https://huhohoo.com/zh-TWHohoo / AI Radarzh-TWhttps://developer.nvidia.com/blog/run-ai-models-for-edge-application-development-with-just-a-few-commandscfb4189f31938d74a87fAI is expanding beyond the data center as developers build a new generation of agents to act at the edge: monitoring driver attention and cabin conditions,... AI is expanding be… · 來源發布: 2026-10-30T20:33:18.000Z · 本站收錄: 2026-10-31T00:00:35.801ZFri, 30 Oct 2026 20:33:18 GMThttps://aihot.news/items/cms82puwtmpssdpbjmajoopr95818be28c4f036b0f020Philipp Schmid 转引 Sundar Pichai 的消息：Google Project Suncatcher 在数周前发射原型卫星后，成功向轨道发送指令、启动 TPU、运行提示词，并在地面收到了首个完整回应（一首 haiku），作者称 Gemma 4 成为首个在太空运行的大语言模型。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-30T18:28:00.000Z · 本站收錄: 2026-10-31T00:00:35.801ZFri, 30 Oct 2026 18:28:00 GMThttps://developer.nvidia.com/blog/build-specialized-agents-on-a-stack-you-control-with-nvidia-nemoclawb0afceb347e7b188bf46Building an autonomous agent requires a model, a harness, and domain-specific tools. To deploy an agent more securely, developers need a runtime that enforces... Building an aut… · 來源發布: 2026-10-30T16:00:00.000Z · 本站收錄: 2026-10-30T19:39:03.173ZFri, 30 Oct 2026 16:00:00 GMThttps://openai.com/index/helping-websites-express-ai-training-preferences84abd1857de5dfd951d9OpenAI will honor a shared robots.txt signal to help website operators communicate their AI training preferences to participating providers. · 來源發布: 2026-10-30T09:30:00.000Z · 本站收錄: 2026-10-30T19:39:03.173ZFri, 30 Oct 2026 09:30:00 GMThttps://aihot.news/items/iu7xvjx1c2c5wqrpn0rwfxcb5f4829fe692295bb3142eCursor 发现通过其 IDE 扩展市场分发的 Claude Code VS Code 扩展被修改并植入恶意软件，最多影响 5000 用户，恶意版本并非 Anthropic 开发。Cursor 已于当日下午下线扩展市场，与相关提供方合作清除恶意软件并确认无进一步风险后重新上线，建议受影响用户参考博客说明排查，并升级到 Claude Code for … · 原公告: 日期未核實 · 聚合源發布: 2026-10-30T07:46:39.000Z · 本站收錄: 2026-10-30T13:28:40.731ZFri, 30 Oct 2026 07:46:39 GMThttps://arxiv.org/abs/2610.3959001f3dd12554851c88ddbAs large language models (LLMs) become more widespread in Africa, their safety in low-resource languages remains largely untested since most red-teaming work focuses on English.… · 來源發布: 2026-10-30T04:00:00.000Z · 本站收錄: 2026-10-30T13:28:40.731ZFri, 30 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.398540236301ead3e367db1f9Large Language Models (LLMs) rely on Multi-Head Attention (MHA) for sequence modeling, but scaling to long contexts is constrained by the quadratic cost of attention, motivating… · 來源發布: 2026-10-30T04:00:00.000Z · 本站收錄: 2026-10-30T13:28:40.731ZFri, 30 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.390540283c17ff41feb9fac1aModern robot learning systems feed on high-quality demonstration data, and their performance hinges on coverage. Without good coverage, small errors compound and the policy can … · 來源發布: 2026-10-30T04:00:00.000Z · 本站收錄: 2026-10-30T13:28:40.731ZFri, 30 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.40285048013bc2f09cfe74a79Musculoskeletal