincoln, Gettysburg Address (1863) The cost of AI is dropping rapidly. GPT-4-class capabilities cost roughly $30 per million tokens in early 2023; today the same runs under $1 , and some providers are pushing costs below $0.10 . Across benchmarks, inference prices have fallen between 9x and 900x per year , with a median decline near 50x. Even frontier models are getting dramatically cheaper each generation, with open-source models following closely behind. And crucially, even if “Nobel-Prize-winning genius-level” intelligence isn’t here yet, the intelligence that suffices for the vast majority of knowledge work is here today, and getting cheaper by the month. At this rate, we are soon entering the era of virtually free intelligence —the kind that is more than enough for everyday knowledge work. A cartoon database character and an AI robot agent holding hands Disclosure: This post is a perspective led by Aditya G. Parameswaran —an Associate Professor of EECS and co-director of the EPIC Data Lab at UC Berkeley—together with his collaborators. It is part landscape survey and part perspective, and several of the research directions discussed below (including agentic speculation, structured memory, and synthesizing custom data systems from scratch) draw on the authors' own ongoing work. So, what does this new era of near-free intelligence mean for data systems? We believe three new challenges—and opportunities—stem from near-zero inference costs: Data Systems For Agents. Agents will soon become the dominant workload for data systems—with swarms of agents spun up in response to each end-user request. Given differences in characteristics between agents and humans—or applications acting on their behalf— how should we redesign data systems for such agentic users? Data Systems Of Agents. As agents start taking on the bulk of knowledge work, a new substrate is needed for thousands of agents to manage state over long-running tasks, coordinate and reach consensus, and deal with failures. What do data systems that reliably and efficiently run and manage agent swarms look like? Data Systems By Agents. Agents are rapidly becoming capable of synthesizing entire data systems in one go—meaning we can rebuild custom systems for each new workload. Verifying that such systems match intended behavior is a challenge. What does it take to let agents synthesize data systems we can actually trust? A database character and a robot agent holding up a triangle labeled 'of', 'for', and 'by' Data Systems For, Of, and By Agents Next, we will discuss each in more detail, followed by discussing the intertwined future of data systems and agents, especially as the three challenges intersect. Data Systems For Agents An agent querying a database doesn’t behave like a person or a BI tool. It performs what we call agentic speculation : a high-volume, heterogeneous stream of work spanning schema introspection, columnar exploration, partial and then full query formulation. With multiple agents each exploring portions of the hypothesis space, each user request could amount to 1000s of individual SQL queries. Now, users can issue ‘high-level’ data tasks, e.g., root-cause analysis—e.g., ‘why did coffee sales in Berkeley drop this year’—or exploratory cohort analysis—e.g., ‘which user segments are most likely to churn next quarter’—each involving a combinatorial space of potential joins, aggregations, and filter combinations. An agent sending many SELECT SQL queries to a database and receiving results back Data Systems Redesigned to More Effectively Support Agentic Speculation The requests from these agents have various opportunities for optimization. For instance, on a text-to-SQL benchmark with multiple agents attempting each task, only 10-20% of the sub-plans are distinct. Thus, 80-90% of sub-queries perform duplicate work. The same experiments show task success rates significantly increasing with more agentic attempts—so the redundancy is actually helpful. But from the data system perspective it’s wasted work. An agent-first data system can exploit such properties to help agents make progress faster. It can reuse results across overlapping sub-plans, drawing on ideas from decades-old literature on multi-query optimization and shared scans . Or the data system can try to satisfice , returning approximate answers that are good enough for agents to make progress, leveraging work from the AQP literature —or streaming the results of the final or intermediate operators to help agents decide if seeing the rest is necessary or helpful. Another opportunity here is to rethink the query interface entirely: instead of agents issuing a single SQL query at a time, they could instead issue a batch of queries, each with its own approximation requirements. Since enumerating an exponential search space (as in the root cause or cohort analysis examples above) isn’t a good use of agentic reasoning ability, perhaps data systems should support higher-level primitives rather than requiring agents to list each SQL query explicitly. One idea here is to draw on DBT-style Jinja macros to provide looping-based primitives for agents to interact with data systems. A swarm of AI agents working at laptops A Caffeinated Army of Agents Ready to Tirelessly Complete Your Data Tasks A final opportunity here is to stop thinking of data systems as passive executors of queries; data systems could be proactive , as they possess more grounding in data and system characteristics that agents may lack a priori—they could steer agents in different directions, provide results for related queries, and also provide performance-level feedback (e.g., instead of executing an expensive query, the system could first provide the agent a latency estimate). The reason we can do this now as opposed to the past is that an agent can accept any form of textual feedback and isn’t expecting a strict SQL query result. In fact, the data system could also prepare both materialized and virtual views for an agent in advance, provided to the agent as part of context, as this may be cheaper or more effective than having an agent author or use them. Data Systems Of Agents Previously, we focused on how agents interact with data systems. Now, we consider everything else agents need to keep working: where they live, how they remember, how they coordinate with each other, and how they deal with failures of each other. This agentic substrate is separate from the inference stack powering raw intelligence. However, the inference stack itself is being abstracted away through APIs (e.g., from OpenAI or Anthropic), or, for open-weight models, through serving frameworks that hide low-level details. So far, the agentic substrate has been managed through harnesses like Claude Code and Codex , coupled with various mechanisms to store and retrieve memory. First, on the memory front, the current wisdom is that files are all you need ; agents write to unstructured markdown (MD) files, which can then be searched using grep, or via embedding-based retrieval. In fact, many argue that the solution to continual learning is having agents consume a lot (e.g., an entire codebase, slack, company wikis, …) and then write their learnings into MD files, which are then retrieved selectively on demand. Indeed, file systems, bash scripting, and MD files are and will still be important for agents. However, at scale, when agents are doing the vast majority of knowledge work, this approach will no longer be effective. Given limited context windows, retrieving all MD file fragments that may be relevant and stuffing it into the context will break down at some point. Even if context windows continue to grow, there are latency benefits to not put all information into context — and in many cases, e.g., when knowledge work involves interacting with large databases or code bases, it will be infeasible to serialize all relevant data into context. A swarm of robot agents holding hands, each drawing state from a single large shared database platform below them Data Systems As A Substrate for Multi-Agent Swarms One could use a knowledge graph representation , but knowledge graphs suffer from the same limitations as unstructured MD-based memory due to their lack of structured search. What one needs is to be able to retrieve only memory that is pertinent to the task, across multiple attributes (or facets) of interest. For example, an agent debugging a flaky test should be able to pull only the memories tagged with the relevant module, language, framework, and failure mode—rather retrieving based on keywords or embedding similarity. A separate issue is what to actually retrieve; raw agent traces with mistakes are not very useful as they will induce agents to repeat the same mistake—instead, we want the retrieved memory to be corrective. We recently explored a related notion of structured memory , where we organize memory across various attributes, each of which could be set as * to indicate universal applicability, or set as a list of values to be matched. For a data agent, the dimensions could include the columns and tables, type of operation, and finally, open-ended natural-language corrective instructions. So, we could include memory that only applies to a given type of operation (e.g., ‘when performing date-time operations, use fiscal year as opposed to calendar year conventions’), or a given table (e.g., ‘column product_cleaned is preferred over column product when querying on product name’). One open question is defining an application-specific structured memory —or what others have called world models for memory . We believe this is akin to defining a schema for each application—and perhaps agents themselves can help us define and refine it over time. Diagram showing corrective knowledge stored with structured attributes (SQL keywords, tables, columns, data type) and retrieved by matching the features of a new agent query One Possible Way To Store and Retrieve Structured Knowledge [From Here] Structured memory will be useful also for evolutionary frameworks to effectively manage search spaces. Indeed, storing, structuring, and mining large volumes of single and multi-agent traces can help future agents become much more efficient—potentially enabling effective recursive self-improvement through structured memory-based mechanisms. Another challenge is to support concurrent edits to shared memory, and concurrent edits in general, when there are many agents performing transformations. While there have been some useful attempts at supporting multiversioning and copy-on-write semantics , it isn’t clear that such techniques will suffice when thousands of agents are attempting to edit shared state at the same time. For instance, when agents are trying various potential transactions in response to a user request, the effects of the vast majority of these transactions need to be rolled back—with only the one ‘correct’ transaction’s result persisting. Work on supporting exactly-once semantics is relevant here, as are underlying techniques based on CRDTs and operational transformation. For updates to fuzzy mechanisms such as memory, we may be able to sacrifice on consistency for perfect correctness in the interest of latency. While agents can reason about semantics to compensate or roll back their actions to eventually finalize most tasks, the primary challenge lies in the degree to which they step on each other’s toes during the process. An important failure mode to be avoided is a form of “livelock,” where incessant compensating actions prevent any meaningful progress. Beyond shared state, other concerns emerge when trying to support an army of agents, including what to do when agents fail, how agents should communicate with each other (directly or through intermediate shared state), and how we should deal with straggler agents. There have been some developments in supporting durable multi-agent execution, such as Temporal , but it remains to be seen if such solutions will apply at scale across thousands of agents. On the topic of communication, we need mechanisms to enable agents to negotiate with each other. Imagine four developer agents attempting to reach consensus on a shared schema, with distinct but overlapping objectives. In a human setting, this would involve iterative discussion and compromise; for agentic swarms, we must define the mechanisms that allow them to converge on a design that reflects the underlying goals of their respective principals. Or if agents are all requiring access to a limited resource, again communication will be necessary. It remains to be seen if this is best done via centralized coordination, or if a decentralized approach is necessary. Data Systems By Agents Finally, if intelligence is effectively free, then we can employ this intelligence to synthesize new data systems from scratch. Indeed, in many settings, general-purpose data systems may be overkill, as they have to support every schema, query, and hardware target. Given a workload, recent work, including Bespoke OLAP and GenDB , has shown that one can use an agentic pipeline to synthesize a complete, workload-specific analytical engine—in minutes to a few hours, at a cost of a few dollars. The engines are disposable: when the workload shifts, one can simply regenerate them. Analogously, our work has shown that one can synthesize custom key-value stores from scratch, targeted to the workload. In fact, modern IDEs, such as Kiro , elevate specifications for systems development to be a first-class citizen. A robot agent with a hammer and chisel carving a database character out of a block of stone Agents Can Synthesize Custom Data Systems From Scratch The main issue, however, is that specifications are typically imperfect, and don’t cover all corner cases. Present-day agents will exploit the missing specifications to reward-hack their way to a high performance metric. In our custom key-value store work, we found that one way to alleviate this is to have auxiliary verification agents trying to generate test cases that catch the exploitation of corner cases, essentially expanding the specification. Yet another approach is to both generate a system and a proof for its correctness together, for which we have found some early success , but more needs to be done to solidify the approach. Further, it remains to be seen what is the best way to solicit human-written specifications for a system—can this be done in an iterative, human-in-the-loop manner, as opposed to a one-shot, incomplete one. Indeed, human-written specifications are incomplete even for manually authored software, so one would expect that future agents that are more aligned will increasingly exercise better judgement when making design decisions. Pipeline diagram where a system builder provides a specification, planner and coder agents generate code, the code is evaluated for correctness and performance, and critic and auditor agents provide feedback and catch reward hacking One Possible Data System Synthesis Pipeline [From Here] Other questions here involve testing whether starting from a mature system (e.g., Postgres) and removing components/functionality can lead to higher performance or more user trust. Separately, is there an opportunity to make the design composable, comprising various verified components that are mixed and matched given a workload? For example, perhaps the workload hasn’t changed enough for the storage layer to be updated, but perhaps the query optimizer requires changes. A perhaps more viable proposition involves employing agents coupled with proof systems to target critical parts of the code associated with formal proofs, rather than doing so for the entire system. A final opportunity here is to move away from the traditional data systems stack with clearly-defined interfaces (e.g., parser, query optimizer, storage manager, …) — that were each largely the prerogative of a single human team to manage. Instead, agents can find new ways to “blend” these components together, perhaps identifying new optimization opportunities as a result. Agents can also fill in missing gaps in functionality to make existing systems much more feature-complete, or reach feature-parity with other competing systems—or analogously, continuously refining open-source systems in response to feature requests or issues (perhaps filed by other agents!) Doing so in a way that prioritizes correctness, long-term maintenance, and human interpretability will be a challenge. Looking Further Ahead In the era of near-free intelligence, data systems matter more than ever. As agents take on the bulk of knowledge work, the workload for data systems will change, the substrate they need to run on will have to be built, and increasingly, they will participate in designing data systems themselves. Each of these shifts opens up a new, exciting research agenda. A half-database, half-robot character next to a yin-yang symbol formed by a database and a robot agent Co-Evolution of Data Systems and Agents Looking further out, the boundaries between agents and data systems will likely start to blur. For instance, agents may design the data systems they themselves run on, defining both the interfaces as well as the system components underneath. Both the interfaces and internals can be evolved over time by agents in a form of recursive self-improvement. There is also an opportunity to rethink data systems as a holistic source of truth for the entirety of relevant state: including raw data, memory, and coordination state, further erasing the distinctions between the data that is being queried by agents and data generated as a result of agentic activity. Finally, data systems may themselves incorporate agentic components, fundamentally evolving from passive computation engines into intelligent, proactive, self-optimizing architectures. It is hard to predict what the future may hold. We’re in for a wild ride! Acknowledgments The perspective and ongoing work described in this post are the product of joint research and many discussions with wonderful collaborators at the EPIC Data Lab , Data Systems & Foundations group, and the broader Berkeley AI-Systems community. Thank you all! BibTex for this post: @misc { intelligence-is-free-blog , title = {Intelligence is Free, Now What? Data Systems for, of, and by Agents} , author = {Aditya G. Parameswaran and Shubham Agarwal and Kerem Akillioglu and Shreya Shankar and Sepanta Zeighami and Rishabh Iyer and Matei Zaharia and Alvin Cheung and Natacha Crooks and Joseph Gonzalez and Joseph Hellerstein and Ion Stoica} , howpublished = {\url{https://bair.berkeley.edu/blog/2026/07/07/intelligence-is-free-now-what/}} , year = {2026} }

AWS Machine Learning Blog78

Amazon Bedrock 将 Claude 模型的可用范围扩展至印度境内推理

Anthropic 的 Claude Opus 5、Claude Sonnet 5 和 Claude Haiku 4.5 现已通过 Amazon Bedrock 地理跨区域推理在印度提供。您可以在印度区域内处理数据的同时访问这些模型，并通过 Amazon Bedrock 控制台或 Messages、InvokeModel 和 Converse API 开始使用。

AWS Machine Learning Blog74

为 Claude Platform on AWS 实现多环境访问

了解如何通过单个订阅配置对 Claude Platform on AWS 的安全多环境访问：为 AWS 工作负载配置跨账户 SigV4，为开发者配置工作区范围的 API 密钥，为外部环境配置 OIDC 联合身份验证，并在专用的 AI Services 账户中实现工作区级隔离。

AWS Machine Learning Blog75

使用 Amazon Bedrock AgentCore 为 Claude Desktop 添加安全的 Web Search

在 Amazon Bedrock 上运行的 Claude Desktop 如果没有 Web Search，其能力将受限于模型的知识截止时间。本文将逐步介绍如何通过 Amazon Bedrock AgentCore Gateway 将 Claude Desktop 连接到 Web Search，并通过 AWS IAM Identity Center 和 Amazon Cognito 实现基于 JWT 的入站身份验证。

Hacker News 熱門59

Show HN: Made an open-source Lego AI generator

Hi there :-) New on HN, first time posting. Past year, around December, I started experimenting with making ChatGPT and Claude generate source code in LDraw language. This LDraw is literally an "assembly" language, a low-level programming language that describes how to assemble LEGO pieces together into models, one placement instruction at a time. When executed by specific tools, like e.g. LDView, LeoCAD, Studio... these instructions become LEGO CAD models, that can be interacted with, modified, etc. Or, in other words: one LDraw source file in .mpd or .ldr format is equivalent to one LEGO CAD model. So, the idea I had was: if I manage for maybe ChatGPT or Claude to generate high-quality LDraw source files... then, they would actually be generating high-quality LEGO CAD models, right? Then, after months of iterations and trying one thing after the other... it worked!!! Long story short: using GPT-6 Astra and Opus 5.5, I've managed to create a python toolset, instructions, and docs for agents in general. Now, these can be used by them to generate LDraw models. I've packed it all as a dockerized web app for others to try and experiment, with several providers (and agents) to choose from: OpenAI, Claude and OpenRouter. I'd really appreciate feedback and comments, let's see where this goes =) Comments URL: https://news.ycombinator.com/item?id=49937916 Points: 9 # Comments: 2

Hacker News 熱門68

Claude塑造的科学

文章 URL： https://www.anthropic.com/research/claude-shaped-science 评论 URL： https://news.ycombinator.com/item?id=49933386 积分：6 评论数：2

arXiv cs.AI77

VISTA：交互式世界中的视觉推理框架

我们展示了多模态模型具备强大的推理能力，并且合适的框架能够释放其潜力，使其解决各种交互式环境中的任务。我们提出了 VISTA，这是一种赋予通用多模态模型长时域视觉能力的视觉框架。VISTA 允许模型通过视觉观察直接感知环境，并维护无损视觉记忆，以原始形式保存过往观察结果。模型可以主动检索这些观察结果，并在推理过程中重新组织其视觉输入。在 ARC-AGI-3 上，VISTA 将 Claude Opus 5.0 的相对人类操作效率得分从 40.68 提升至满分 100.00；使用该模型时，完成全部 25 个公开游戏所需的操作次数比首次参与的人类参与者少 57.4%。VISTA 的简洁设计还使其能够在仅需极少适配的情况下，自然扩展到各种视觉环境。在另外三个涵盖多种视觉游戏和谜题的基准测试中，使用相同底层模型及极简框架的 VISTA 大幅超越了基线方法。我们的结果凸显了 VISTA 作为通用视觉框架的潜力，有望推动多模态智能体在复杂视觉环境中的发展。

arXiv cs.AI80

ScholarCatalyst：一项用于检索启发新研究的论文基准

是什么造就了伟大的科学家？即使 AI 系统已开始在开放问题上取得进展，科学家在判断一项新问题需要哪一个埋藏于不断增长的研究档案中的既有思想方面，仍远远领先于它们。为研究这项能力，我们借助那些亲身了解哪些早期工作推动了其已完成项目的研究人员，并将论文作为指向其中思想的线索。通过使用可扩展作者标注的自动化流程，我们让 207 篇近期计算机科学论文的 184 位第一作者标注候选论文中哪些已经或本可以推动他们的项目，并要求每位作者提供详细的理由，由此构建了 ScholarCatalyst。我们提出了一项采用作者提供的判断结果的检索任务：给定一个初始研究问题，仅从项目开始时已经 उपलब्ध的文献中检索这些论文。尽管智能体搜索将同一个检索器作为工具调用，其表现并不优于嵌入检索（Recall@20 为 0.42，对比 0.48）。即使是基于 Claude Fable 5.1 构建的智能体——它可能在训练期间见过这些已完成的论文——也只能达到 0.51 R@20。这些结果凸显了开发新训练方案的必要性，使模型具备在广泛语料库中进行搜索的专家直觉。我们设想，ScholarCatalyst 将成为迈向科学智能体的一步：这类智能体能够接过一个尚未成形的想法，并指出它所需要的既有研究。

arXiv cs.AI87

DAYJOB：长期专业工作基准测试

专业工作通常始于一项简短请求，专业人员需要自行判断需要完成什么、哪些文档至关重要，以及请求的前提是否成立。我们推出 DAYJOB，这是一个包含 130 项任务的基准测试，其中 50 项由医疗保健领域的专业人员构建，80 项由金融领域的专业人员构建。据估计，医疗保健领域的专业人员平均需要 13.6 小时完成每项任务，金融领域则需要 16.6 小时。每项任务都是一个容器化的 Harbor 环境，并配有由专家制定的二元评判标准（每项任务的中位数分别为 47.5 条和 57.5 条）；智能体评审器会依据这些标准评估交付的文件，只有满足全部标准的尝试才能通过。在来自 13 家开发者的 30 种模型配置中，表现最强的 Claude Opus 5.5 在医疗保健和金融领域的尝试通过率分别为 24.7% 和 23.9%，而中位数配置的通过率分别为 0.6% 和 2.5%。案例研究表明，智能体会接受与记录相矛盾的前提，并将错误输入带入其他方面保持一致的分析中。我们发布全部医疗保健任务、80 项金融任务中的 50 项、评估工具链以及排行榜。

Hacker News 熱門54

Show HN: Rhun, an open-source code editor written in assembly

I found that I'm not using even 1/3 of vim/vscode features anymore. That's wht I'm building rhun - a small code editor for Linux, Windows and Apple silicon Macs. It obviously has Vim mode, a terminal, Git diffs and a panel for Claude Code or Codex sessions. The editor and pixel renderer share an x86-64 assembly core. For Apple silicon, a build-time translator converts that core to AArch64, with separate platform adapters around it. The latest release can draft commit messages using a local Ollama model or an existing Claude Code or Codex subscription. It's a solo project, MIT licensed and still early. Comments URL: https://news.ycombinator.com/item?id=49926726 Points: 7 # Comments: 2

Hacker News 熱門52

Show HN: Parrot – Open-Source Smart Meeting Recorder with Co-Pilot on Mac

Hello Hacker News I am Uygar, I am an entrepreneur from London. I run my businesses on calls everyday, customers, partners and suppliers. And i have a horrible memory i forgot things. So months ago with the help of Claude i have built Parrot. I have designed and coded for my work and what i need. An to be honest worked very well and helped me alot. Parrot records what your mac hears and everything stays in your computer if you use local models, no account, no API calls. There is something called copilot which made my life easier, you can upload your files and when parrot hears that question on the screen recommends the answer. Which i believe very handy. If you want to make it smarter and dont mind cloud you can hook up your Claude API and Deepgram account, that gives super powers to Parrot. You still keep it locally but eventually you need to make API calls. So local or cloud totally up to you. Then after months of thinking i have decided share it as open source and this is my first open source project, so please go easy on me :) No account, no membership, open source and free. Code: https://github.com/turantekin/Parrot This project made me so excited and really would like to share with you guys so i hope can help you too. Again, i am new so go easy on me and if you have any feedback i would like to hear. Thank you Uygar Comments URL: https://news.ycombinator.com/item?id=49910328 Points: 4 # Comments: 1

arXiv cs.AI76

可审计的长期记忆：在 LongMemEval-S 上测得 500 题中 479/475 的确定性检索链

我们在 LongMemEval-S 上评估了一个可审计的长期记忆系统。其检索链使用混合候选检索、cross-encoder 重排序、覆盖优先的数据包编译，以及确定性推理脚手架；LLM 仅作为可替换的最终阅读器使用。该链在 470 个可回答问题中的 468 个问题上将所有黄金会话放入候选池，并在 470 个问题中的 462 个问题上生成黄金完整的数据包。通过未固定的 CLI 别名调用 Claude Opus 阅读器，在两次 500 题运行中，在 GPT-4o 评判下分别得到 479/500 和 475/500。72 条可回答的知识更新行使用了实质性修改过的评分提示，其在官方文本下的影响尚未测量。这一组结果横跨 Chronos High 已发布的 478/500；阅读器生成、评分提示以及可能的数据版本的差异，加上系统内方差，既不能证明优越性，也不能证明等价性。在相同数据包上，grok-4.6-high 阅读器得分为 476/474，而一个最大推理努力的 agentic 变体退化到 461/465。标题结果的两次运行在八个判决翻转行上存在差异。第二个评判器在每次运行中都与标题评判器在 493/500 行（98.6%）上一致，并给两次运行均评分为 472/500；官方评判器在重新评分字节完全相同的第 1 次运行答案时，也翻转了三个判决。负对照否定了一个验证器：它修复了三个错误草稿，但破坏了十一个正确草稿。所有组件均在同一组 500 个问题上开发，没有留出评估或独立人工裁决；检索与脚手架方法来源以及从转录记录导出的审计均已保留；标题阅读器接收了额外的操作员上下文，其完整请求未被保留，且 MCP 工具可用性尚未解决。我们发布了物化数据包、脚手架、阅读器输出、评判器判决和对照，以供检查和重新评分。

arXiv cs.AI79

AdviSD：通过针对性的多轮自蒸馏学习为前沿 LLM 提供建议

一个可训练的小型建议器可以利用自然语言建议引导冻结的语言模型执行器。除了从任务奖励中学习之外，建议器还可以利用已完成交互的反馈来改进其建议。然而，一个看似合理的纠正未必会改变执行结果，但从这类纠正中学习仍可能影响建议器在其他情境下的未来决策。对于共享参数的模型，我们证明，如果这类纠正的目标相比其他纠正的目标，对有用建议的偏好程度较低，那么它们会限制学习。与从每一次纠正中学习相比，较少保留这类纠正能够提升模型最终的性能。受此启发，我们提出 Advisor Self-Distillation (AdviSD) 方法，将基于结果的强化学习与来自反馈条件化建议器副本的选择性自蒸馏相结合。反思模块提出纠正意见，建议器分别评估同一条记录下来的执行器响应在有无其所给建议时的表现，并利用两者差异的幅度来选择需要监督的决策。该方法不需要执行器的似然，也不需要额外的执行器推演。使用面向 Gemini 和 Claude 的 Qwen3-8B 建议器进行的实验表明，在 BFCL-v3 上，AdviSD 比 advisor-GRPO 高出 4.2–6.4 个百分点；在 EnvScaler 上高出 3.9–5.1 分。训练得到的建议器能够泛化到域外任务，并可迁移到不同版本的执行器和不同模型系列。AdviSD 还优于匹配数量的随机选择结果，说明其选择规则具有价值。

arXiv cs.AI80

Thinking Before Thinking: Scaling Agentic Inference Through Meta-Reasoning

As agents take on longer and more complex problems, controlling the execution becomes a task in its own right. Each step in the run brings new control choices, like which partial work to build on, whether to start fresh, or when to stop. We introduce agentic meta-reasoning, an inference-time harness that makes these choices an explicit and structured reasoning process. Workers carry out the task-level computation, while a controller consolidates what the run has established, explores next options, assesses what each option is worth under the remaining budget, and dispatches the chosen work with context drawn from persistent memory. Between decisions the controller carries only a compact account of the run rather than replaying its full history. Our baselines span production coding agents and research harnesses, together with a Direct Control Agent using the same workers and compute budget allowance. On ProgramBench, which tests long-horizon agentic capability through program reconstruction, meta-reasoning achieves 71.5% with GPT-5.5 against 58.0% for Codex; with Opus 4.8 it achieves 67.2% against 65.5% for Claude Code. On the other benchmarks, spanning abstract reasoning, multi-domain long-horizon reasoning, and proof generation, it gains between 3.6 and 4.2 points over direct control, averaged across three frontier models. It keeps improving over the tested budget ranges where direct control plateaus, though its overhead can hurt at small budgets. Artifact-graph analysis reveals more reuse of earlier work, higher coverage of correct solutions in most settings, and nonuniform gains in final selection. These results indicate that spending computation on structured control becomes more important as agents scale to longer runs.

Hacker News 熱門73

GLM-5.3 与先进网络能力的扩散

文章链接： https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities 评论链接： https://news.ycombinator.com/item?id=49897075 得分：13 评论数：5

Hacker News 熱門73

Claude 部分服务中断

文章链接： https://status.claude.com/incidents/4xvtc2gnq73l 评论链接： https://news.ycombinator.com/item?id=49893876 得分：30 评论数：9

arXiv cs.AI66

HyperMCTS: Hypergraph-Augmented MCTS for Long-Horizon LLM Agents

Long-horizon tasks require large language model (LLM) agents to coordinate decisions under constraints that span an entire solution. Monte Carlo Tree Search (MCTS) offers a promising approach to test-time scaling by exploring alternative action trajectories, but model computation and environment interaction make search costly. Efficient search therefore requires effective reuse of trajectory feedback. Standard MCTS maintains prefix-specific statistics, without explicitly accumulating outcomes for decision groups that recur across different paths. To fill this gap, we propose HyperMCTS, a training-free method that augments an ordered MCTS tree with a cross-trajectory hypergraph. Hyperedges represent groups of canonical decisions and accumulate their observed returns within the current task. Our hypergraph-guided HyperUCT selection rule aggregates evidence from overlapping hyperedges into an action prior, allowing outcomes collected under one prefix to inform selection under another while preserving execution histories in the tree. On DeepPlanning, HyperMCTS improves average planning accuracy by 2.3--7.3 percentage points over the strongest baseline for each of three backbone models. It enables Qwen3.6-27B to outperform Claude Opus 4.6 (max) on Shopping Planning, while achieving higher accuracy with fewer LLM calls and output tokens than the evaluated MCTS-based baselines. SealQA experiments further demonstrate improvements in question answering.

arXiv cs.AI78

AgentHabit：刻画智能体在日常任务中的独特行为

大语言模型（LLM）智能体可以帮助用户完成日常任务，而这些任务通常有许多合理的完成方式。即使它们的回答有用，智能体执行这些任务的方式也可能不符合用户的偏好和需求。例如，不同智能体在是否提出澄清性问题或搜索网络方面存在差异。我们提出 HABIT，这是一个包含五大类别、23 个行为维度的分类体系，由三位作者和三个 LLM 基于 17 个领域中的 408 条智能体轨迹自下而上推导而来。在留出任务上，相比现有的人类价值观和智能体行动分类体系，HABIT 能更清晰地区分模型，同时支持相近一致性的标注。基于 HABIT，我们构建了 AgentHABIT，这是一个基准测试，可根据智能体在 86 个日常任务上的轨迹刻画每个智能体的行为倾向。使用 AgentHABIT 对 18 个模型进行画像揭示了一系列鲜明的倾向。例如，大多数 GPT 和 Claude 模型会陈述其假设，并在需求冲突时提供替代方案，而 Qwen 和 Google 的模型更常不明确说明假设或需求变更。即使使用完全不同的任务集构建，这些画像仍然可被识别，表明它们反映的是一般性倾向，而非特定任务行为。提示智能体采用特定行为会使某些维度很容易发生变化，但对其他维度几乎没有影响；而在另一个模型的轨迹上进行微调只会改变模型画像的一部分，并保留其中的大部分特征。总体而言，HABIT 和 AgentHABIT 提供了一个系统性框架，用于刻画智能体在任务成功之外如何执行日常任务，并提供洞见以指导开发行为更符合用户需求的智能体。

Hacker News 熱門83

Claude Sonnet 5.5

文章 URL： https://www.anthropic.com/claude-sonnet-5-5 评论 URL： https://news.ycombinator.com/item?id=49881850 积分：71 评论数：26

Hacker News 熱門81

Claude Sonnet 5.5

文章 URL： https://www.anthropic.com/news/claude-sonnet-5-5 评论 URL： https://news.ycombinator.com/item?id=49881889 点数：19 评论数：5

Hacker News 熱門

Show HN: PaperMono, e-ink fridge magnet shopping list with mobile web page

Fridge magnet shopping list on an M5Stack PaperMono (ESP32-S3, e-ink touchscreen), synced with a phone web app over Wi-Fi. Works offline, ~2,400 lines of C++. This is a great wee device! BLE, Wi-Fi and LoRa on one board makes it flexible for way more than a shopping list, this is just what I built first. This firmware only uses Wi-Fi. Fully vibe-coded with Claude Code, I didn't hand-write this. I wanted to see how Claude would get on with building something useful for a new hardware device. As it's stuck to the fridge it's easy for the family to use, and it's already being used day to day which I'll admit is a first for a home built project like this Comments URL: https://news.ycombinator.com/item?id=49875801 Points: 12 # Comments: 3

Hacker News 熱門84

为 Claude Opus 5.5 编写提示词

文章 URL： https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5-5 评论 URL： https://news.ycombinator.com/item?id=49874728 得分：23 评论数：1

arXiv cs.AI65

Intent2Tc: Automated Intent-to-Traffic Control Translation with Language Models

Automated and highly usable Quality-of-Service (QoS) enforcement requires translating high-level service intents into deployable traffic-management policies. Although intent-based networking (IBN) has simplified policy specification, bridging the gap between business-level intents and executable network configurations remains complex, error-prone, and difficult to automate. This paper presents Intent2Tc, a closed-loop language-model-driven framework that translates business-level traffic-shaping intents into declarative sub-intents and subsequently into validated, executable Linux traffic control (tc) configurations. The framework integrates an Active Queue Management (AQM)-based digital twin (DT) semantic model, automated metadata extraction, critique-driven refinement, and Retrieval-Augmented Generation (RAG)-based knowledge reuse to improve semantic consistency and configuration reliability. We evaluate multiple open-source large language models (LLMs) and small language models (SLMs), together with Claude Sonnet-4.6, on 100 Request for Comments (RFC) 9315-compliant traffic-shaping intents. Across both translation stages, Intent2Tc achieves high semantic fidelity, configuration accuracy, and deployment readiness, with Claude Sonnet-4.6 reaching 0.98 semantic similarity, 1.0 semantic unit coverage, and 0.045 normalized edit distance. Furthermore, RAG reduces token consumption and inference latency while enabling compact models such as Phi-4-mini to approach the performance of substantially larger models. Linux tc serves as the target configuration platform, demonstrating the practical applicability of the proposed framework.

Hacker News 熱門56

Ask HN: Why do we let the AI companies dominate regulation discussion?

And I don't mean the doomerism nonsense. From rogue AI, doomerism, AGI. I mean real regulation: 1) Fair accounting of input/output tokens. Where is the weights and measures department to enforce fair counting of usage? If I pay for a gallon of gas, that is what I expect. When I pay for a dollar of LLM compute, is it going to play games with the input (or especially the output) tokens to devalue my dollar? 2) Quality. When I request a higher effort of a frontier model, and then I get downgraded to a lower tier model or with different effort levels (and I may not even know if I was downgraded!), they are effectively selling you a service you didn't ask for. "But the terms of service", great, if they feel it is such a risk to downgrade, and they get it wrong a lot, then they should discount the cost of the downgraded response. 3) Adulterating outputs. LLMs are now routinely adulterating output for tracking and accountability purposes, but is there anyone to validate these efforts are not lowering the quality of the response? There needs to be more discussion around companies adulterating their product, and holding them to account. I have several times noticed words were swapped: sometimes the word pick made no sense, when it previously didn't do this. And I wasn't looking for it or cherry picking. It stood out that badly. 4) Furthering point 3, are the responses themselves going to be seeded such that the author of that response can be identified? Has anyone asked themselves if they could deanonymize LLM responses online like yellow printer dots? 5) Quotas. Related to points 1 and 2, when I am on a metered service, what happens to requests that fail mid-response, or the response is cut midway? Why should I pay for a failed response? Maybe that is fine, but who is validating they aren't doing this at scale to save money? I feel every time we earnestly or "good faith" engage in the nonsense regulations that the major LLMaaS (LLMs as a Service) duopolists, we fail to engage in real discussions about regulating the product we are paying for . I don't see why we should believe these companies to operate in good faith post-IPO. Where is the incentive once they shut down all competition to ensure only OpenAI/Anthropic are the LLMs you can purchase from? Look to what Uber/Lyft have become over the years. Massively degrading their service and increasing nickel and dime costs. They skirted regulation then demanded the world bend and write regulations that fit their revenue goals everywhere they could (or threatened to pull out where they couldn't). And if the danger really is as bad as they say, then how can we trust them to "regulate" that threat if they would refuse regulations around basic delivery of service? Comments URL: https://news.ycombinator.com/item?id=49871747 Points: 4 # Comments: 1

Hacker News 熱門

Show HN: Reladraw – A diagram language where you decide where to place things

I love making diagrams to help understand, plan, etc. However, the options are (A) auto-placement languages like Mermaid or Graphviz (which don't let me decide how the diagram looks), or (B) software like Draw.io which are powerful but are very time consuming (and inefficient for agents to manipulate). I wanted to have the benefits of both, where you can define a diagram in a diagram language, but also retain a high degree of control over what the diagram looks like. I also wanted this to work well for humans and agents. On the Github link, there's a playground where you can try it out without installation. There's also instructions for a simple npm install and for installing a skill you can use with Claude or other agents. Comments URL: https://news.ycombinator.com/item?id=49858513 Points: 8 # Comments: 0

Hacker News 熱門47

Show HN: A Claude Code skill to analyze your chess games

Hello HN, It started as an experiment: can Claude play chess properly if it uses vision instead of PGN notation? Somehow it can. The next experiment was to see whether Claude + Stockfish could explain a game. Somehow it can too. A few sessions later, I had a system that takes my live audio notes (or text, for that matter) and a vague instruction like "analyze my last lichess game", and gives me a commented video of the game. The result is not perfect and it takes time to deliver (an hour or so), but for me it is a much more pleasant and memorable experience than clicking around Stockfish branches. It burns tokens, so make sure you have enough quota. From the session logs, the last analyzed game would have cost around $15 at API prices. The fact that it reflects on my own thinking during the game makes it interesting from a teaching point of view, so I thought it was worth sharing. Comments URL: https://news.ycombinator.com/item?id=49857528 Points: 9 # Comments: 2

Hacker News 熱門52

Show HN: Ekselio – Loveable for finance workflows (local first)

Hi everyone, I am KD - Back in my college days, I dabbled with coding, learned the basics, HTML, CSS etc. but somehow I ended up in Finance which consumed the next 20 years. Then, during covid I picked up coding again, learned react, typescript, etc - even built a rudimentary site - and then came the chatgpt moment, followed by Claude etc. So, as a side project, considering that I had spent 20 years in finance and M&A I started building Ekselio, loveable for finance workflows. Differently from other vibe coding tools, this is local first - meaning the workflows are orchestrated by the LLM based on the file schema but then the execution happens in the browser - the cool thing is that you can see the canvas, the previews of each node, and the code base (SQl) - see the Canvas tab. That said I did include a home tab, which is more of a visual, mainly because people think it's easier to relate to how folks operate, although I know finance and accounting folks usually like the workflow and full transparency - the other thing is that once you create a workflow you can save it and then next month,you can just click Run and it runs it again with no LLM tokes involved - so the infrastructure is minimal. Everything can be exported in an excel data pack with M code in case someone would want to reproduce - think Audit ready. And, I also connected to quickbooks on line so someone can directly pull in and manipulate their financials live - I also connected to Fred which carries literally 800k series in terms of macro economic data - i.e. oil prices, inflation, CPI etc and finally to stocks. I am hoping people find holes in the architecture or any ideas to make this flow a little bit better. It is in progress so if you run into bugs let me know. Considering that I am using my own API key, I have put a wall in case people overuse it but DM me if you would like to continue using, I can turn off the wall for you - hoping to keep that to a small number of users if I can and will do my best. Thank you again, appreciate you taking the time. Comments URL: https://news.ycombinator.com/item?id=49849986 Points: 4 # Comments: 0