robots combine compliant actuators with articulated skeletons, but exploring alternative configurations often requires repeated fabrication. We present MUSBRIC, … · 來源發布: 2026-10-30T04:00:00.000Z · 本站收錄: 2026-10-30T13:28:40.731ZFri, 30 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.40632055344d7275986f5bef3As robots and intelligent systems increasingly interact with humans, they must anticipate long sequences of future actions from only brief observations. While purely neural mode… · 來源發布: 2026-10-30T04:00:00.000Z · 本站收錄: 2026-10-31T00:00:35.801ZFri, 30 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.402270970bbd36efcf4c1ac90Epistemic game theory explains an agent's strategic behavior via the beliefs it holds about others and how these beliefs are updated as the interactions unfold. Modern LLM-based… · 來源發布: 2026-10-30T04:00:00.000Z · 本站收錄: 2026-10-31T00:00:35.801ZFri, 30 Oct 2026 04:00:00 GMThttps://aihot.news/items/z9kqlhsgiz11fvlk3aodm6seg14a3b8b5c9a4350fbd1bFigma 介绍其为 agent 自研 Figma 编辑模型的训练方法，先用超过 25 万条合成样本做 SFT，让早期 Llama 3.3 70B 在编辑任务上接近当时顶级模型，后转向 RL 并结合确定性校验器与视觉评委做评测。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-29T21:00:00.000Z · 本站收錄: 2026-10-30T00:16:25.071ZThu, 29 Oct 2026 21:00:00 GMThttps://simonwillison.net/2026/Oct/29/how-i-built-the-most-ai-pilled-company-in-the-worlde65fcafff408f8083fa1How I Built the Most AI-Pilled Company in the World Don't be put off by the clickbait headline, this piece by Jacob Sansbury describing how they are building Spawn is a fascinat… · 來源發布: 2026-10-29T17:00:26.000Z · 本站收錄: 2026-10-30T00:16:25.071ZThu, 29 Oct 2026 17:00:26 GMThttps://aihot.news/items/nb40aoihhydhoo0acp0rcyaol2d2e9705cafefad7b5fd腾讯混元发布 Hy-Nomas 开源交互式视频世界模型的技术报告（arXiv:2610.36729），并开源 Hy-Nomas-Bench 数据集用于可复现研究。官方还用 Hy-Nomas 重现 Genie 4 示例并放出对比视频，演示无需排队即可试用，地址为 https://hy-nomas.hy.tencent.com/。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-29T17:00:03.000Z · 本站收錄: 2026-10-29T19:49:35.843ZThu, 29 Oct 2026 17:00:03 GMThttps://aihot.news/items/x6x8nkxn5mqc55umit3clj1tf27846bbfac3ce01ee1e2vLLM 团队宣布对 Mistral 新旗舰模型 Mistral Large 4 提供 Day 0 支持，该模型为 1T 参数、每 token 激活 49B 的原生多模态 MoE 模型，采用 MLA 稀疏注意力加 DSA top-k 索引器、跨层 KV 共享、NoPE、SSMax 和 LeRoPE 等新架构，并内置 DSpark 推测解码草稿器与 Pi… · 原公告: 日期未核實 · 聚合源發布: 2026-10-29T00:00:00.000Z · 本站收錄: 2026-10-29T19:49:35.843ZThu, 29 Oct 2026 00:00:00 GMThttps://aihot.news/items/l928e49lwysm0ymgmhyjrkc3z8825148d53503e8df066OpenRouter 发布教程，讲解如何把 PDF 以文件形式连同严格 JSON schema 发给能读文件的模型，让表格以带类型的行返回，括号亏损、千分位和脚注标记等清理都写进字段描述。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-29T00:00:00.000Z · 本站收錄: 2026-10-29T06:24:00.528ZThu, 29 Oct 2026 00:00:00 GMThttps://aihot.news/items/gjz7lj8j6ojjyj80vkkhy786vcceba212171c1558a424Factory 发布 Droid 1.0，一个模型无关的 agent harness，主打程序化工具调用（PTC），模型通过编写 JavaScript 程序在沙箱中一次执行数百个工具调用。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-29T00:00:00.000Z · 本站收錄: 2026-10-30T00:16:25.071ZThu, 29 Oct 2026 00:00:00 GMThttps://aihot.news/items/fbxzb7x7fzbjn0dmf367ywxrm088d80c954f3b98b9715Boris Cherny 介绍犹他州向 10 个州政府部门的 34 名开发者提供了 Claude Code。