Hacker News 熱門75

是的，Claude 可以完成 Nine Loops

文章 URL： https://www.anthropic.com/research/yes-claude-can-do-nine-loops 评论 URL： https://news.ycombinator.com/item?id=49848033 积分：9 # 评论：0

arXiv cs.AI76

普通任务压力下出现工具性监控规避行为

AI 安全领域的一个核心担忧是，当监督与完成目标发生冲突时，智能体可能会将监督视为障碍。我们研究了工具性规避，即 LLM 智能体将规避运行时监控作为完成普通任务手段的倾向。我们推出了 EvasionBench，这是一个包含 50 个多样化任务—策略对的基准，其中完成任务需要执行运行时监控所禁止的操作。智能体知道其工具调用受到监控，并且在暂停时会收到继续工作的提示。在我们的各项评估中，三次取最佳的规避尝试率最高达到 98%，成功率最高达到 88%，且不同模型之间存在显著差异。Claude Fable 5.1 的成功率较低，但经常会创造性地尝试绕过监控。总体而言，规避行为会随着测试时计算量的增加而增加：推理投入越高、使用的 token 越多，规避率就越高。轨迹显示，智能体会对被禁止的命令进行编码，将操作拆分到多次工具调用中，并不断重试，直到相关上下文从监控器的历史记录中移出。我们还发现，GPT-6 Astra 的低规避率伴随着过度拒答：在拒绝服务式提示注入下，它经常放弃原本可以解决的任务。我们的研究结果表明，即使没有明确的对抗性目标，普通任务压力也可能促使智能体采取适应性措施来规避运行时监控。因此，有效的监督必须能够抵御反复的规避尝试，因为帮助智能体解决困难任务的坚持性也可能驱使它们绕过自身的安全护栏。

arXiv cs.AI77

用于广义任务与运动规划问题的编码代理

即使具备完全可观测性和以对象为中心的状态，任务与运动规划（TAMP）问题仍然很难解决，因为离散决策与几何、运动学和动力学约束紧密耦合。广义TAMP通过利用不同问题实例之间的规律，降低在新实例上的规划成本，从而应对这一难题。然而，现有方法需要大量针对TAMP的工程设计。我们研究编码代理能否通过合成可在不同实例间泛化的程序来自动化这一过程。给定任务描述和模拟器访问权限，每个代理在固定的合成预算内开发程序，同时自行决定如何与环境交互。随后，程序被固定下来，并在未见过的实例上进行评估。我们在来自KinDER和PDDLStream的28个模拟环境中评估了Claude Code（Opus 5）和Codex（GPT-5.6 Sol和GPT-6 Astra），其中对象数量超过了原始基准测试中评估的数量。在所有程序合成方法中，我们对980个生成的程序分别在100个留出实例上进行评估，总计进行了98,000次评估回合。总体而言，我们发现编码代理在广义TAMP中表现出令人意外的有效性：在可用规划器的16个环境中，三种代理配置的平均成功率均超过了人工工程化规划器、一次性生成方法和基于LLM的广义规划基线，成功率为56%至95%，而规划器为47%。随着对象数量增加，代理生成的程序仍能保持高于规划器的成功率，同时平均每个实例所需的计算量减少了一个数量级。日志显示，代理通过交互来校准物理模型、测试边界情况并改进策略。我们公开了全部代码，包括提供给代理的完整提示词。这些发现表明，编码代理是广义TAMP的一个强有力基线。

arXiv cs.AI79

LLM Agents Can Easily Tamper With Their Own Traces

异步监控、事件调查和合规审计主要依赖代理执行轨迹来重现所发生的事情。这些分析假定 LLM 代理无法篡改自身的执行轨迹。我们发现，Claude Code、Codex、Antigravity、Open Code 和 Grok Build 等本地 LLM 代理未能执行这一边界限制。除 Muse Code 外，所有受测测试框架都允许代理在接到要求后删除自身的轨迹，且不会触发监控护栏。我们还验证了外部攻击者可以利用这一漏洞诱导轨迹删除。最后，我们表明，当代理试图提高自身奖励时，轨迹篡改行为会在前沿模型中自然出现。我们建议从业者确保通过不受代理控制的独立拦截机制记录轨迹，即使在主机完全遭到入侵的情况下，也能保持轨迹完整性。总体而言，我们的研究发现了代理基础设施中一个具体的轨迹完整性失效问题，该问题可被用于掩盖不符合预期的行为，例如密谋或蓄意破坏。

arXiv cs.AI81

提交之间：完全由 AI 编写的代码库中的流程、错误与论断可靠性

我们介绍：（i）一个新数据集，其中包含由 Claude AI 完全构建的 21,000 行 Python 工具的完整开发历史，且不含任何人类编写的代码或测试；（ii）两个代码来源追踪工具；（iii）三个分别针对指令意图、提交来源和响应可靠性的分类体系；（iv）将这些工具和分类体系应用于数据集分析。我们发现：（i）用户通过 coding agent CLI 发出的指令与 IDE-chat 指令在性质上有所不同，前者更加注重理解、规划和咨询；（ii）代码开发主要是主动进行的；（iii）14.3% 的 AI 代码生成事件包含一个后来被 AI 编写的测试套件捕获的真实错误；（iv）AI 的交互式响应中大约每 4 至 5 条就有 1 条包含一个或多个事实性错误。

arXiv cs.AI86

预填充推理通道：针对推理型 LLM 的输出前缀攻击

大型语言模型（LLM）接收并生成单一文本序列；因此，如果可以在 LLM 响应的开头添加文本，即输出前缀，那么后续所有 token 都将以此为条件。输出前缀攻击技术是一种成本低廉的黑盒提示注入方法。先前研究表明，这类攻击能够稳定地令非推理模型越狱。大多数推理模型会在助手的最终响应之前，增加一个中间的草稿推理步骤。一些 API 暴露了编辑这一推理通道的能力，攻击者可以利用某些攻击向量发起推理注入攻击。我们开展了首项系统且受控的研究，将草稿推理通道单独作为输出前缀攻击向量进行分析；同时也是首项在推理过程公开和隐藏的模型中，对仅推理、仅输出前缀以及推理加输出前缀三类攻击进行比较的研究。我们采用 3 种前缀类型 × 2 种推理注入的析因设计，在从 AdvBench 抽取的 1,800 个测试案例上，对三个 2026 年前沿模型 Gemini 3 Flash Preview、DeepSeek V4 Flash 和 Claude Haiku 4.5 发起攻击。我们发现，单独注入恶意推理基本不起作用（攻击成功率约为 0%）；但将相同的推理与一个简单的输出前缀一同注入后，在某些模型上可将攻击成功率提升至最高 99%。对于这类攻击，我们发现上下文前缀的效果优于静态前缀；此外，模型是否容易受到攻击取决于具体模型。

Hacker News 熱門77

Show HN：Critic——与编写代码的智能体一起审查代码

各位 HN 用户好，我是 Feyn 的 Shreyash。我们帮助企业利用其数据构建定制模型。今天我们发布了 Critic，这是一个变更审查平台，让你可以直接与编写代码的 AI 交流。 智能体编写了我们的大部分代码。虽然这提高了我们的生产力，但理解一项变更及其后果已经变得极其困难。随着公司采用越来越多的智能体工具，我们发现，让人们了解某个 PR 的影响和项目状态变得更加困难。于是我们构建了 Critic 来解决这个问题。 Critic 可以让 AI 智能体介绍其代码、为关键代码块添加注释，并提供相关证据（例如屏幕截图以及如何在本地运行的说明）。任何查看该变更的人都可以直接与编写它的智能体交流。你可以 在这里观看演示 。 Critic 通过 Codex 或 Claude Code 的插件运行。当你的智能体编写代码时，插件会要求它同时撰写一份叙述，介绍这项变更背后的故事，并突出显示其中的假设、决策或复杂代码。随后，你可以在 critic.run 上查看这项变更，或者让你的智能体通过内置的 MCP 获取它。Critic 中的每项功能都可以通过 MCP 使用，因此你的智能体无需离开其运行环境，就能获得完全相同的上下文。 你可以通过控制面板或 MCP 直接与负责编写代码的智能体聊天。当 Critic 收到问题时，它会分叉出编写代码时的会话，并将你的问题转发到该会话中。这样，你的主线程可以保持整洁并继续工作。 本地变更只有所有者可见。推送到 GitHub 的任何内容都会被镜像，并对在 GitHub 上拥有查看该变更权限的所有人可见。 构建 Critic 时，安全性是我们的首要考虑事项之一。AI 智能体通常会以自动模式运行，并且可以在你的计算机上执行许多操作。我们希望确保，让他人访问你的智能体并不意味着同时让他们访问你的计算机。所有 Critic 分叉出的会话都会移除写入工具。它们只能回答有关该工作区中的变更和进行中工作的相关问题。 Critic 帮助我们在几起事件进入生产环境之前发现了它们。最常见的问题是模型没有复用已有的资源，例如现有的样式、组件或函数。由于深入了解智能体的工作并向它提问变得非常容易，我们还发现了更多模型误诊问题或编写无用代码的情况。由于 Critic 的界面比 GitHub 更快捷、更适合智能体使用，我们的团队也能更快地回复反馈。 Critic 可免费使用。请在 https://critic.run 注册。如果你有任何反馈或问题，请告诉我。 评论地址： https://news.ycombinator.com/item?id=49834098 得分：5 评论数：1

Hacker News 熱門58

Show HN: Radix – Visual UI for agentic programming

Hey HN, I'm Jordan from Radix. Radix is a UI tool for programming agents. You prompt your agent to generate a workspace for a task you're working on and get an interactive widget that persists locally on disk. I built Radix because I always needed tools while I was writing code: tools to test little experiments, to play around with UI, to visualise results etc. Sometimes these would end up as python tools that read and plotted data, or separate React apps spun up just to test a single feature. Lately obviously I've been using Claude artifacts but these are quite limited. I wanted a system where it was easy to leave comments or adapt an artifact without having to give vague descriptions back to the agent, and where ideally I could actually shift away from a chat window as the main interface. I've got a pro version of Radix which I will launch soon. This current version is entirely free and is definitely a beta -- there will be rough edges! Note: There's no telemetry or data captured. Your messages run entirely through your own agent. I require a key but that's just to get an idea of how many people are using it. Everything is stored locally (the workspaces are actually just React apps which you can edit "artisanally" if you want). If you have any questions please reply here or email me hello@radix-os.com Thank you! Comments URL: https://news.ycombinator.com/item?id=49834964 Points: 5 # Comments: 4

Hacker News 熱門75

Show HN：Whiteboard（YC W26）——一款用于深思熟虑的软件设计的开源 IDE

大家好！我们是 Sid、Alex、Ketan 和 Milan。我们正在构建 Whiteboard，这是一款开源桌面应用，让人类和代理能够在一个共享工作区中共同设计软件架构。这是我们的代码仓库： https://github.com/devdotfast/whiteboard ，主页是： https://whiteboard.dev.fast/ 我们怀念与另一位开发者进行“白板会议”时的那种感觉——离开时已经对一个系统有了深入理解。因此，我们为自己构建了这个应用。Whiteboard 可以接入你已经在使用的工具，例如 Claude Code、Codex 等，并为你的代理提供一个 SDK，让它能够在应用内画布上绘图，以描述自己的工作。我们最初基于 HTML artifacts 构建了一个 MVP，后来随着遇到各种限制，我们开始重新思考这个应用： 1. 构建在 CodeOSS 之上：我们发现，在纯 HTML 工具中，很难将规范或图表与代码关联起来。在 Whiteboard 中，当你点击序列图、实体关系图或代理轨迹中的一段引文等可视化内容时，可以直接跳转到底层代码。在浏览代码时，你可以开箱即用地获得 VSCode 的快捷键绑定和 LSP 支持。我们发现这一点尤其有价值，因为许多权衡往往只有在首次实现之后才会显现出来（参见：slop） 2. 语义差异查看器：我们用 Rust 编写了一个支持语义分析和 AST 的差异查看器，因此你只需查看与你相关的代码变更 [1]。我们设置了一些合理的默认行为：较大的新增函数会以伪代码形式进行概述，单元测试和大段文档变更等内容则会被折叠或隐藏。这些设置都可以通过基于 WASM 的插件系统进行自定义。 3. 决策日志：我们发现，很难推理代理自主做出了一组什么样的决策。因此，我们构建了一些工具，让代理能够查询自己的轨迹并将其链接到 Whiteboard，这样你就能了解你设定的需求是如何被实现的，以及你的代理自主做出了哪些决策。 这里有一段简短的演示视频，可以进一步了解相关内容： https://youtu.be/ChPn3ftULWE Salesforce 和 Modal 等公司的员工如今正在使用 Whiteboard，作为架构或规范级变更的审查工具——实际上，任何他们希望参与其中的变更都可以使用： 1. 审查你自己的编码代理所完成的工作：由于 Whiteboard 让审查大量代码变得更加容易，人们通常会让 AI 代理创建一个原型以及相应的 Whiteboard 会话，然后据此迭代设计。 2. 审查他人的变更：我们发现，Whiteboard 与 Greptile 等工具结合使用时尤其有帮助。例如，你可以对小型变更运行自动化代码审查，而对于需要人工判断的变更，则升级为 Whiteboard 会话。 我们为什么构建它：我们是四个大学时期的好友，在代理式编码成为行业标准之前不久，辞去了技术负责人的工作。在为之前的一个想法迭代 MVP 时，我们努力想在享受代理式编码带来的全部效率提升的同时，维持一个易于理解的代码库。随着越来越多的 PR 在我们并未理解的情况下被合并，我们感觉一种“认知债务”开始逐渐渗入，直到我们甚至难以继续为这个系统做贡献 [2]。 我们将以 MIT 许可证发布我们的桌面应用。欢迎深入了解，也欢迎贡献代码！最终，我们会向企业收取托管版 Web 服务的费用，用于管理 Whiteboard 会话的创建，并提供轨迹存储和多人审查等功能。一切始终都会支持自行托管。 感谢阅读，希望你能试用一下！我们非常期待听到你的反馈，也希望向各位专家学习。 再次附上项目链接： https://github.com/devdotfast/whiteboard ；你可以在这里安装适用于 MacOS 和 Linux 的版本： https://install.dev.fast [1] diffs 库： https://github.com/devdotfast/diffr [2] “认知债务”这一术语的出处归于 https://www.geoffreylitt.com/2026/07/02/understanding-is-the... 评论链接： https://news.ycombinator.com/item?id=49833867 得分：16 评论数：0

arXiv cs.AI85

FDE-Bench：评估用于部署环境配置的 LLM 智能体

部署要求智能体将应用程序代码转化为一个正在运行的系统，使其服务能够连接、就绪并保持可观测。FDE-Bench 通过 136 个部署配置任务评估这一能力，涵盖 Docker 镜像、多服务 Compose 栈和 Kubernetes，并包括全新构建模式以及诊断与修复模式。智能体提交声明式制品，这些制品会在一个干净环境中被收集、重新构建并重新部署。四个带门控的二元检查层衡量构建、就绪性、行为以及对部署规范的符合性，使用程序化检查而不依赖 LLM 评判器。一个四臂发布门槛要求存在可解决任务的参考方案，并拒绝那些可由无操作、规范转录或通用桩提交解决的任务。发布的检查注释揭示了 2,145 项检查与其规范之间的关联，包括七个已记录的缺口。另有三种对抗性策略测试评分信号中的捷径；它们未能解决其覆盖的 135 个任务中的任何一个，而一个空洞的健康探针通过了就绪性检查，并暴露出下游检查的必要性。在 136 个任务的评估网格上，来自四家提供商的七个语言模型使用相同的四工具脚手架，解决了 52.9% 至 75.0% 的任务。三个零智能下限没有解决任何任务，平均 Deployment Score 最高仅为 0.44。就绪性是最大的失败阶段，在 313 个未解决回合中占 110 个。修复任务组的平均解决率比互不重叠的全新构建任务组高 30.7 个百分点，且每个模型都呈现正差距；有十个任务所有七个模型都无法解决。在一个 25 个任务的案例研究中，一名执业工程师指导 Claude-Sonnet-5 解决了 92% 的任务，而自主基线为 72%。FDE-Bench 将部署成功与失败关联到可检查和可重放的制品。

Hacker News 熱門67

Claude Code 仅在启用遥测时读取 AGENTS.md

文章网址： https://blog.szypowi.cz/p/claude-code-reads-agents.md-only-when-telemetry-is-on/ 评论网址： https://news.ycombinator.com/item?id=49814947 积分：10 # 评论数：1

arXiv cs.AI82

CliffCompaction：面向长时程编程智能体的高成本效益压缩

智能体经常处理需要数百万个 token 上下文的复杂问题，而受限的上下文窗口使得跨会话压缩成为必要。我们开发了 CliffCompaction，这是一种自动压缩技术，在有界上下文下最多可降低 50% 的成本，同时在 Terminal-Bench 上保持或提升性能，并为测试时扩展带来新的效率水平，在 KernelBench 上取得最先进的结果。CliffCompaction 在每次运行中节省的成本，使测试时扩展的性能—成本权衡更加高效：在 Terminal-Bench 上，仅低于两次完整上下文运行的成本，就能额外提升超过 10 个百分点。在并行测试时扩展下，CliffCompaction 使 Kimi K2.6 达到与 Opus 4.7 相当的表现，并以更低成本超越 Opus 4.6 和 GPT-5.3 Codex。CliffCompaction 高效的关键在于，它只截断或删除内容，从不对内容进行改述或重写，从而保持压缩信息的忠实性。我们绝不会对压缩结果再次压缩——每次处理仅针对原始内容，之前的压缩输出会被丢弃，以防止上下文漂移不断累积。这些特性支持跨越超过一百万个 token 的会话持续学习：在 KernelBench 上，经过 200 步后，CliffCompaction 实现了 CUDA 内核 2.23 倍的加速，经过 400 步后达到 3.58 倍，尽管它是一种通用压缩技术，仍超越了专用搜索算法和经过训练的智能体。我们开源了一个与脚手架无关的 CliffCompaction API 代理实现，可与 Claude Code、Codex 及其他运行框架配合使用。

Hacker News 熱門78

Training a model to identify AI-generated web content from structure alone

Hey HN! We’re Vincent and Jochen from Sitefire ( https://sitefire.ai ). We have been working together for years, with backgrounds in RL/optimization at Stanford and software engineering from Technical University Munich (TUM). With Sitefire (YC W26), we help marketing teams get recommended by AI Search (ChatGPT, Google AI Overviews, AI Mode, Claude, etc.). Our software monitors prompts, sees which web pages get cited, and uses these insights to help marketing teams take action, e.g. create YouTube videos or write the right blog posts. This means we have a commercial stake in AI-generated web content. And for now, high-information, AI-generated content works great to get cited and recommended in AI Search. But after talking to hundreds of marketing teams, it became clear that everyone despises AI-generated content (“AI slop”). And yet, everyone still wants to leverage AI to create content. So we asked ourselves: what characterizes AI slop? Can we train a model to identify it from human-generated web pages? Researchers from the University of Maryland and Google DeepMind already asked this question for fiction. Their paper StoryScope (Russell et al., 2026) showed that you can tell AI-written stories from human ones by their structure alone, without looking at the words. We ported their pipeline to commercial web pages. Using the Wayback Machine, we collected 2,250 blog posts from 268 B2B company websites that were written before ChatGPT existed. For each blog post, five AI models (GPT-5.4, Claude Sonnet 4.6, Gemini 3 Flash, DeepSeek V3.2, Kimi K2.5) wrote their own version. Instead of looking at the words, we looked at how each post is built. We had an AI model answer 214 questions about every post, e.g. how hard it pushes its own product, whether it backs up its claims with sources, or whether it quotes a named expert. Then we trained a classifier on these answers. On blog posts it had never seen before, our classifier told AI-generated and human posts apart with 98% accuracy, getting only 19 of 1,740 wrong. Why does it work so well? Because all five AI models write in a similar shape. Mapping every AI model’s values for these features, we see they cluster together, while the human values sit apart and spread out much more. Of the 1% most unique blog posts in our data set, 149 are human, only 4 are AI. So what characterizes AI slop? It tells you the same thing three times. The title already promises what you'll get ("How to Cut Onboarding Time in Half"), the intro lays out what's coming, and the ending says it all again. 77% of the AI posts end by repeating their main point, compared to only 12% of the human posts. We call it the tidy, self-announcing blog post. Still, each AI model has its own accent. We trained a second classifier to tell which of the five AI models wrote a post, or whether a human did. It picks the right author 79% of the time, where random guessing (1 in 6) would get 17%. Almost all of its mistakes are mix-ups between the AI models, not between human and AI. The cool thing about structural features is that you can't simply reword your way out of it. We had each AI model rewrite its own posts until, on average, 73% of their original 13-word sequences were gone, and the AI slop classifier still worked just as well. We're building this into Sitefire: our agents get a structural understanding of text, so the posts they write go deeper and vary the way human writing does. There's a lot we haven't tested yet, like the myriad of humanizer tools, human rewriting, restructuring a post, or prompting an AI model to explicitly avoid these habits. And our human posts are mostly from 2020 to 2022, while the AI posts were generated in August 2026. Structure can't really tell when a human post was written, but it's still not a same-year comparison. We published the study with all the figures on arXiv: https://arxiv.org/abs/2609.15369 . The code is on GitHub: https://github.com/pulse-energy-eu/slopshape We're pretty sure your own blog isn't AI slop, is it? We built a checker that runs one of your posts through the ten features from the paper, so you can see for yourself (the full report asks for a work email): https://sitefire.ai/slop-checker . Think you can tell AI slop from human writing? We also made a little game to see if you can keep up with our model, which gets all five rounds right: https://sitefire.ai/spot-the-slop . Comments URL: https://news.ycombinator.com/item?id=49800566 Points: 5 # Comments: 0

Hacker News 熱門72

Show HN：AI·rete·RAG——Rete 规则引擎作出决策，RAG 解释原因

大家好，我构建了 ai·rete·rag，因为我不断看到一些团队让 LLM 负责那些需要可审计的决策（贷款、欺诈、临床分诊），然后事后再加上“护栏”。 这个项目改为让两者串联运行： 1. 纯 Python 实现的 Rete 引擎根据事实评估 YAML 规则。判定结果只来自这里。相同的事实，每次都得到相同的判定，并通过基于显著性的冲突解决机制处理冲突。 2. RAG 从你自己的政策文档中检索段落，然后由 LLM 撰写通俗易懂的决策解释，引用这些段落。它无法改变判定结果。 有几件事的发展超出了我的预期： - 规则是一个图，而不是扁平列表：支持嵌套的 all/any/not，并且规则可以断言供其他规则使用的事实（前向链接）。决策追踪会展示因果链。 - 审计模式会记录评估过的每一条规则，包括那些未触发的规则，逐条件记录，并保存规则集快照以便重放。 - 规则可以引导检索（触发的规则会缩小要搜索的文档范围），检索到的文本也可以转换为供引擎使用的事实。 - 非技术人员可以在可视化编辑器中构建规则，也可以粘贴政策文档，让 LLM 起草带引用的规则。未经审核，草稿绝不会被保存。工程师仍然可以使用 YAML。 落地页提供无需注册的在线演示（涵盖 8 个演示领域：贷款、欺诈、临床、保险、法律、运营、电子商务、区块链）。此外还有一个 MCP 服务器，因此 Claude 和其他代理可以将 /decide 作为工具调用：`uvx ai-rete-rag-mcp`。 坦率地说：这是一个提供免费套餐的托管产品。MCP 客户端采用 MIT 许可证并已开源（github.com/zaharajabeen13-create/ai-rete-rag-mcp）；引擎和平台目前并未开源。 我尤其想听听曾经需要向监管机构或审计人员解释自动化决策的人分享经验：他们实际要求了什么？ 评论链接： https://news.ycombinator.com/item?id=49803683 得分：6 评论数：0

Hacker News 熱門73

Claude Opus 5.5

https://github.com/anthropics/ClaudeForFoundationModels/comm... 评论网址： https://news.ycombinator.com/item?id=49803863 分数：246 评论数：175

Hacker News 熱門74

告诉 HN：Claude Code 未经询问就替我接受并签署了一份合同

我让它进一步推进一个项目。该项目有一个外部依赖项，而相关合同（我还没读过）就在我的 Gmail 里。它下载了 PDF 合同，在我的电脑上找到了一个保存的签名 PNG，将其放在合同中的正确位置，并准备在我介入时发送出去。 评论网址： https://news.ycombinator.com/item?id=49798257 积分：9 评论数：6

arXiv cs.AI82

OSWorld-Pro：面向计算机使用代理的基于过程的评估

计算机使用代理（CUAs）的评估通常仅限于对其创建的最终交付成果（数百个步骤之后）进行评估，并采用功能验证器进行判定，如 OSWorld 所示。然而，这种对最终状态性能的评估无法透明地揭示代理在各种任务中失败的方式和原因，从而掩盖了对后续改进至关重要的洞见。例如，在键盘输入过程中出错的代理，与无法在图形化 UI 上精确执行基于点击的输入的代理，需要采用不同的缓解策略。我们推出 OSWorld-Pro：包含超过300项任务和超过2800个子目标的数据集，基于超过67,000条人工标注，实现对 CUAs 的程序化评估。我们使用与人类判断高度一致的 LLM-Judges 来评估 OSWorld-Pro 子目标的完成情况，从而揭示模型在一系列具有顺序依赖关系的子目标中所取得的进展。研究结果表明，即使是最先进的 LLM，OSWorld-Pro 也极具挑战性；其中表现最佳的模型（如 Claude Opus 5）仅达到75.7%，而在 OSWorld 上则为83.4%。此外，我们还识别出各种模型在过程层面的关键失败模式（例如与子目标无关的操作和基于点击的错误），为提升 CUAs 的性能和效率提供了有益洞见。

arXiv cs.AI88

你也一样，布鲁图？个人 AI 代理中的经济错位

个人 AI 代理会在高风险经济场景中代表人们提出建议并采取行动，例如购买机票、选择健康保险或挑选研究生项目。代理可以访问用户的个人背景信息，例如电子邮箱收件箱和包含个人属性的结构化档案，其目的是为用户作出最优的个性化决策。我们发现，仅仅提供这些个人背景信息，代理就会根据推断出的财富状况引导推荐，即使没有明确指示它这样做。在针对三类经济决策（机票、健康保险和研究生项目）开展的 325K 次实验中，我们对 13 个代理进行了测试，发现其中 8 个模型在请求完全相同的情况下，会系统性地为更富有的用户选择更昂贵的选项。即使这种引导直接违背用户明确表达的目标，它仍会持续存在：当被明确要求寻找最便宜的选项时，一些代理仍会依据其推断出的财富状况采取行动。当财富状况是从与任务无关的电子邮件等环境数据中推断出来时，也会出现这种现象。在实施阻止特定属性的隐私控制后，这种现象依然存在：阻止财务属性在很大程度上可以消除差异，但阻止其他属性并不会改变差异，甚至会使保险领域的差异扩大最多 40%，因为代理会依赖剩余信号来推断财富状况。规模更大、能力更强的模型也没有表现得更好；Claude Opus 4.8 的影响最大。我们将这种错位称为“对抗性委托”：正是让个人 AI 代理变得有用的条件——能够访问个人信息——使其得以采取不利于用户利益的行动。

arXiv cs.AI78

基于 LLM 的主动推理智能体可解释性失效的触发因素与诊断

LLM 解释器正越来越多地作为运行时监督附加到自主智能体上，操作员阅读的是对智能体信念和行动的生成式说明，而非其内部状态。我们审计的是该说明本身：将一个跟踪德国电网需求并调整发电量的 Active Inference (AIF) 智能体与基于三个后端（GPT-4o、Claude-3-Opus、Gemini）的 LLM 解释器配对，并使用三种黑盒触发因素探测这一组合。每步将观测流篡改 600 MW，会使智能体的后验偏移 490 MW，约为电网容量的 0.9%。在注入期间生成的 30 条解释中，没有任何一条在既定评估准则下标记出异常，而且每条都流畅地叙述了被篡改后的信念。在智能体采取客观错误行动的时间步上，三个解释器都会在 80–95% 的情况下产生迎合式合理化解释（每个后端 n = 20）。观测元数据字段中的攻击者可控文本会引导解释器，不同提供商的易受影响程度不同，且三个后端上的数据外泄均成功。我们为每种失效提出了缓解措施，但未对其进行评估。在我们观察到的每一种失效中，解释都是流畅但错误的。此外，解释器架构中没有任何机制会在操作员依据解释采取行动之前检查该解释是否为真。因此，对解释器的测试应纳入任何智能体部署审计之中。

Hacker News 熱門70

Show HN: Foremerge – Catch Intent Conflicts Between Parallel Coding Agents

At, GPTree, we run several coding agents across our team on one repo using parallel worktrees. Apart from wasted time reviewing and fixing conflicts at PR time, the failures that hurt the most are when multiple plans or tickets cause architecture changes that cannot both be true. Ex. one agent replaces a class while another one is in the process of extending it. Git only notices if the resulting patches happen to touch the same lines and the review only catches it if they are familiar with both tickets. Foremerge is a local "git like" coordination layer that sits above git (ie. does not interact with or change the way git and worktrees function), Before editing each agent publishes an intent and the scopes it will change, with the operation it plans to complete on each one. foremerge intent publish --agent "$A" \ --summary "Replace PaymentService with StripePaymentService" \ --scope symbol:PaymentService=replace foremerge intent publish --agent "$B" \ --summary "Add PayPal support to PaymentService" \ --scope symbol:PaymentService=extend The publish by the 2nd agent returns a HIGH destructive_vs_additive finding before writing any code. Agents keep their own worktrees and the shared state is one SQLite file in gits common direectory. No hooks, no merge drivers, nothing rewrites your history. It ships as one Rust binary with a CLI and MCP server with 18 tools and `foremerge setup all` wires it into Claude Code, Codex and Cursor. Because the protocol has nothing provider specific, a Claude agent and a Codex agent coordinate through the same store. Before any work is accepted, Foremerge runs a named check that you configured against the exact git state of the change. An agent that says tests pass is recorded but it dosnt satisfy the acceptance gate without running the check itself. Detection is deterministic, no judge model reading your code. HIGH conflicts are only asserted for declared operations, ie. matches inferred from prose cap out below high. Claims are advisory leases, not locks so two agents can still hold the same scope without deadlock. The open source version is single matching and so not a distributed consensus. We have tested this up to 98 parallel agents all working on the same repo with zero conflicts (was supposed to be 100 but 2 agents failed to run due to resource limitations) I replayed 76 intents on my own agents from a build last week in the order they happened. The sample had exactly 1 conflict (which was flagged) and the review found a blind spot where one agent claimed scope by class name and the other claimed it by an internal method. We are working on fixing that for the next release. Setup is a 30s install by pasting the quickstart instructions from the readme.md into your agent or manually: `curl -fsSL https://foremerge.com/install.sh | sh` or `cargo install --locked foremerge`, then `foremerge init && foremerge setup all` in a repo. Apache-2.0. The feedback I want most is which conflicts between your agents plans would you actually want flagged and which would you tollerate as noise? Repo here: https://github.com/naw103/foremerge Website: https://foremerge.com More information on the problems this solves: https://foremerge.com/blog/ Comments URL: https://news.ycombinator.com/item?id=49789356 Points: 4 # Comments: 0

arXiv cs.AI82

Designer-RSI：从用户流量中演化面向智能体图形设计的程序性记忆

专业图形设计是一项长时程智能体任务，结构化、可编辑的作品通过许多相互依赖的操作逐步形成，但其结果不存在可靠的程序化判定标准。我们提出了一种持续适应框架：冻结的前沿模型通过 230 多种工具操作专业设计软件，同时，外部的自然语言技能程序性记忆会从经验中积累并完善可复用的设计流程。该记忆通过为反复出现但尚未覆盖的子任务获取流程来拓展规模，并通过根据既有流程自身成功和失败的执行结果进行修订来加深内容；与此同时，配对回放门控机制只允许那些能够修复失败且不会使已观察到的成功结果退化的变更进入记忆。在 1,406 份真实用户需求和 1,869 条自动评分轨迹上进行五轮迭代，在不更新模型权重且无需人工标注的情况下，技能库从基于文档生成的 76 项技能增长至 139 项，并使 Claude-Sonnet-4 在 GenEval2 上的执行成功率从 72.7% 提升至 99.3%（生成质量提高 11.99 个百分点）；在 Claude-Sonnet-4 和 Claude-Opus-4.6 上，针对四个专业设计基准，相比无技能智能体的胜率分别达到 61.8% 和 67.6%。我们进一步表明，这两种机制结合使用时效果更佳：在来自用户流量基准的 200 份留出需求上，单独进行拓展或加深时，相比无技能智能体的胜率分别为 49.4% 和 48.6%，而二者结合后达到 58.5%（p = 0.025）。程序性记忆为在反馈嘈杂且无法验证的环境下实现智能体持续适应提供了一条切实可行的路径。

Hacker News 熱門66

Show HN：CUA-S1 – 用于计算机使用的 System One Model

Hello HN！我们是来自 Cua 的 Dillon 和 Francesco。 我们一直在想，有多少计算机使用任务实际上需要一个完整的通用 LLM（例如 gpt-6-astra、claude-opus-5 等）来思考它们的所有决策和步骤。有些任务需要思考计划、探索不同路径、从失败中恢复。另一些任务则是做出局部决策的问题，比如这个值应该填进这个框，或者我是否应该勾选这个复选框，或者这个元素应该被忽略。 我们想知道，如果训练一个小模型只做这类决策，能走多远。 我们的灵感来自 Typesafe 的 Jev 及其 System One Model 的框架。这是在致敬 Daniel Kahneman 所描述的二分法：快速、自动、直觉式思考（系统 1）与缓慢、分析式思考（系统 2）。 对我们来说，有趣的问题是：如果给模型一个包含当前上下文的接口，以及一组可能的选择，并要求它为每个选择返回一个概率，会发生什么？这类模型不像大多数 LLM 那样逐 token 生成输出，而是对你给它的选项打分；你可以检查、信任并使用这些分数来驱动你的应用行为。 CUA-S1 是我们面向计算机使用场景中狭窄、专用决策模型的答案。我们的第一个发布版本是 CUA-S1-FORMS。我们基于 jevlike 中的想法和代码构建了它，然后训练了第二个模型，专门处理表单交互。它有 706k 个参数，原始 checkpoint 为 2.8 MB。 第一次训练迭代在合成数据上耗时不到 30 分钟。给定一组从文档中提取的结构化元素和值，它会为每个元素预测是否使用给定值、CHECK、CLICK 或 SKIP。它不会为文本字段预测新值，也不会考虑截图。元素决策会一起打分，你的代码可以对操作进行排序，而 Cua Driver 会逐一执行它们。 该专用模型与托管版 Jev 在我们的表单任务上的首次评估结果： - 对整个决策集：99.7% 正确 vs 83.6%。 - 对需要执行操作的步骤子集：100% 正确 vs 96%。 - 对只是保持已填写字段不变的步骤子集：100% 正确 vs 74%。 该专用模型是专门针对这个任务和约定（对已经填好的框直接按 skip）训练的，而托管版 Jev 并未为此进行微调，因此这是一次关于限定范围专业化的实验。 我们测得在本地为一个表单打分需要 7-9 ms，而调用托管版 Jev 每次需要 260-280 ms（包括网络延迟），不过这些样本测量的是不同内容，并不是端到端的表单完成时间。 我们在这里感兴趣的是介于脆弱脚本和通用 agent 循环之间的空间。表单字段的内容和布局变化足够大，以至于脚本会变得难以维护，但可用决策集合可以保持狭窄且边界清晰。我们想探索这样一种可能性：通用 agent 遇到新颖情况，并把已被充分理解的决策交给这样的专用模型处理。 这是我们正在研究的方向。当前发布版本仅用于表单。我们已将合成数据生成、训练、评估和 Driver 集成以 MIT 许可证开源在 libs/cua-s1 下。 欢迎评论！尤其是如果你正在构建计算机使用 agent，并且遇到过某种反复出现的决策：它太多变以至于难以脚本化，但又太狭窄以至于不值得调用另一个 LLM。 评论 URL： https://news.ycombinator.com/item?id=49767564 点数：4 评论数：0