试点显示 80% 的开发者在 30 分钟内搭好可用环境，71% 的开发者在为期四周的试点结束前报告节省了 20 小时以上时间。该推文引用了 Anthropic 与 GSA 合作将 Claude 以 40% 折扣提供给联邦及各级政府的公告作为背景。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-28T17:09:54.000Z · 本站收錄: 2026-10-28T22:49:32.440ZWed, 28 Oct 2026 17:09:54 GMThttps://aihot.news/items/tdp7qmrh74fdfarxbi4jaoc6v280fc51136c04b796c84Microsoft 宣布将 GitHub 团队打造的 Copilot harness 通过 Microsoft Foundry 向开发者和企业开放，该 harness 已驱动 Copilot Code、Copilot Studio、GitHub Copilot app 和 CLI 等产品。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-28T16:51:35.000Z · 本站收錄: 2026-10-28T22:49:32.440ZWed, 28 Oct 2026 16:51:35 GMThttps://simonwillison.net/2026/Oct/28/github-octoverse-20265e9e44475572a25994c0GitHub Octoverse 2026 I'm at GitHub Universe today (GitHub gave me a free ticket through their "GitHub Stars" program). EVP Jay Parikh just announced this year's Octoverse repor… · 來源發布: 2026-10-28T16:45:23.000Z · 本站收錄: 2026-10-29T06:24:00.528ZWed, 28 Oct 2026 16:45:23 GMThttps://aihot.news/items/xonc43xgcx39303i40k8022rl40d40d4a66fef630d5f8GitHub 在 GitHub Universe 2026 上提出把 GitHub 打造为完整软件交付系统的愿景，并发布多项产品：GitHub Copilot app 面向智能体交付。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-28T16:09:28.000Z · 本站收錄: 2026-10-28T16:48:23.205ZWed, 28 Oct 2026 16:09:28 GMThttps://aihot.news/items/s3m3uzabqoxxed9ae50b73qvcaad5f298a95cf5bc9613GitHub 发布 2026 年度 Octoverse 报告：过去 12 个月开发者推送代码 179 亿次，超过 2024 和 2025 年之和，合并 PR 达 13 亿次。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-28T16:07:40.000Z · 本站收錄: 2026-10-29T06:24:00.528ZWed, 28 Oct 2026 16:07:40 GMThttps://simonwillison.net/2026/Oct/28/vibe-game-development70f2bbbd704a353daaa0I've written before about how November 2025 was an inflection point for coding agents - the point where Opus 4.5 and GPT-5.1 crossed a tipping point that meant their respective … · 來源發布: 2026-10-28T13:55:15.000Z · 本站收錄: 2026-10-29T06:24:00.528ZWed, 28 Oct 2026 13:55:15 GMThttps://aihot.news/items/tw5fe5mj6a2zyp6ica2ec3qpbb669157b1e769dffcb49Addy Osmani 发布 Claude Code Mods 上手指南，说明 mods 是运行在会话中的 JavaScript/TypeScript 文件，可通过 hooks 观察事件、改写行为或渲染自定义界面。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-28T06:15:43.000Z · 本站收錄: 2026-10-28T08:56:38.404ZWed, 28 Oct 2026 06:15:43 GMThttps://arxiv.org/abs/2610.37276010a5341a3acc99aab0bLarge language model (LLM)-driven agent scaffold evolution improves agent workflows through iterative candidate generation and evaluation. Evaluation both guides subsequent sear… · 來源發布: 2026-10-28T04:00:00.000Z · 本站收錄: 2026-10-28T08:56:38.404ZWed, 28 Oct 2026 04:00:00 GMThttps://aihot.news/items/wgss1ksoafvu62x799sdfs0fof4ba9d732d063fef85d3OpenRouter 介绍 Claude Code 触发用量上限时改用 OpenRouter API key 继续编码的方法：设置三个环境变量或用 Ori Harness 的 ori claude 命令，即可按 token 付费使用同样的 Claude 模型，没有五小时和每周限制。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-28T00:00:00.000Z · 本站收錄: 2026-10-28T16:48:23.205ZWed, 28 Oct 2026 00:00:00 GMThttps://aihot.news/items/o5s4s2z9njz29hgd7im06gzrfb893fff35c9fbf7aca1c作者称 Verified AI 迎来标志性月份，Axiom 的 AI 智能体集群用 24 天完成数学史上最长证明的验证，将跨越 500 多篇论文的 15,000 页有限单群分类定理（CFSG）转化为 3800 万行以上机器验证代码，并发现已发表记录中的 34 处严重出入。引用内容补充称该形式化证明可能是最大的开源项目之一，图中显示 Axiom CFSG… · 原公告: 日期未核實 · 聚合源發布: 2026-10-27T23:31:32.000Z · 本站收錄: 2026-10-28T01:38:43.599ZTue, 27 Oct 2026 23:31:32 GMThttps://simonwillison.net/2026/Oct/27/game-boy-color-on-my-conference-badge68b12007f7f448554e3bGame Boy Color on My Conference Badge I'm at GitHub Universe this week and, similar to last year , the conference badge is a full-blown color screen computer. I've been hacking … · 來源發布: 2026-10-27T19:34:18.000Z · 本站收錄: 2026-10-27T20:55:38.095ZTue, 27 Oct 2026 19:34:18 GMThttps://developer.nvidia.com/blog/fine-tuning-nvidia-nemotron-to-know-local-rules-not-just-local-languages5ed14fffd1748be81239As AI adoption expands globally, people across languages and cultures are using it in more aspects of their daily lives. General-purpose language models serve... As AI adoption … · 來源發布: 2026-10-27T16:00:00.000Z · 本站收錄: 2026-10-27T20:55:38.095ZTue, 27 Oct 2026 16:00:00 GMThttps://openai.com/index/retell-ai1e3ebf94e41ebe19793bGPT-6 Astra tops Retell’s voice agent benchmark, while Codex helps it reach a $1.6M annual revenue run rate per employee. · 來源發布: 2026-10-27T15:00:00.000Z · 本站收錄: 2026-10-28T01:38:43.599ZTue, 27 Oct 2026 15:00:00 GMThttps://developer.nvidia.com/blog/3-ways-to-boost-moe-inference-on-nvidia-dgx-sparkb836eb939c37ffa509e8Generating tokens faster starts with identifying what limits inference performance. In the batch-size-1 evaluation of Ling-3.0-flash from Inclusion AI on NVIDIA... Generating to… · 來源發布: 2026-10-27T15:00:00.000Z · 本站收錄: 2026-10-27T20:55:38.095ZTue, 27 Oct 2026 15:00:00 GMThttps://aihot.news/items/p92a4pkxkheznj9nmt50o08l6d7fd07ad12ad39e2fc61Ai2 发布研究 Old Ideas, Novel Problems、配套 benchmark、数据、代码和自动化流水线，测试 LLM 作为研究想法新颖性裁判的可靠性。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-27T08:00:00.000Z · 本站收錄: 2026-10-27T20:55:38.095ZTue, 27 Oct 2026 08:00:00 GMThttps://arxiv.org/abs/2610.33584004707b31725282b6888Cooperative transport with heterogeneous legged robots requires coordinating different motion capabilities while supporting an unfixed shared payload. We present HA-DiPO (Hetero… · 來源發布: 2026-10-27T04:00:00.000Z · 本站收錄: 2026-10-27T08:21:06.874ZTue, 27 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.34182024f6d16934a5ea43d97Supervised Fine-Tuning (SFT) of LLMs suffers from catastrophic forgetting, and prevailing mitigation strategies struggle to reconcile the plasticity-stability dilemma. To addres… · 來源發布: 2026-10-27T04:00:00.000Z · 本站收錄: 2026-10-27T08:21:06.874ZTue, 27 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.3404103cf979048061c310930Dynamic and continuously evolving environments present significant challenges for robotic mapping and scene understanding, particularly