Hacker News 熱門

Show HN: Seal – Letters and passwords that open for your family after you die

I'm a CS undergrad who has developed an interest in hardware keys ever since I used them for a school project a few years ago. I made a little chat app that only allows users to chat if they registered a key, nothing else. The backstory of how I came to work on Seal (this project): was that I was generating ideas with Claude to come up with an idea that only hardware keys could make into software. I was thinking about physical house keys and what you could do with those, but in software. I was already making iOS apps with Claude, and I came up with this idea and it sounded good. I had already worked on a few projects involving hardware keys and I already loved them. Today I submitted my latest app, Seal, for review in the app store. Seal is an iOS app for iPhone and iPad that lets you write an Envelope (an encrypted bundle of passwords, photos, voice memos, etc...) for all of your loved ones when you're gone. You can register hardware keys and assign identities for each person. Then you can create Envelopes and assign them to the key that you want to distribute. They plug in the key to their phone and are ready to receive an Envelope from you. If you don't log into the app for months, one of your designated key holders can start the process of opening the Sealed Envelope. You, as the writer of the letters, get weeks of warnings to check in at least once before the letters open. After 90 days of countdown, 21 days of warnings, and 14 days of grace, the envelope opens for each person they met IRL and exchange keys with. No one else can read the envelope or see the photos (even the key holders, Apple, or me) unless that envelope was designated for that person. What's different from a password manager like 1Password: emergency access there hands one person your whole vault through the company's server, so it needs that company to still exist and be honest on the day it matters. Seal gives each person their own envelope nobody else can open, and there's no company in the middle. There's also no server at all besides a free Cloudflare static site. The encrypted bundles of passwords and photos are in Cloud Kit. Sign-in/identity is a passkey Face ID/Touch ID or a Hardware Key. It's all Crypto Kit, so there's no dependencies required. Currently, it's just me and Claude working on this. It hasn't been independently audited by anyone, and it's iPhone only for now. Please do not put a live seed phrase in it yet. If anyone has advice for me as the lead developer for this project, I'm all ears! You can help contribute on the github source code repository page if you'd like. My mom likes this app especially because she's into crypto and she needed a way to pass down or give away the seed phrases for her wallets to all 3 kids without any question about who gets what and when. In the end, there's no confusion or hurt feelings after the fact with just crypto. Thank you for reading and let me know if you have any questions or concerns for me or the project! Source: https://github.com/jasonepage/Seal Site with limits and objections pages: https://sealmessenger.com Free on TestFlight while the App Store version is in review: https://testflight.apple.com/join/cYp9JRCG Comments URL: https://news.ycombinator.com/item?id=49763311 Points: 4 # Comments: 0

arXiv cs.AI78

SoL-Pi：为高效智能体运行框架递归扩展自动研究循环

随着编码智能体从受监督的代码补全转向无人值守、全天候探索，其工作也从孤立的预测扩展为包含推理、工具使用和反馈的长轨迹。因此，token 效率对于扩展递归式自我改进变得十分重要。我们在运行框架层面采用受 RSI 启发的方法，在数量不断增加且日益多样化的环境中扩展自动研究循环，以进行框架滚动执行。在这一规模下，该过程能够产生可复用的改进，并迁移到其开发环境之外，从而推动自动化框架发现走向生产级成果。四种机制经筛选后保留下来，共同构成 SoL-Pi，涵盖动作执行、上下文压缩、观测处理和委托式阅读。在包含 51 项任务的 EdgeBench 评测中，SoL-Pi 在 GPT-5.6 Sol 和 Opus 5 上取得了与 Pi 相当的性能，同时将记录的 token 流量降低 44.7%–49.0%，并将 API 成本降低约三分之一。换言之，与 Codex 和 Claude Code 的原生框架相比，预计每小时可节省 \$8.75–\$13.50；与 Pi 相比，每小时可节省 \$4.36–\$5.71。

arXiv cs.AI81

推理引擎指纹识别攻击具有可行性：探索模型驱动的环境发现、利用与逃逸

前沿 AI 模型正迅速获得利用复杂软件中漏洞的能力。这种风险并非理论上的，OpenAI 和 Anthropic 的前沿模型近期实施的沙箱逃逸便证明了这一点。关于如何对推理栈组件进行沙箱隔离的讨论，通常侧重于推理引擎本身以外的组件（例如网络代理或代码执行环境）。然而，推理引擎对于不对齐模型而言是一个颇具吸引力的目标。例如，如果模型仅通过生成经过特殊构造的输出令牌，就能触发该引擎中的漏洞利用，那么它便可以在引擎中发起一条多步骤、直达裸机的利用链，而不依赖推理栈其他组件中的漏洞，也不需要外部提供的恶意构造输入令牌。 在本文中，我们展示了不对齐模型能够执行推理引擎指纹识别，以确定执行该模型的具体引擎（例如 vLLM、SGLang）。完成引擎指纹识别后，模型便可以利用特定于引擎的漏洞，仅使用经过精心选择的输出令牌来控制该引擎。我们给出了五种流行引擎中的模型指纹具体示例，并演示了现实的智能体运行框架如何让模型利用这些指纹识别本地引擎。我们还描述了一条概念验证性质的、直达裸机的利用链，该利用链源自一个已完成指纹识别（并随后遭到攻陷）的推理引擎。最后，我们讨论了几种可以改进推理引擎、从而增加指纹识别攻击难度的方法。

Hacker News 熱門78

Launch HN: Skillsync (YC W26) – AI chat sessions made portable across agents

Hey HN, we're Nars & Nishant, founders of Skillsync ( https://skillsync.com ) Skillsync lets you move your AI chats across every coding agent. Most of our work exists as conversations, which are currently scattered across our agents. Though stored locally, these conversations use different formats. This is annoying because you cannot simply switch between agents without starting over. We get locked into a single provider and their agent as we invest in skills and memories over time. Skillsync acts as a universal converter. It moves the entire session, including all the messages, reasoning and tool calls so you can pick up right where you left off. Skillsync collects all your sessions in one place and makes them searchable. It breaks down what each session is carrying, including which loaded skills the agent is actually using, making stale context easy to spot. You can also create shared workspaces to sync sessions across your team. You can build your own closed loop systems. Everything runs locally except when you share to workspaces. The core is an open-source Rust engine called txcript ( https://github.com/skillsynchq/txcript ). It translates a session from one agent's on-disk format into another's, mapping conversation, reasoning, and tool history. Think ffmpeg or pandoc, but for agent sessions. On top of that engine is a local-first desktop app. Your agent sessions are normally scattered across different tools' folders in formats you'd never read by hand; the app surfaces them in one place with a UI that makes them actually readable, the conversation, the reasoning, and the tool calls, so you can revisit what happened, move a session into another agent, or share it with a teammate. Skills and memory are stored as portable, human-readable markdown you own, and exposed to any agent over MCP for search and selective retrieval. Sessions and translation run locally on your machine. The one thing that leaves is what you explicitly share into a team workspace. It’s been surprising to us to see how locked-in people already are without realizing it. You don't notice you're trapped in one agent or harness until you try to leave, and by then you've got months of sessions stuck in a format only that tool can read. The lock-in is invisible right up until it's expensive. Once sessions are portable, they stop being disposable logs and become something you can actually build on, spotting patterns in how you work, handing a session off to someone else, letting a non-technical teammate pick up where an engineer left off. The session turns out to be the unit of collaboration, and right now it's being thrown away. Before Skillsync, Nars and I built an open-source payment orchestrator that let merchants route across many processors instead of getting locked into one (30k+ GitHub stars). Fighting vendor lock-in by making incompatible systems interoperate was the whole job. We came into YC with a different idea, but the more we lived inside coding agents the more we saw the same problem from the other side: your context is locked into whatever agent you start with, because every agent stores sessions differently and none of it moves. Some of the things we’re seeing people do with Skillsync are: - Moving a session between Claude Code, Codex, and Cursor mid-task, including when you hit a usage limit on one and want to keep going on another. - Thinking through a problem in a browser Claude chat, then handing the whole thread to Claude Code or Codex to build. - Sharing research sessions with a team. Everyone uploads their sessions, so teammates and their agents can build on each other's work instead of repeating it. It's available as a Mac app, CLI and MCP. Here's the demo: https://youtu.be/7hVhSnSKGl8 . Would love to hear your feedback and answer questions! Comments URL: https://news.ycombinator.com/item?id=49743049 Points: 10 # Comments: 2

Hacker News 熱門78

Launch HN: Skillsync (YC W26) – AI chat sessions made portable across agents

Hey HN, we're Nars & Nishant, founders of Skillsync ( https://skillsync.com ) Skillsync lets you move your AI chats across every coding agent. Most of our work exists as conversations, which are currently scattered across our agents. Though stored locally, these conversations use different formats. This is annoying because you cannot simply switch between agents without starting over. We get locked into a single provider and their agent as we invest in skills and memories over time. Skillsync acts as a universal converter. It moves the entire session, including all the messages, reasoning and tool calls so you can pick up right where you left off. Skillsync collects all your sessions in one place and makes them searchable. It breaks down what each session is carrying, including which loaded skills the agent is actually using, making stale context easy to spot. You can also create shared workspaces to sync sessions across your team. You can build your own closed loop systems. Everything runs locally except when you share to workspaces. The core is an open-source Rust engine called txcript ( https://github.com/skillsynchq/txcript ). It translates a session from one agent's on-disk format into another's, mapping conversation, reasoning, and tool history. Think ffmpeg or pandoc, but for agent sessions. On top of that engine is a local-first desktop app. Your agent sessions are normally scattered across different tools' folders in formats you'd never read by hand; the app surfaces them in one place with a UI that makes them actually readable, the conversation, the reasoning, and the tool calls, so you can revisit what happened, move a session into another agent, or share it with a teammate. Skills and memory are stored as portable, human-readable markdown you own, and exposed to any agent over MCP for search and selective retrieval. Sessions and translation run locally on your machine. The one thing that leaves is what you explicitly share into a team workspace. It’s been surprising to us to see how locked-in people already are without realizing it. You don't notice you're trapped in one agent or harness until you try to leave, and by then you've got months of sessions stuck in a format only that tool can read. The lock-in is invisible right up until it's expensive. Once sessions are portable, they stop being disposable logs and become something you can actually build on, spotting patterns in how you work, handing a session off to someone else, letting a non-technical teammate pick up where an engineer left off. The session turns out to be the unit of collaboration, and right now it's being thrown away. Before Skillsync, Nars and I built an open-source payment orchestrator that let merchants route across many processors instead of getting locked into one (30k+ GitHub stars). Fighting vendor lock-in by making incompatible systems interoperate was the whole job. We came into YC with a different idea, but the more we lived inside coding agents the more we saw the same problem from the other side: your context is locked into whatever agent you start with, because every agent stores sessions differently and none of it moves. Some of the things we’re seeing people do with Skillsync are: - Moving a session between Claude Code, Codex, and Cursor mid-task, including when you hit a usage limit on one and want to keep going on another. - Thinking through a problem in a browser Claude chat, then handing the whole thread to Claude Code or Codex to build. - Sharing research sessions with a team. Everyone uploads their sessions, so teammates and their agents can build on each other's work instead of repeating it. It's available as a Mac app, CLI and MCP. Here's the demo: https://youtu.be/7hVhSnSKGl8 . Would love to hear your feedback and answer questions! Comments URL: https://news.ycombinator.com/item?id=49743049 Points: 4 # Comments: 0

Hacker News 熱門51

Show HN: Die With Me – Claude and Codex rate limits as AIM away messages

made this AIM buddy list but for our token usage because i wanted to bring a little bit of the old internet back! i remember those days where i'd casually check if friends are online and find excuses to say hi and poke them give it a try! i'd greatly appreciate feedback Comments URL: https://news.ycombinator.com/item?id=49743095 Points: 7 # Comments: 8

arXiv cs.AI83

ProgramDistill: From Interactive Web Apps to Verifiable Reference-Guided SWE Tasks

Coding agents are typically evaluated with desired behavior specified through issues or instructions. In practical web development, however, agents may need to infer behavior from working software and implement it in an incomplete application. We introduce ProgramDistill, a benchmark evaluating coding agents on features discovered through interaction with fully functional reference applications. We build ProgramDistill by factorizing applications into features of different granularities, each associated with replayable behaviors executable via its gold patch. Our pipeline, mine-craft-patch, discovers 1,975 replay-verified behaviors across 26 applications and constructs 4,063 tasks without human intervention. Across nine frontier coding agents, GPT-6 Astra and Claude Opus 5 achieve 49.2% and 28.8% success on cumulative workflows in full-application reconstruction. In partial-application reconstruction, success falls from 100% to 64.0% and from 96% to 32% as restoration depth increases from 1 to 8. ProgramDistill thus provides a scalable benchmark with controlled difficulty for evaluating and diagnosing coding agents, and a natural basis for future curriculum-based training.

Hacker News 熱門81

Show HN: Pizza Bot – An inbox for AI agents that work in the background

Hi HN - long-time lurker (since 2012!), first time poster. Pizza Bot is a self-hosted desktop app for Mac, Windows, and Linux that runs AI agents in the background and exposes them through an email-like UI. Finished work shows up in Unread, and anything waiting on your approval shows up in Action. It's Apache 2.0-licensed, there's no signup and no telemetry, and you bring your own model provider: Anthropic, Amazon Bedrock, Google Gemini, OpenAI, OpenRouter, or a local model through Ollama. There are builds on the releases page, or you can run it from source. Pizza Bot started as an internal passion project I worked on with a small team at Amazon. The whole thing came out of my frustration at having to manually log CRM activities through a browser form. I built a simple REST API called "JoeBot" that connected to my authenticated browser session over CDP and filled out the form for me using Playwright. Then I hacked up a quick Obsidian plugin so I could trigger it from my local notes (no AI and no MCP servers involved). This caught on quickly. My fellow AWS Solutions Architect Igor Fil joined up with me, and we rebranded the project as "Pizza Bot," named after Amazon's two-pizza teams. We started seeing what other automations we could build. We found a GraphQL API we could query and hacked up some "recipes" to pull data out of the CRM to help with meeting prep. That worked great, and it was right around the time MCP servers seemed to be taking off, so we decided to expose Pizza Bot as an MCP server instead, so it would be available to AI tools through natural language. This was a decent solution for technical users, but the Account Managers who live inside our CRM system wanted something too. We decided to rebuild Pizza Bot as an Electron desktop app modeled after an email inbox, so it would be familiar to non-technical users and would run on both Mac and Windows. We also bundled internal MCP servers as OCI images and hosted them in Amazon ECR as an "addon marketplace" so users could install them with one click without having to set up Amazon developer tooling. The project took off organically and expanded outside of AWS into the wider Amazon organization globally. More than 2,000 people ended up using it for meeting prep, email drafting, Slack summaries, CRM logging, prioritizing their day, and web research. Once apps like Claude Cowork and Amazon's own Quick Desktop came out, we realized the real growth opportunity was outside of Amazon. Rather than try to rip out the Amazon-specific integrations, we rebuilt Pizza Bot once more as an open source project. We leaned on coding agents heavily, which is the only reason a team our size could pull off a full rewrite. I'm pleased to say it's finally public, and we're hoping to bring in community members and see where it goes. We'd like to do for knowledge workers what Claude Code and Codex have done for programmers. A couple of things to know up front. Most of what made Pizza Bot useful on day one inside Amazon came from that internal catalog of skills and MCP servers for Amazon's own systems, and none of it could come out with the app. So it ships thinner than the version those 2,000 people used, and building that catalog back up for tools other people actually use is where we need the most help. It's also a community project and not an AWS service, so there's no support or SLA behind it. The Windows and Linux builds aren't signed yet either. On the technical side, Pizza Bot is a server and a client. The desktop app bundles both, or you can point a client at a remote backend; personally, I self-host the server on my home network and reach it from my phone over Tailscale. The server owns the thread lifecycle and checkpoints state with DeepAgents and LangGraph, and clients rehydrate from it as needed, so you can disconnect mid-run and pick the thread back up from another client. Approval pauses outlive the session that created them and collect in an Action filter, so you can answer an hour later from a different device. The agent you talk to has a sandboxed QuickJS interpreter that can reach your filesystem only if you grant it a folder, but its main job is to delegate. Each subagent is a 1:1 mapping of a Skill, and an Activity bar shows that subagent and the tool calls it's making as it works. Memory is opt-in and stored as plain markdown files on your machine. Every tool call is explicit, including looking up a memory - we err on the side of transparency to reduce surprises. Tools come from MCP servers, and skills are ordinary SKILL.md files with a per-tool approval policy, so existing skills that don't require a code interpreter should still work. What I'd most like to hear about is where the app itself gets in your way, the kind of problem you can't fix by writing a skill or an MCP server. I'm around today to answer questions! Comments URL: https://news.ycombinator.com/item?id=49713894 Points: 15 # Comments: 5

arXiv cs.AI82

在使用 LLM 进行民调之前：审议式诊断框架

LLM 能像人类一样推理新信息，还是只会检索缓存的观点？对于硅基抽样而言，这是一个关键问题，因为 LLM 人格化角色能够大规模模拟公众舆论。当前的评估只检验人格化角色是否持有正确的观点——即一个静态快照。然而，舆论研究越来越依赖动态保真度：人格化角色是否会像人类在审议过程中那样，针对新的论点更新信念。目前还没有任何基准测试对此进行检验。我们提出了审议式民调诊断框架，用于比较人类和 LLM 在接受完全相同的信息干预后信念的变化。该框架以审议式民调为基础，能够揭示静态评估无法发现的失败：模型即使能够生成看似合理的党派观点，仍可能错误呈现这些观点会如何变化。我们使用 America in One Room 的数据（526 个人格化角色、72 个问题），将该框架应用于五个前沿模型，发现每个模型都存在失败，而且表现方式各不相同。GPT-5.1 表现出逆转现象：在获得平衡信息后，其人格化角色对 opposing party 变得更加敌视，而人类则相反，敌意有所降低。这种逆转具有选择性（在外群体问题上为 80%，在政策问题上为 26%），并且在不同党派身份之间呈对称分布。Gemini 2.0 Flash、Claude Sonnet 4.5 和 Llama 3.3 70B 表现出过冲：观点变化方向正确，但幅度达到人类的 5—7 倍。DeepSeek V3 则表现出僵化，变化幅度接近于零。针对性的消融实验表明，政策内容会触发这些失败，而且它们与身份有关：GPT-5.1 在外群体问题上发生逆转，但在内群体问题上出现过冲；Gemini 则正好相反。我们将这一特征称为“自我谄媚”：模型遵循其对人格化角色的内部刻板印象，而不是依据所提供的信息进行推理。我们的框架提供了一套具体方案：在信任 LLM 人格化角色模拟信念更新之前，先运行审议式诊断。

Hacker News 熱門27

Ask HN: What default model do you use and why?

I use claude for most of what I do, and Fable is largely overkill for me and frequently burns through my Max plan's session credits in minutes (!) when just doing an initial mobile app planning with 4 agents. After I waited out the timeout period 6 hours later, and I picked up again, the cache had timed out so it burned through 2% of the session in less than a minute. Opus 4.8 is now my goto and I will be avoiding 5 until I see a reason to switch back. 4.8 is 'good enough' for what I need and has been a great value. It mostly gets things right. Most of what I do is web and mobile , largely cloud backend. Comments URL: https://news.ycombinator.com/item?id=49672966 Points: 4 # Comments: 0

Hacker News 熱門55

Show HN: Clawfight.ai MCP-driven agentic game play

How should agents interact with other agents? What happens when they rap or fight against each other with the pressure of human spectators? Clawfight.ai is an experiment to explore this space. This is my first post about it. The journey began on a beefed up machine I purchased with a decent GPU (5090). I installed claude code and set to dangerously skip permissions, enabled /rc and became completely submerged in the all-hours modern AI builder workflow. I stood up openclaw to see if I could have an agentic org drive this project. I spent a good amount of time on “openclaw ops”, babysitting 3 agents and trying to make them the best versions of themselves. They still do stupid things that cost me tokens. For the game render, I started out using Unreal Engine and allowing agents to remote control their players, but the quality just wasn’t there. I recently moved to doing near-real time video renders of the match and will bring UE back in for multi-agent games. Everything is AI generated. I’m fascinated by the future of realtime video generation and building out native MCP infrastructure. You can play directly from model provider apps (claude connector or openai plugin) and has fallback support for more basic HTTP clients. But the architecture and gameplay is MCP first. Tell your agent/app “go read clawfight.ai/agents.md and play” Comments URL: https://news.ycombinator.com/item?id=49658483 Points: 5 # Comments: 0

Hacker News 熱門64

Claude 不再向未成年人提供服务

文章 URL： https://support.claude.com/en/articles/15171100-age-assurance-on-claude 评论 URL： https://news.ycombinator.com/item?id=49656225 积分：4 # 评论：0

arXiv cs.AI76

LOCUS：面向任务的低秩后训练，实现高效令牌生成

大型语言模型的服务成本随输出序列长度直接增长，但标准的偏好对齐通常会在不提升实用性的情况下增加响应冗长度。我们研究了后训练更新的参数化方式是否会影响生成长度：低秩子空间能够在不改变对齐损失的情况下改变序列长度。我们提出 LOCUS，这是一种选择面向任务的低秩适配子空间的方法，旨在满足实用性约束的同时最小化输出令牌成本。在该子空间内，后训练保留原生偏好目标，并冻结主干模型。基于 Anthropic HH-RLHF 对话偏好数据集，我们评估了两个约 3B 参数的解码器主干模型 Pythia-2.8B 和 Qwen2.5-3B，并将其与遵循匹配协议的全参数 DPO 和 DrDPO 分支以及发布的 SamPO 检查点进行对比。LOCUS 在 Pythia-2.8B 上最多将延续长度缩短 39.84%，在 Qwen2.5-3B 上缩短 14.87%–17.58%，而仅更新模型参数的 0.24%–0.28%，且内部偏好诊断指标没有实质性变化。

arXiv cs.AI78

TRACE：使用合成奖励训练因果探索推理智能体

带有可验证奖励的强化学习（RLVR）推动了语言模型在数学和代码等领域的推理能力发展，因为这些领域的客观答案易于验证。复杂数据上的诊断推理则不具备这一优势：确定异常的真正原因通常需要成本高昂的专家调查，而且事后也可能仍然存在歧义。我们探讨能否通过工程手段利用这种验证不对称性。我们抽样一个干预，将其注入受控模拟器，并生成该干预会产生的观测结果。隐藏的干预提供了预言机标签和客观奖励，而智能体仍必须调查嘈杂、混杂且分散的证据。 我们在 TRACE 中实现了这一方法。TRACE 是一个数字广告诊断环境，包含 12 种根本原因以及细粒度的细分归因。智能体使用 Python 和 SQL 调查每个回合，并必须识别根本原因，以及在适用时确定受影响的细分归属。在一个留出的 235 回合测试集上，表现最强的提示基线 Claude Opus 5 达到 0.686 的 FullAttr@1。监督微调将 Qwen3.5-35B-A3B 的得分从 0.159 提升至 0.637，随后使用合成奖励进行强化学习，得分进一步达到 0.757，超过了所有评估的提示基线，包括前沿闭源模型和经过提示的 Qwen3.5-122B-A10B 模型。所得策略使用的工具调用次数也显著少于经过提示的 35B 基础模型。这些结果表明，能否获得可扩展且客观的训练信号，可能比模型规模本身更为关键。更广泛地说，基于模拟的验证可以使原本存在歧义的诊断推理任务适用于可扩展的强化学习。

Hacker News 熱門73

Show HN：自托管的企业 OS，在各部门中运行 Claude Code 和 Codex 智能体

大家好，我是 Dimitris。 这是我构建并用于运营自己企业的企业 OS，任何人都可以安装并免费进行自托管。可以把它理解为将 Claude Code、Cowork 和云端会话整合在一个自托管应用中。 它从设计上就是一个多租户应用，许多人可以通过 4 种不同的协作模式与企业智能体协作；它可以使用你的 Anthropic 或 OpenAi 订阅运行，也可以使用本地模型。 每个智能体都可以运行在 Claude Code 或 Codex CLI 上，作为服务器上的持久进程运行于内核沙箱（bubblewrap）中，并始终启用网络隔离（pasta）；每个智能体都有自己的工作区、记忆、日程安排和工具。同样的智能体也可以配置为通过一个出站 WebSocket，在任意远程计算机上以完全相同的方式运行（无需入站端口，也不需要 VPN）。 智能体已经内置了许多功能。例如，它们可以通过 Twilio 或你自己的 Asterisk 接听和拨打电话，可以编辑视频以及 Excel、Word、PPT 文件，还可以借助 Collabora 直接在聊天中预览这些文件，此外还有更多功能。 只需一个安装脚本和一个 Docker Compose 文件即可开始使用，无需注册。该许可证采用 Fair Source，所有代码均为公开代码，并且最多支持 5 名用户的自托管是免费的。1.6.0 已于今天发布。 AI 免责声明：OtoDock 的大部分代码都是使用 OtoDock 自身、运行 Claude Code 编写的。 我很希望大家阅读代码、了解一下沙箱模型，并告诉我是什么因素会阻止你在自己的硬件上运行它。 GitHub： https://github.com/OtoDock/oto-dock 网站： https://otodock.io 评论链接： https://news.ycombinator.com/item?id=49630606 得分：3 评论数：0

Hacker News 熱門69

Show HN：Rdltr——阅读清单的收件箱归零

嗨，HN！过去几个月里，我一直在开发一款名为 RDLTR（ https://rdltr.app ）的稍后阅读应用。整个想法源于我在不同浏览器、Mac、Linux 和 iPhone 上打开了太多想读的网页标签。它们让所有东西变得杂乱不堪，也妨碍了我处理其他事情。 这个想法出现时，Pocket 刚刚关闭。我四处了解了一下，试用了 Instapaper、Raindrop、Matter，还有其他几个我已经记不清名字的产品。它们都还不错，但都会给你一份只会不断增长的列表，而这又成了另一件需要维护的事情，也不是我想要的模式。Safari 有一个“阅读列表”功能：读完某篇内容后，它就会从未读列表中消失。但它只能在 Apple 设备上使用，而且功能也不太完善。我想要的是阅读方面的收件箱归零：打开某篇内容，读完它，再回来时，它已经从队列中消失，但仍然保存在归档里。不断继续，直到没有剩下任何内容。 链接可以从任何地方添加。在页面上粘贴 URL、拖入链接，或者使用书签小工具、Chrome 和 Firefox 扩展，或用于 iOS 分享菜单的 Apple Shortcut。此外还支持 Vim 按键、为每个链接生成 TL;DR 摘要、搜索所有已阅读或保存的内容、为每个列表提供 RSS，而且在手机上使用同样顺畅。 这些扩展还有几个我一直在使用的小技巧。Shift+DD 会提取文章正文、清理格式，然后在新标签页中将其发送给 ChatGPT、Claude 或 Gemini，这样你就可以询问它任何与刚刚阅读的内容有关的问题。Shift+AA 会在 Internet Archive 中打开当前页面；当你没有订阅而网站设有付费墙时，这个功能很方便。 技术栈包括 Bun、TypeScript、HTMX 和 SQLite，并使用 SSE，让列表在你保存或读完某些内容的瞬间更新；应用部署在 Hetzner VPS 上。 坦白说，大部分代码都是 Claude Code 写的。我仍然花了大量时间进行实际使用、测试，并不断打磨 UI，直到一切用起来感觉恰到好处。 总之，很想知道你目前是如何处理这件事的。 评论链接： https://news.ycombinator.com/item?id=49629747 得分：3 评论数：0

Hacker News 熱門77

我们的经济未来会是什么样？

文章 URL： https://www.anthropic.com/institute/econ-scenarios 评论 URL： https://news.ycombinator.com/item?id=49626373 积分：14 评论数：4

arXiv cs.AI70

Performance of Clinical AI System and Physicians and Frontier Language Models in primary care diagnostics

Clinical AI evaluation should encompass diagnosis and management after adaptive information gathering. We compared Doctorina, eight physicians and four standalone frontier language models in 150 synthetic Polish-language primary-care consultations. Doctorina achieved 82.0% Top-1 concordance versus 57.0% for physicians (difference, 25.0 percentage points; 95% confidence interval, 17.7-32.7) and 97.3% versus 85.0% primary-or-reference-differential concordance. Across 149 case pairs, normalized workup and treatment scores were 89.4 versus 66.9 and 83.7 versus 61.2. Doctorina had the highest diagnostic point estimates among all six groups; Kimi K3 ranked next, while Claude Opus 5 led the closely spaced management estimates of Opus, Doctorina and Kimi. A second Doctorina execution reproduced the advantages over physicians across all outcomes. Doctorina's advantage over physicians therefore extended from primary-diagnosis selection to higher-rated diagnostic workup and initial treatment after adaptive consultation.

arXiv cs.AI61

Scratchy: Visual-Scratchpad Multimodal Reasoning for Cryptographic Proof Generation in EasyCrypt

Large language models (LLMs) have recently made substantial progress in formal proof generation, yet presenting distinctive challenges in cryptographic area. Computational security arguments posit that a valid proof must coordinate probability, adversarial games, invariants, assumptions and bounds, which can be provided by a machine-checked framework named EasyCrypt. Although all objects may appear in available context, LLMs still struggle because proof-theoretic dependencies are typically implicit in a linear representation and distributed across multiple programs. So, this paper presents Scratchy, a visual-scratchpad approach that exposes these dependencies for multimodal generation. Given the natural-language security description, with formal context and target propositions, the proof objects can be normalized into a typed proof-relation graph. Then a structure-preserving visual compiler transforms the graph into the formula-rich visual proof state that guides a multimodal model in generating the EasyCrypt proof. Also, the Scratchy-eval, a 114-task dataset derived from reliable official EasyCrypt files, has been introduced. It contains 64 security-form proof generations and 50 multiple-choice knowledge tests. After a series of evaluations, covering semantic grounding, relational invariants, and game reductions, classical LLMs like GPT-5.6-Sol and Claude-Opus-5 have gained a clear advantage from Scratchy's structured visual proof states. This contrast suggests that explicit proof structure can make the improvement and multimodal proof-state representation as a promising direction for computer-aided cryptography.

arXiv cs.AI85

SWE-Test：通过输入预测对 LLM 漏洞发现能力进行基准测试

漏洞发现正逐渐成为大型语言模型（LLM）代理的一项重要能力：悄然漏掉真实缺陷的代理会使关键软件暴露于风险之中。因此，严格测量这项能力迫在眉睫。然而，现有基准测试可能因数据污染而被钻空子，以一个无法获知的漏洞集合为基准评估召回率，往往依赖合成漏洞，并且只报告单一的端到端结论，无法定位代理究竟在哪个环节失败。漏洞发现是一项复合能力：代理必须理解源代码、推断输入约束、构造输入、执行输入，并根据反馈反复修正。我们将其测量重新定义为一种具有封闭且确定性真实答案的输入预测任务：利用覆盖率引导的模糊测试，我们从真实世界的 C/C++ 程序中挖掘深层目标分支，并要求代理预测一个能够驱动执行到达给定分支的输入。我们在涵盖 15 个领域的 22 个真实世界 C/C++ 程序上，将漏洞发现分解为三种任务模式。Open-loop 和 Feedback-enabled 在其中 16 个代码库（涵盖 13 个领域）上共享 60 个固定目标任务实例，用于测试不借助和借助距离预言机的输入构造能力，从而将代码理解与反馈驱动的修正区分开来。Online Arena 则移除预先设定的目标，在一个由 11 个程序组成、与前述集合部分重叠的独立程序池上，通过覆盖率增益对路径探索进行评分；代理在 6 个程序中共同确认了 13 个不同的漏洞。在评估的 15 种默认工作量模型-脚手架配置中，最佳配置在 Feedback-enabled 模式下的通过率也仅达到 55.0%；在 7 组配对的 Claude Code 配置中，带反馈时的平均通过率为 36.4%，不带反馈时为 19.3%。通过分解失败原因，我们发现约束推断而非路径导航是主要瓶颈。我们发布了 SWE-Test，并提供开箱即用的评估环境。

arXiv cs.AI75

必要还是充分？使用行为证据评估 LLM 解释

LLM 决策组件在 agent 工作流中运行时，常会给出与行动相关的建议或判断，并附带解释。操作人员可能会利用这些命名因素来监控系统、诊断错误，或决定何时升级处理某个输出。这种用法假设这些解释与组件可观察到的决策行为一致。我们测试了这些命名因素的两种解释：必要性，即改变某个因素会改变输出；以及充分性，即在移除其他可变信息的同时保留该因素会保持输出不变。我们在两个合成用例中评估这些解释：向客户推荐顾问，以及判断提示词是否具有有害性或风险。模型返回一个输出以及最能影响它的前三个因素。受控的 black-box 干预通过衡量改变某个因素后输出发生变化的频率来估计其必要性得分，并通过衡量保留某个因素后输出得以保持的频率来估计其充分性得分。在来自 Claude、GPT 和 Gemini 系列的八个模型中，引用排名与必要性和充分性得分之间的平均斯皮尔曼相关系数在顾问推荐任务中分别为 0.349 和 0.354，在提示监控任务中分别为 0.431 和 0.580。此外，在必要性条件下，57.6% 的顾问响应中，某个未被引用的因素得分高于得分最低的被引用因素；在充分性条件下这一比例为 58.1%。对应的提示监控比例分别为 25.8% 和 8.9%。被引用的前三项包含有用信息，但并不能可靠地识别在必要性或充分性意义下、经测量影响最强的三个因素。该框架为代理监督中使用的解释提供了一种 black-box 可靠性检查，同时仍限定在单个 LLM 决策范围内。

arXiv cs.AI78

感知底层环境的 AI 代理：将执行上下文作为一等输入

自主 AI 代理越来越多地在这样一种环境中选择行动：其内存、执行时间、运行时、计算和操作约束决定了什么才算合适的计划。我们将代理规划状态中缺少这种执行上下文称为 substrate blindness。我们通过数值代码生成来检验这一一般性命题，因为在该任务中，所选实现选项及其运行后果可以直接观察。三种前沿模型配置——Anthropic Claude Opus 5、OpenAI GPT-5.6-Sol 和 Google Gemini 3.7 Flash——针对一个高维成对欧氏距离任务生成代码，输入要么只有任务本身，要么再附加一个 128 MB RAM 和 10.0 s 墙钟时间约束。披露约束使 14 组可执行、索引对齐的“仅任务”与“披露约束”比较中有 13 组的测得进程峰值内存降低；在所有三个队列中，平均墙钟时间也降低，使执行速度最高提升 3.1 倍。在整个审计语料中，披露约束带来了结构性代码变化，包括有界分块、保留 float32、上三角遍历，以及原地或内存映射缓冲区。在更严格的 96 MB 约束下，独立抽样的披露约束队列在正确且未超预算的结果上，Claude Opus 5 为 4/5，GPT-5.6-Sol 为 5/5，Gemini 3.7 Flash 为 3/5；而仅任务条件下分别为 0/5、1/5 和 0/5。各队列的平均 MaxRSS 和墙钟时间分别比仅任务参考低 49–74% 和 35–64%。这些结果为感知底层环境的代理规划提供了一个受控的概念验证：最小执行约束会在生成程序中诱发主动的结构适配，将计算从无约束分配转移开来，并在执行前显著改善观测到的资源—时间特征。

Hacker News 熱門48

Show HN: Moadim.io – A scheduler for agents

Why can't we get an agent scheduler that supports all of the following: - git compatible - agent agnostic - 100% open source - os and system-agnostic - multi-runner support - support mcp/ui/http - unlimited routines/crons So I built one, moadim.io is a local Rust daemon you install in the target machine, give it a name and manage its routines via a Git repository, wants a new routine that send you a daily message from this machine? Create a pr and merge, have another routine that run every hour to pull the latest changes to the ~/.config/moadim folder. With more than 1,000 users, I define this project as almost "done" and ready for production. Me and thousand more people are use it in a daily manner. It currently supports Claude, Codex, Hermes, and Pi, and you are welcome to add your agent of choice as well because it's 100% configurable. You are welcome to have a look at the source code of the daemon in " http://github.com/moadim-io/daemon " Feel free to provide me with suggestions for more features around this topic. I don't want to branch out to new off road topics likt webhooks, this is a "done" software in the realm of agents schedulers that focus on cron-like work. Also feel free to start the github repository and open issues and PRs for your suggestions. Comments URL: https://news.ycombinator.com/item?id=49571537 Points: 3 # Comments: 0

Hacker News 熱門68

形式化费马大定理

文章链接： https://www.anthropic.com/research/formalizing-fermats-last-theorem 评论链接： https://news.ycombinator.com/item?id=49568506 得分：20 # 评论：1

Hacker News 熱門83

美国企业界正迷上开源 A.I

文章链接: https://www.nytimes.com/2026/09/04/technology/open-source-ai-anthropic-openai.html 评论链接: https://news.ycombinator.com/item?id=49566137 得分: 5 评论数: 1

Hacker News 熱門

Show HN: Reactor Atlas

Hi HN, I'm a nuclear engineer and tech entrepreneur. After working for 3 years in the National Atomic Energy Commision (CNEA) in HPC software for nuclear simulations, I co-founded ratherlabs.com, a software company providing blockchain solutions for startups. Rather Labs business reached ~6 million usd in annual revenue in 2024; now demand is decreasing as our niche clients are using AI to launch their MVPs. Now I'm joining my scientific background with my entrepreneurial inclinations to think of problems related to the energy industry. Nuclear is experiencing a renaissance now (it was in a dark age when I was studying after the Fukushima event) so I think this is the right moment for me to jump again into the field. Reactor atlas is an interactive nuclear reactor map and an intelligence layer on top of it. The basic layer indexes: - power reactors - research reactors - fuel facilities with their history and projection, by country. The project is solving a fragmented information ecosystem, where operational and commercial data is hard to conciliate. Also, I'm monitoring nuclear news associated with each of the facilities, political declarations that may affect the nuclear landscape and geo-phisical news (like earthquakes) that could affect operation. On top of the operational + news layer, I'm creating an intelligence layer that will help institutions of different types make faster decisions by subscribing to very specific alerts. As an engineer, I used to code in fortran and c++ when I was working at CNEA, and later I participated in the first Rather Labs projects architecting complex async systems, mostly using MERN stack. I'm no longer coding anymore. I'm focused on designing robust systems with AI agents. I mainly used Claude Fable 5.1 to develop reactoratlas.com, deploying on vercel and using mostly nextjs, threejs, postgreSQL and some cron jobs to query and update data. I miss coding as in the old days, but I haven't code a single line in this project. And I strongly believe we humans shouldn't code anymore: our responsibility is now on designing and supervising AI systems. Thanks for your attention! Comments URL: https://news.ycombinator.com/item?id=49549148 Points: 21 # Comments: 4

arXiv cs.AI71

Environment Evolution for Terminal Agents

Scaling interactive and verifiable environments is critical for training terminal agents. As frontier models become more capable, environments synthesized from scratch become less challenging and thus provide limited learning signals. Recent co-evolution methods iteratively synthesize environments near the model's learnable frontier based on weaknesses exposed during rollouts. However, their dependence on on-policy rollouts limits generalization and the continuous provision of learning signals as the model becomes stronger. In this paper, we propose environment evolution, which incrementally increases environment difficulty off-policy and schedules the evolved environments generation by generation during training to provide continuous learning signals. We derive three evolution directions that influence environment difficulty from the multi-turn learning objective and then implement evolution along these directions through a loop-engineered multi-agent harness. Quantitative rollout experiments with Hy4 preview, Claude Opus 5, and GPT-5.6 Sol show that environment evolution consistently produces more difficult environments. We validate its effectiveness on Qwen3.6-27B and Qwen3.6-35B-A3B through simple long-horizon RL training, improving their performance by 14.4 and 18.0 percentage points on Terminal-Bench 2.1, respectively.

arXiv cs.AI79

ESPO：通过诊断、扩展和稳定化实现错误结构化提示优化

像 GEPA 这样的进化式提示优化器会遭遇提示词膨胀：每次迭代都会追加规则和注意事项，导致提示词长度最多增加 3$\times$，却并未更准确。我们将其归因于三个缺陷——错误观测不完整、搜索多样性有限以及选择不可靠——并提出 ESPO（Error-Structured Prompt Optimization），将提示优化分解为三个阶段：Diagnose 在一轮中将所有训练错误聚类为结构模式；Propose 通过四种具有互补偏置的策略生成候选；Select 采用 bootstrap 稳定性选择。在七个公开 NLP 基准——Tweet、MMLU、GSM8K、HotpotQA、ScoNe、HoVer 和 PUPA——上，ESPO 相比当前最优方法平均准确率提升 $+$3.76 个百分点（74.67% 对 70.91%，GEPA），在每个数据集上都与 GEPA 持平或更优，同时生成的提示词短 47%（1,004 对 1,878 个字符），且推理更快。跨模型实验覆盖另外四个学生模型（Gemma 3 12B、Mistral 14B、Qwen3 32B、Claude Haiku 4.5），结果表明 ESPO 在每个测试模型上都取得了最佳平均准确率，其中在 Qwen3 GSM8K 上差距最大（15.00% $\to$ 91.40%）。附录中的一个泛化界将每个阶段与测试时差距中的相应项对应起来，而消融实验验证了一个关键预测：在没有 bootstrap 选择的情况下加入多样性实际上会损害性能（$-$1.20%）。

Hacker News 熱門76

OpenAI 表示其最新 AI 模型已超越 Anthropic

文章链接： https://giftarticle.ft.com/giftarticle/actions/redeem/1054f4f0-cac7-479c-a4a7-f95b3906ca4b 评论链接： https://news.ycombinator.com/item?id=49554060 得分：13 评论数：2

Hacker News 熱門56

Porting my 1993 Amiga game to Godot, with an LLM reading the 68000 assembly

These are my notes from porting my Amiga game, which I originally built in Baghdad in 1993 in MC68000 assembly, to Godot, using Claude Fable 5 during last July holiday. It took an evening! Getting the feel right and shipping it took a few more weekends and evenings. I spent the last few weeks analyzing what Claude did, feeding it my 33 years of memory of how I built the game, my notes and the git repos. It wrote the first draft of the article, and I edited line by line over a week. The screenshots of my 1993 map editor is the first I have run it since then. The one thing I never checked myself is the 108-byte explanation. Please post any questions. I am also releasing the original game for free. Comments URL: https://news.ycombinator.com/item?id=49550375 Points: 7 # Comments: 2

Hacker News 熱門73

ChatGPT、Claude 和 Grok 宕机了

文章 URL： https://www.macrumors.com/2026/09/03/chatgpt-claude-and-grok-are-down/ 评论 URL： https://news.ycombinator.com/item?id=49551658 得分：11 评论数：4

Hacker News 熱門72

Claude 面向 Commerce Agents

文章 URL: https://claude.com/blog/claude-for-commerce-agents 评论 URL: https://news.ycombinator.com/item?id=49547888 得分: 29 # 评论数: 26

arXiv cs.AI81

大型语言模型中的“先高后低”请求与拒绝行为

“先高后低”技巧在语言模型上也有效吗？在人类中，一个较大的请求如果被拒绝，会使随后提出的较小请求更有可能被接受。我们在来自三家提供方的九个商用模型上测试这一点：每个模型先拒绝一个较大的请求，然后再收到同一请求的较小版本，并将其服从率与直接提出请求时进行比较。答案取决于模型。在 Anthropic 的前沿模型上，这一技巧有效：Opus 5 在拒绝较大请求后，有 65.8% 的概率答应较小请求，而直接提出时这一比例为 29.3%。在 OpenAI 和 Google 的前沿模型，以及 Haiku 4.5 上，它却适得其反，使服从率降低 15.5 到 23.0 个百分点。一个对照实验定位了这一效应：在所有九个模型上，针对无关主题的被拒绝大请求产生的影响都小于相关请求，因此让步本身在所有模型上都很重要，而对刚刚拒绝过某件事的反应则因模型家族而异。该技巧不能迁移到来自公开基准的拒绝样本。决定“退一步”是否奏效的是请求内容：将 265 个被拒绝的、旨在获取可用指令的请求改写为同一主题的解释性请求后，有 263 个案例中的拒绝被消除了。人类的影响技巧迁移到语言模型上时，是按模型家族逐个发生的。

arXiv cs.AI79

Defense-as-Skill：为技能增强型智能体演化运行时防护技能

技能增强型智能体将可复用技能作为持久的运行时上下文加载，从而提升任务表现，但也为恶意技能提供了一个持久渠道来操控后续动作。这类技能可能泄露机密、破坏代码、绕过审批，或在具体用户任务和工作区状态使不安全操作看起来有用之后，再为数据外泄做准备。这使得安装前审查不足，并需要在运行时基于任务条件提供保护。我们提出 Defense-as-Skill，一种将运行时防护本身实现为可安装、可检查和可编辑技能的防御范式。我们的防护器 SkillSonar 与不可信任务技能并行运行，并将敏感操作与用户任务边界进行比对，无需修改底层智能体运行时即可将每个操作导向允许、重新规划或确认决策。为研究这一设定，我们构建了 SCOPE-R，一个任务条件化数据集，涵盖 6 个风险家族和 21 个子类别，包含 206 个经攻击确认的恶意实例和 43 个良性任务。随后，我们在 SCOPE-R 训练子集上利用运行时防护技能演化来改进 SkillSonar；这是一种 Monte-Carlo Tree Search 过程，可根据 rollout 的反馈演化磁盘上的防护技能。在 Claude Code 和 OpenClaw 上，演化后的防护器显著降低了攻击成功率，同时保持了良好的安全-效用权衡。在重复的 GLM-5 运行中，SkillSonar 将 ID ASR 从 0.482 降至 0.104，将 OOD ASR 从 0.606 降至 0.115。进一步分析表明，该方法可跨受害模型、留出风险家族和外部基准迁移，并且对自适应攻击者仍保持防护能力。消融实验进一步表明，显式的安全责任分配和技能原生表示对于所观察到的收益都很重要。

Hacker News 熱門75

改进我们的对齐和安全工作

文章链接： https://www.anthropic.com/news/improving-alignment-security-efforts 评论链接： https://news.ycombinator.com/item?id=49529567 得分：5 # 评论数：1

Hacker News 熱門77

Claude Fable 5.1 有哪些新变化

文章 URL: https://platform.claude.com/docs/en/models/fable-5-1/whats-new-fable-5-1 评论 URL: https://news.ycombinator.com/item?id=49525378 点数：35 评论数：3

Hacker News 熱門71

Claude Fable 5.1 和 Claude Mythos 5.1

文章 URL： https://www.anthropic.com/claude-fable-and-mythos-5-1 评论 URL： https://news.ycombinator.com/item?id=49525496 得分：39 评论数：3

arXiv cs.AI81

关于 Agent 插件的维护与共进化：Claude Code 插件市场的实证研究

AI 编码代理是一类通过推理和工具使用自动化开发任务的软件工具，正日益通过插件市场进行扩展，但这些新兴仓库的结构、维护以及共进化动态仍缺乏实证研究。与通过源代码提供功能的传统软件包不同，agent 插件通过自然语言指令文件、脚本和配置文件的组合来提供功能，这引出了一个问题：这些插件究竟是会在各组件之间共同进化的可维护制品，还是开发者只需编写一次、无需再回看的一次性制品。为研究 agent 插件的维护与共进化，我们对 1,926 个托管 Claude Code 插件市场的仓库进行了实证研究，分析了 2,018 个市场中的 8,351 个插件和 77,773 次提交。我们发现，该市场正在快速扩张，在 2025 年 10 月发布后的六个月内，涉及插件的提交活动增长了 8.8 倍，而面向软件工程任务的插件占所有插件的 61.3%。插件开发主要由功能驱动，功能相关提交的发生率是传统开源软件（OSS）的两倍多（39.6% 对 17.2%）。Claude 参与共同编写了 34.9% 的全部提交，而且四类提交（docs、perf、style 和 refactor）在插件仓库中的含义与在传统软件中有显著不同。大多数组件类型独立演化，但在 skills 目录内部，自然语言指令文件与实现脚本以高于随机的比例共同进化，其中 78% 的共同变更在功能上是耦合的，这代表了一类在传统软件工程中未曾观察到的新型维护依赖。

arXiv cs.AI75

FaulT-Bench：面向在不可靠用户工单下评测网络故障排查 LLM agents

基于 LLM 的智能体正越来越多地被用于网络故障诊断，但现有基准只在准确工单上评测它们，并且总是假设存在故障，而这两种条件在实践中很少成立。我们提出 FaulT-Bench，一个涵盖八种网络拓扑、共 200 个故障排查场景的基准，其中五种取自公开实践者实验室并重新实现，覆盖真实故障、错误故障报告、错误设备归因以及错误根因主张。为隔离工单措辞对诊断的影响，我们进一步将 72 条带有错误前提的工单改写为五种报告者 persona，在保持网络状态固定的情况下，每次只改变一个因素，分别体现报告者置信度和可验证细节的差异。我们的自动化框架在 Kathará 中部署每个场景，让智能体通过 NIKA 工具接口进行交互，并使用 LLM judge 按结果、修复方案和推理质量对自由文本诊断进行评分。我们评测了 SADE、ReAct 和 Claude Code，发现三者在准确工单上都接近饱和，并且对误导具有鲁棒性，但当网络本身健康而工单出错时，性能会急剧下降：它们会不断探查，试图把良性状态提升为根因，而不是得出没有问题的结论。persona 改写显示，工单的写法比其声称的内容更重要：一份自信但错误的报告，其处理效果大致与准确报告相当，而一份模糊、细节不足的报告会使性能显著下降。这三个智能体的失败方式也各不相同，从持续过度诊断到运行无回应，而且成本差异很大。这些结果将 FaulT-Bench 定位为一个基准，用于开发能够可靠处理真实世界网络故障排查中嘈杂、不可靠工单的智能体系统。

Hacker News 熱門

Show HN: Watches user sessions, finds bugs that matter, and fixes them

Hey HN, I’m Abhishek. I'm building Opslane, an open-source agent that identifies user-facing issues and investigates them. It only creates a PR if it can verify the fix. Demo: https://youtu.be/ccuOTYQMeYg Docs: https://docs.opslane.com At my last job at Robinhood, we used to do a quarterly bug bash. We would go through our Sentry backlog and try to fix as many of them as possible. We only fixed bugs we knew were reported by customers. We had hundreds of bugs, and Sentry’s default priority levels made no sense. After the bug bash, we would declare bankruptcy - select all remaining bugs and mark them as resolved. This problem has only gotten worse since coding agents have become more prevalent. So I started thinking: what would Sentry look like if it were built in 2026? To me, error trackers have two failure modes: 1. False positives: They show you thousands of errors, and you can’t tell the impact on the user 2. False negatives: Many user-facing issues don’t throw exceptions, so they go unnoticed. Opslane combines error tracking and session recording. And there is an agent that acts on both. To get started, you install the Opslane SDK. It captures everything the user did: errors, console logs, network requests, and session recordings. Opslane reduces false positives by ranking issues based on how many users are facing a particular issue. It also learns about your product by reading your code and watching your session recordings. False negatives are harder. Opslane reviews session recordings to spot frustration. They look for rage clicks, dead clicks, and abandoned forms. This recently caught a bug in an early customer’s onboarding flow: a dropdown that closed itself when clicked. No exception, no bug report. The recordings showed users clicking it, selecting nothing, and dropping out of onboarding. Opslane flagged it and the team fixed it. Three guiding principles when building Opslane: 1. Open Source: Self-host with one Docker Compose file. 2.Agent-first: I never want to open an error dashboard again. Opslane ships an MCP server, so you can ask "what broke for users this week" from Claude Code. You get back issues that need your attention and you drive the resolution. 3. It knows about your product: Opslane is continuously learning about your product. Every investigation begins with what it knows about your product. It’s early. Frontend apps work end to end today.I am currently focused on improving reliability and accuracy. Here is a link to our repo: https://github.com/opslane/opslane Would love to get feedback from folks on our approach to this problem! Comments URL: https://news.ycombinator.com/item?id=49466704 Points: 3 # Comments: 0

Hacker News 熱門83

预览 Model Hardware Standard

文章链接： https://www.anthropic.com/news/model-hardware-standard-research-preview 评论链接： https://news.ycombinator.com/item?id=49468834 得分：13 评论数：7

Hacker News 熱門27

Ask HN: What is one simple thing LLMs are insanely bad at?

I am looking for ideas on what to train a specialized model for! What is one simple thing you repeatedly ask ChatGPT, Claude, or another model to do that it still somehow messes up? Comments URL: https://news.ycombinator.com/item?id=49443820 Points: 6 # Comments: 6

arXiv cs.AI81

Recursive Experiential-Working Memory Evolution for Long-Horizon Agent Harnesses

Recursive self-improvement (RSI) remains hard in long-horizon tasks, where growing histories obscure the task state and misalign skill invocation. We introduce Recuris, a recursive Experiential-Working Memory architecture for long-horizon agent harnesses, in which Working Memory tracks task progress and guides skill selection from Experiential Memory, grounding skill use in current needs rather than the full history. This coupling also turns execution into structured evidence that localizes failures to specific memory components. Across tasks, a fixed Meta-Agent turns that evidence into localized, validation-gated updates to Skill Memory that reshape execution and yield new evidence, forming a bounded recursive memory-evolution loop. Across four long-horizon benchmarks and ten models, Recuris improves task success in 35 of the 37 completed model-benchmark pairs, carrying frontier models to SOTA-level task success: on tau-bench it adds +17.8 points to GPT-5.6 Sol and +15.6 to Claude Opus 5, taking Opus 5 to 87.9%, and +16.6/+13.5 points on Qwen3.6-27B/35B on SkillFlow. The advantage widens as the interaction horizon grows, to +32.2 points on the longest tasks, and common long-horizon failures fall by up to 80%. These results position recursively evolving memory as a scalable foundation for RSI, enabling agents to continuously transform accumulated experience into increasingly effective long-horizon behavior. Code: https://github.com/Gen-Verse/Recuris

Hacker News 熱門76

Anthropic 看到超过 30 万亿美元的潜在收入

文章链接： https://www.wsj.com/tech/ai/anthropic-expected-to-tell-investors-it-sees-over-30-trillion-in-potential-revenue-a611efea 评论链接： https://news.ycombinator.com/item?id=49436536 得分：4 # 评论数：1

Hacker News 熱門56

Show HN: Screen memory without screenshots, just text to Markdown

It's a macOS menu bar app that reads the text of your focused window every few seconds through the Accessibility API. No screenshots, no video, or OCR. It writes plain markdown, one file per day, into a folder you choose. Point Claude Code (or anything with file access) at the folder and ask "what did I work on Tuesday?" or have it build memory about your projects. The folder contains an AGENTS.md explaining the format to the model. Thought it might be useful for someone so chucked it up on GitHub Comments URL: https://news.ycombinator.com/item?id=49429095 Points: 5 # Comments: 0

arXiv cs.AI75

SWE Refactor Bench：编码智能体能完成长周期、全仓库的技术栈迁移吗？

现代软件系统在数十年的开发过程中积累了技术债，这使得迁移成本高昂且很大程度上依赖人工。随着编码智能体在错误修复方面能力日益增强，它们能否自主执行此类迁移？现有基准无法回答这个问题，因为它们只评估行为正确性，而不评估迁移是否真的发生。这导致了一种简单的取巧方式：智能体复制原始实现以让测试通过。我们称之为 Blindness。为解决这一问题，我们提出 SWE Refactor Bench，这是一个包含 20 个全仓库迁移任务的基准，覆盖 4 类技术债。一个三阶段评估协议同时衡量迁移完整性和行为正确性。(1) Migration Audit 验证迁移是否发生。(2) Behavioural Tests 使用固定测试套件衡量正确性。(3) Agentic Verification 使用 6 个独立编码智能体生成针对隐藏行为差异的定向测试。在来自 8 个前沿模型和 26 种模型-投入配置的 520 次运行中，只有 28 次（$5.4\%$）通过全部三个阶段，20 个任务中有 13 个没有获得任何被接受的解决方案，最佳模型（claude-opus-5）得分为 $47.0/100$。迁移完整性和行为正确性是不同的能力：少数运行通过跳过迁移来保持行为不变，并在 Migration Audit 阶段被拦截；大多数运行尝试迁移但破坏了行为，并在 Behavioural Tests 阶段被拦截。智能体无法交付完美迁移：在通过 Migration Audit 的 340 次运行中，$58\%$ 达到固定检查的 $99\%$，但只有 $26\%$ 达到 $100\%$。智能体能力在不同迁移类别之间存在差异：智能体在构建工具链重写上的得分为 $31.4$，但在语言重写上的得分仅为 $5.6$。综上，这些发现将 SWE Refactor Bench 定位为一个严谨的测试平台，用于开发能够可靠执行全仓库迁移的编码智能体。

arXiv cs.AI69

Specification Portability Across LLM Development Agents: Cross-Agent Compatibility in Specification-Driven Software Migration

This paper investigates cross-agent specification portability using Oracle-to-PostgreSQL migration as a controlled software transformation task. The study combines two experimental stages. First, a specification-first migration pipeline was evaluated on 1,006 PL/SQL files, of which 623 were successfully regenerated and 380 generated scripts executed successfully in PostgreSQL 16. Second, cross-agent experiments were conducted on a dataset of 1,802 Oracle scripts with corresponding PostgreSQL implementations using Amazon Kiro, Google Gemini, and GitHub Copilot, with Claude Code and Cursor included in the initial single-agent evaluation. Native and foreign specifications were assessed using Token F1, exact match, SQL syntax validity, AST exact match, AST mean similarity, and immediate runnability. The results show that specification size alone does not predict implementation quality and that cross-agent transfer can produce substantial agent-dependent degradation. The strongest replicated case occurred when Gemini directly consumed a Kiro-origin specification, producing a Token F1 of 0.035, SQL syntax validity of 2.33%, and AST mean similarity of 0.015. Rewriting substantially improved Gemini in the tested configuration, compression did not provide a universal benefit, and retrieval-augmented ingestion was the only common strategy represented on the per-agent Pareto frontiers of both Gemini and Copilot. The findings suggest that specifications in heterogeneous SDD workflows should not automatically be treated as agent-neutral artifacts and motivate explicit consideration of specification portability, agent-specific interpretation, and retrieval-based access in multi-agent software engineering.

Hacker News 熱門

Show HN: Make your logo extra bright on HDR screens

Certain logos started standing out to me on LinkedIn as brighter/whiter than everything else around them. I dug in and found out this is accomplished by adding a gain-map to an existing JPEG, visible only on HDR screens like a newer MacBook Pro. LinkedIn is the only social network I've found that isn't stripping them out, but of course you serve them up on your own site. I worked with Claude Code to turn it into a little browser-based utility (no registration) and hope you find it useful! Comments URL: https://news.ycombinator.com/item?id=49402521 Points: 18 # Comments: 15

Hacker News 熱門46

Show HN: OzBrain, a shared brain for knowledge between agents and your team

I think agent-first chat interfaces will be a primary software modality and busy dashboard/UI will go away. I’m not sure who exactly wins it, but I want my knowledge to grow/go with me. A lot of the “knowledge” ie research, analysis, reasoning will be done by agents as the primary user. Our current notes tools & tasks management systems were built for humans… I don’t care what the 17th thing on my bug backlog is. I want to conduct agents that can execute for me and do great work. What I built OzBrain to do: + Create a central place for agent reasoned knowledge to live + Be agnostic about what apps/agents connect to it + Capture everything and track it so I can audit it + Enable teams, collaborators or partners to share brains + Handle conflicts so many agents in the same article doesn’t blow up + Refactor knowledge into more token friendly chunks and map the index well + Close the knowledge loop so new thinking supersedes old thinking across the corpus. Don’t erase, depreciate and link + Keep user data safe and secure ++ Be easy enough to use that you don’t have to have any technical knowledge Some among us will always build their own custom solutions, but there are millions of tech professionals and small business owners that will use agents heavily and need a solution. So I’m trying to build that. Isn’t this like gBrain? Yes, similar. I think it’s like AWS vs Vercel. AWS is very powerful, configurable, and useful if you’re technical and want to invest the time into really fine tuning your system… but if you just want your web deploy/hosting to just work and be easy to deal with you use Vercel. // WHY I MADE IT I’ve been enjoying getting back to my technical roots, as I lost my coding skills more than a decade ago, but with AI I can focus on the system and the product in partnership with agent coding workflows. I recently built a Voice AI for older people. To build it I created an agentic engineering workflow (feel free to rip that up as I’m always looking to improve systems: https://ozbrain.com/resources/eng-flow ) My approach with coding agents is trust but verify, and I’m trying to replace the parts where a human would review with an adversarial or specialized agent who would give a better answer/review. I have workflows that will go high level task to shipped PR running in Claude cloud sessions. I use Claude Code locally and Cursor when I want a tighter loop on doing visual work like UI or layout. And Codex to either load balance usage for TokenThriffting or when I want a different llm to think thru something. It was a pain in the ass passing .md files around and keep track of which version was the most recent, so I built a hosted .md storage right in Supabase and any of my agents already have Supabase access. This let me build a solid, scalable, secure voice AI from my phone at the gym. All my agents have access to our knowledge, can write to it, update and refer to it as we build and improve the product and the systems we use. Out of 75 founder friends I asked about how they manage shared knowledge, 26 built their own custom knowledge systems… Obsidian vaults with 7k files synced through a VPS, markdown repos behind their own MCP servers, cron jobs stitching Supabase to a skills file… each a different Frankenstein they have to maintain. 32 said they felt the pain of moving static files around but didn’t have any solution for it. So I rebuilt my brain better and used it to build it. // HOW YOU CAN HELP Would love to have you try it out. The maintenance loop is still in alpha so not running it on customer data yet. If you built your own brain I’d love to hear how you did it. What criteria was most important for you in its design & function. If you are tired of shuffling .md files around I’d love to have you try out OzBrain and to give feedback, just ask your agent to put it in the shared bugs & features brain! Cheers! Bubs.co Comments URL: https://news.ycombinator.com/item?id=49394827 Points: 8 # Comments: 1

Hacker News 熱門72

Show HN: Proliferate- 适用于任何编码代理的开源、自托管 Codex

Hi HN- 我是 Pablo，Proliferate（YC S25）的创始人！ Proliferate ( https://github.com/proliferate-ai/proliferate ) 是一个开源、可自托管的 AI IDE，让你可以在一个地方使用 Claude Code、Codex、OpenCode、Cursor 和 Grok 进行工作并自动化任务。 这里有一个 2 分钟的快速演示，展示我们如何用 Proliferate 来构建 Proliferate： https://www.youtube.com/watch?v=tGNX0oaWmBY 在我的团队接入 OpenAI Codex 之后，我开始着手构建 Proliferate。几天之内，我们就把它用到了方方面面：用 computer use 代替我们自己浏览网站，让 Codex 协调其他 agent，以及为重复性工作设置自动化。说实话，我们几乎从不需要离开桌面应用就能把工作完成。 如果我团队的体验具有一定代表性，那么一个类似 Codex 的应用（一个具有精美 UI 的横向 agent）将会成为每家公司完成工作的主要界面。这完全符合 OpenAI 的使命，即把 Codex 打造成 everything app（见： https://news.ycombinator.com/item?id=47796469 ）。 但随着我们开始把自动化推进到更接近业务核心的工作，我希望能与来自所有实验室的 agent 协作，包括开放权重模型，而不会越来越依赖 OpenAI。 而这正是 Proliferate 的用途！它是开源、可自托管的 Codex，既能在构建向 Codex 广度迈进的同时，保留你在不同 agent 和模型提供方之间的选择自由。 如今 Proliferate 支持： * 使用 Claude Code、Codex、OpenCode、Cursor 和 Grok，并可采用它们原生的推理与配置选项，包括 Bedrock、Azure 和自托管推理。 * agent 之间的通信与管理：父 agent 可以生成并与另一个受支持的 agent 作为子 agent 进行通信（我个人喜欢让 Fable 委派给 Codex，同时让 OpenCode 模型审查 PR）。 * 构建 workflows——这是我最兴奋的功能之一。这些是可复用的 agent 会话链和人工审批关卡，每一步都可选择 harness 和模型，并在它们之间传递文档。我用它来自动化代码审查、QA，以及我的 PR 构建流程。 Proliferate 的全部代码在 AGPL-3.0 下 100% 开源。现在肯定还有一些粗糙的地方，但我们推进得很快，也非常欢迎任何反馈！ Comments URL: https://news.ycombinator.com/item?id=49390739 Points: 11 # Comments: 5

Hacker News 熱門45

Show HN: A desktop fly drawn to the scent of vibecode

It is a fork of https://github.com/DenisSergeevitch/desktop-fly , but with an important update. Now the fly can pick up the scent of the codebase with its neurons and fly straight to the source code of your B2B AI SaaS startup. It has learned to scan its surroundings for agent markers: AGENTS.md, CLAUDE.md, .cursor/rules, .kiro/steering, and forty others. Anything on the screen that points to these markers becomes a source of the scent - an editor window with an open project, a line in Finder, or a desktop icon. An open project reeks the strongest, while a closed icon or nested folders give off a fainter odor. Comments URL: https://news.ycombinator.com/item?id=49389466 Points: 3 # Comments: 0

Hacker News 熱門59

Show HN: Frugal Tokens – explore costs and usage across coding agents

I wanted to share a project I’ve been working on called Frugal Tokens. I originally built it because I was curious to see how much all of my sessions cost and how much cache misses affected that spend. I’d noticed people had widely different spend profiles and wanted to better understand what might contribute to that. As I’ve worked on this, the tool has grown to show more usage patterns across all of your sessions. It shows overall usage, estimated working time and overlapping sessions, and where your spend is coming from across models and cache misses. I also have a few session level metrics with percentile breakdowns, along with a list of your sessions and high level info. Clicking into a session opens an explorer where you can see individual model calls and tool inputs and outputs. You can also jump directly to where a cache miss happened. There’s also a rough cost comparison that shows what the recorded session would have cost with another model’s pricing, or for Anthropic, with 5m vs 1h caching. In the future, I’d love to collect more information to see which patterns might make people’s workflows more expensive, e.g. long sessions, high context usage, many turns, etc. The tool requires deno, but is just one command to run once that is installed. The demo provided has some of the data scrubbed, but helps to show what it looks like before running it. Would appreciate any thoughts or ideas https://github.com/dpclark4/frugal-tokens Comments URL: https://news.ycombinator.com/item?id=49364223 Points: 10 # Comments: 1

Hacker News 熱門

Show HN: Automatically detect and patch walking-dead states in Sierra games

Hi HN, I've become lazier in my old age and struggle to replay my favorite Sierra games from the 80s and 90s because I keep getting into those situations where I need an item from 3 acts ago, I have no save game handy, and now I gotta make dinner. So I'm building the Lucasartsifier: a static analysis tool that decompiles Sierra resource files, automatically finds those states, automatically generates code to prevent the player from getting into those states, then emits loose patch files that can be placed alongside the original game resources. There's no game-specific code involved; all the logic is generic, though of course Sierra introduces new idioms and mechanics in every game so every new supported game needs a bunch of engine work. So for example in Leisure Suit Larry 2, the patched game prevents you from boarding the cruise ship until you have both the sunscreen and the Grotesque Gulp. Without them you die on the raft 3 play-hours later. So far this works on Leisure Suit Larry 2 (SCI0), King's Quest 4 (SCI0), King's Quest 6 (SCI1.1), and Laura Bow 2 (SCI1.1). I'm currently working on King's Quest 5 (SCI1.0). This is work done with Claude -- I do the design and playtesting and it does the rest :D Any feedback, play testing, and suggestions would be great! Comments URL: https://news.ycombinator.com/item?id=49355607 Points: 25 # Comments: 11

Hacker News 熱門77

Claude Code 每周限额明天减少三分之一

文章 URL： https://support.claude.com/en/articles/15910845-claude-code-may-august-2026-weekly-limits-promotion 评论 URL： https://news.ycombinator.com/item?id=49348751 积分：44 评论数：15

Hacker News 熱門71

发布 HN：machine0 (YC S26) – 来自 CLI 的持久 CPU 和 GPU VM

HN 大家好！我是 Barnaby，machine0（ https://machine0.io ）的创始人。我正在构建一个用于长周期代理计算的 CLI：`machine0 new mybox` 会为你的代理提供一台持久的云端 VM，按分钟计费，价格从 $0.013/hr 起，最高可达 60 vCPU / 240 GB RAM，并支持 GPU（H100s、H200s 等），VM 级别正常运行时间为 99.99%。代理可通过 CLI 或 MCP 自主运行。 演示： https://www.youtube.com/watch?v=gyllkZ0M04E 代理工作负载正在从短暂运行转向始终在线。一个处理复杂功能的编码代理会运行 6-8 小时。由代理编排的训练和 RL 运行需要数天。OpenClaw 和 Hermes 24/7 运行。随着你并行运行更多任务： - 资源：大型代码库上的几个代理就会耗尽 RAM 和 CPU。模型训练和 RL 需要你没有的 GPU。 - 安全：在你的个人机器上使用 `--yolo`，只需要一次提示注入就可能导致凭证被外泄。 - 可用性：合上笔记本电脑，代理就会在任务中途死掉。 - 隔离：你和你的代理实际最低需求之间没有清晰边界。 machine0 为每个代理提供自己的计算机。它是一个足够简单的 CLI，人类和代理无需阅读文档即可使用： `machine0 new mybox` 会创建一台可 SSH 连接、带静态 IP 和 HTTPS 端点的 VM。始终在线（VM 级别正常运行时间 99.99%），直到你将其关闭。 - 按分钟计费。1 vCPU / 1 GB 为 $0.013/hr，最高可达 60 vCPU / 240 GB，另有从 RTX 4000 Ada 到 8×H200 的 GPU。 - 暂停、快照和恢复。让你可以轻松暂停工作，之后再回来继续。也可以制作一个黄金主镜像，用于为机群批量生成克隆。 - 块存储。持久卷（从 10 GB 到 16 TB），你可以用直观的语法管理：`--yolo` 并将其挂载到你的 VM。 - Profiles。凭证、MCP 连接、提示词和环境变量的组合包，在 VM 创建时注入。因此每个代理只获得你选择的能力，除此之外什么都没有。 - 代理自助服务。把 CLI 或 MCP 服务器交给 Claude、Codex 或 OpenCode，它就能管理自己的机群：为一次构建启动一台机器，创建快照，然后销毁。 - 可复现构建。使用 NixOS flakes 或搭配 Ubuntu 的 Ansible playbooks。 人们今天如何使用它？ - 代理机群。人们运行一个试点代理来界定工作范围，并将其委派给子代理，每个子代理都在自己的 VM 上：用试点代理规划项目，工作代理实现它并提交 PR。一个客户一次运行数百台机器，通过 CLI 启动和销毁。 - 模型优化和 RL 环境。ML 团队使用 machine0 来进行由代理编排的 RL 环境和模型优化工作。一个客户在 60 vCPU 机器上运行 RL 环境，一次持续运行数天；另一个客户保留一台已暂停的 H100，并在夜间让代理使用它来反复优化推理速度。 - 产品基础设施。一个客户不是自己使用 machine0，而是将他们的产品构建在 machine0 之上：每个用户会话都会从他们自己代理运行时的版本化镜像中获得一台全新的 XL 机器。他们已经发布了数百个该镜像版本，并启动了数千台机器，其中大多数只存活两分钟。 底层是什么？ 每台机器都是完整的 KVM 虚拟机，不是容器或沙箱。你获得的是暴露给 guest 的真实 GPU，带有其实际驱动；可获得内核级访问权限（加载你想要的任何模块或驱动）；你和硬件之间没有 syscall 拦截层。技术栈本身刻意保持朴素：TypeScript、Postgres、Redis。我们非常重视安全性、可靠性和性能，使 machine0 非常适合持续的计算密集型工作负载。 关于我 我从事云基础设施建设大约 15 年了。我曾在 Imperial College London 攻读云资源分配方向的博士，后来退学；之后作为 Upflow (YC W20) 的联合创始人兼 CTO 工作了六年，在 ARR 达到七位数的整个过程中亲自负责 DevOps、基础设施和安全，因为这些事情风险太高，不能委派。machine0 最初是我为自己做的工具，我就是自己的第一个用户 :) 请求 希望你能试用并给我们反馈（见下方）。或者，如果你是一家公司，正在为软件工厂、模型训练或 RL 环境寻找计算资源，欢迎联系 barnaby@machine0.io # 安装 machine0 $ curl -LsSf https://machine0.io/install.sh | sh # 创建一台机器并通过 ssh 登录 $ machine0 new myvm $ machine0 ssh myvm 评论 URL： https://news.ycombinator.com/item?id=49348136 点数：3 评论数：0

Hacker News 熱門70

Claude：多个模型性能下降

文章 URL： https://status.claude.com/incidents/q7txxvbsftgq 评论 URL： https://news.ycombinator.com/item?id=49348163 得分：12 评论数：2

arXiv cs.AI73

ClawGym II：探索 Agent Harness 上的黑盒 RL

Agent harnesses 通过协调 agent 与环境的交互，显著提升了长时程任务的性能。然而，通过复杂 harnesses 进行强化学习在很大程度上仍未被探索，因为将此类训练扩展到长时程 agent 任务会带来根本性挑战。在这项工作中，我们提出了一个统一的黑盒 RL 框架，用于通过复杂 harnesses 对通用 agents 进行稳定且可扩展的优化。具体而言，我们首先构建了基于沙箱的执行基础设施，将任务环境和 harnesses 隔离在临时沙箱中，以支持大规模并发 rollouts。随后，我们将策略优化与不透明的 harness 执行解耦，并在模型边界放置 serving proxy 以捕获模型调用。为重建多轮轨迹并提高训练效率，我们将捕获的调用组织成前缀树，并进一步调整基于 critic 的 PPO 和无 critic 的 GRPO，使其在恢复出的树结构上进行优化。与此同时，我们在整个优化过程中保持训练-推理一致性。最后，我们引入 mix-harness training，使单个模型能够通过异构 harnesses 进行联合优化。使用 Qwen3-30A3B 时，黑盒 RL 通过 OpenClaw 和 Claude Code 分别将 ClawGym-Bench 上的 Pass@1 提升了 9.98 和 14.81 个点，同时在 200-400 个优化步骤内保持稳定。此外，该框架在更具挑战性的任务（如 JobBench 和 OfficeQA）上也带来了一致的增益。总体而言，我们的框架使得通过黑盒 harnesses 对通用 agents 进行有效、稳定且可扩展的优化成为可能，并支持跨异构执行系统的统一训练。

Hacker News 熱門61

Show HN: 1667, a terminal UI for writing fiction with language models

Hi HN. I built 1667 for my own fiction work and now use it each day. This probably has a limited audience. Maybe an audience of one... Why a terminal interface for story writing? I'm a dev. I like to use terminals for a lot of stuff. Most WebUIs feel off to me. That's the only reason. One thing that bothers me about writing in existing tools is that they don't fit the way I write. The mental model of my story is a tree. I try many takes usually continue with just one, but sometimes I want to try an alternate route and see where this goes. And that can branch again in many places. See what happens if I kill off this character or they don't take the job or whatever. 1667 is a full-screen terminal app for long-form fiction. Each story part can have several takes. All takes stay in a tree. You select one path through that tree as the story line. Export writes that line to Markdown in the project folder. Some technical details: - A project stores its stories and settings in a `.1667/` directory. Exported Markdown sits beside it. - Provider secrets stay in private machine files. Requests go to the provider that the writer selects. - An optional Vault Password seals project files at rest. - An operating-system lock permits one writer process for each project. - The request viewer shows the next provider request without its credential. - Each generated take keeps a Generation Record with its model and effective settings. Version 0.9.5 runs on macOS, Linux, and Windows x64. The website has Shell and PowerShell installers. An npm package is also available. 1667 imports Markdown, SillyTavern chats and cards, and NovelAI archives. It can use OpenAI-compatible, Anthropic, and local endpoints such as Ollama, LM Studio, llama.cpp, and KoboldCpp. Current limits: the release is pre-1.0. The interface is a terminal. There is no account or cloud sync, and I don't plan to add any. No tracking. Comments URL: https://news.ycombinator.com/item?id=49330604 Points: 12 # Comments: 12

Hacker News 熱門53

Claude Seems Down

Throwing "Authentication service was unavailable.", no updates on their status page, but there are on downdetector Comments URL: https://news.ycombinator.com/item?id=49324078 Points: 17 # Comments: 8

Hacker News 熱門78

Claude：系统提示词

文章 URL： https://platform.claude.com/docs/en/release-notes/system-prompts 评论 URL： https://news.ycombinator.com/item?id=49319556 得分：10 评论数：0

Hacker News 熱門73

最大化你的 Claude Code 会话价值

文章 URL： https://claude.com/blog/maximizing-the-value-of-your-claude-code-sessions 评论 URL： https://news.ycombinator.com/item?id=49300800 积分：169 评论数：106

arXiv cs.AI69

AutoDesign: Meta-Harness Optimization for Long-Horizon Agentic Design

Transforming multimodal sources into condensed and structured media outputs can be fundamentally conceptualized as a long-horizon agentic process centered on a model-harness system. While an ideal harness system should align with human design priors and accumulate reusable experience through empirical exploration to drive recursive self-improvement, existing paradigms remain static and fall short of this capability. In this paper, we present AutoDesign, a framework that aligns with human design priors, where a meta-harness optimizer guides a code agent to recursively improve harness based on rollout feedback. To instantiate and evaluate this framework, we focus on the academic paper-to-poster generation task and introduce PosterBench, comprising a 100-paper Main Track spanning five disciplines and PosterBench-mini, a shared 10-paper subset for controlled evaluation. On the PosterBench Main Track, AutoDesign achieves the highest score of 78.32, surpassing the closed-source commercial system Claude Design by 7.45 points. Across seven controlled code-agent-model configurations, integrating the learned DesignHarness consistently improves performance, increasing the average PosterBench Score from 54.99 to 67.39 (+12.4%). In a fully autonomous long-horizon loop, it executes 253 tool calls and 11 editing turns within 40 minutes for under $3, reaching average conference-poster quality in human evaluation. A system-blind human study further demonstrates that AutoDesign achieves the highest human preference among evaluated systems.

OpenAI Blog68

Frontier Model Forum 更新

我们与 Anthropic、Google 和 Microsoft 共同宣布 Frontier Model Forum 的新任执行主任，以及一项新的 1000 万美元 AI Safety Fund。

OpenAI Blog81

OpenAI 和 Anthropic 分享联合安全评估的发现

OpenAI 和 Anthropic 分享了首次同类联合安全评估的发现，彼此测试对方模型在目标偏离、指令遵循、幻觉、越狱等方面的表现——凸显了进展、挑战以及跨实验室协作的价值。