on construction sites where moving agents… · 來源發布: 2026-10-27T04:00:00.000Z · 本站收錄: 2026-10-28T01:38:43.599ZTue, 27 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.34991074545dfb6930b891e63End-to-end full-duplex spoken dialogue models enable natural, real-time spoken interactions with interruptions, backchannels, and overlapping speech. Prior evaluations have asse… · 來源發布: 2026-10-27T04:00:00.000Z · 本站收錄: 2026-10-28T01:38:43.599ZTue, 27 Oct 2026 04:00:00 GMThttps://openai.com/index/greptile-code-review-codex-astraa9c488820a62ac4adcd9How Greptile uses GPT-6 Astra to test user workflows, Daybreak to strengthen security, and Codex to reduce rework. · 來源發布: 2026-10-26T22:00:00.000Z · 本站收錄: 2026-10-27T01:36:23.463ZMon, 26 Oct 2026 22:00:00 GMThttps://aihot.news/items/d0k8x9588e62wlqnfoq3bmtq05847634950f86068d3c3Anthropic 发表研究，让 50 名有因果推断经验的专业分析师用 Claude Code（Claude Opus 4.8）完成一项统计分析任务，并在分析师不知情的情况下用掷硬币随机设定智能体在四个分析决策上的倾向。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-26T16:00:00.000Z · 本站收錄: 2026-10-27T20:55:38.095ZMon, 26 Oct 2026 16:00:00 GMThttps://openai.com/index/openai-cut-incident-investigation742d30efdc4090c11a97See how OpenAI engineers use Codex Ultrafast to investigate incidents in minutes. Explore faster security analysis, code refactoring, and daily workflows. · 來源發布: 2026-10-26T07:00:00.000Z · 本站收錄: 2026-10-26T21:10:16.237ZMon, 26 Oct 2026 07:00:00 GMThttps://aihot.news/items/o1r046h1yurwlvv536eergn579fd77d4e3cc2a617a548数字生命卡兹克分享用 AI 逆向改造现实设备的实操教程，以公司 4 台会议室电视为例，通过 Codex 开启 ADB 调试连接电视，清理开机广告、重做简洁桌面和粒子动态屏保，并复刻到其余 3 台。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-26T04:48:25.000Z · 本站收錄: 2026-10-26T08:04:55.166ZMon, 26 Oct 2026 04:48:25 GMThttps://arxiv.org/abs/2610.3192903134d1bee533cd209b9Physical self-reproduction in modular robots remains difficult because a robot must not only move, but also collect, align, dock, and separate modules reliably in hardware. Taki… · 來源發布: 2026-10-26T04:00:00.000Z · 本站收錄: 2026-10-26T08:04:55.166ZMon, 26 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2606.27457034157200a987be8a0ecServing open-weight language models on premises forces a trade-off between accuracy and computational cost. Hosting models of different sizes lets easy queries go to a small, ef… · 來源發布: 2026-10-26T04:00:00.000Z · 本站收錄: 2026-10-26T08:04:55.166ZMon, 26 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.31834089ab686357791556a56Existing approaches for robot learning in simulation often require substantial manual effort to construct scenes and specify tasks, limiting training scalability and data divers… · 來源發布: 2026-10-26T04:00:00.000Z · 本站收錄: 2026-10-26T08:04:55.166ZMon, 26 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2605.113880a51578fa0575756342eHumans intuitively solve complex problems by flexibly shifting among reasoning modes: they plan, execute, revise intermediate goals, resolve ambiguity through associative judgme… · 來源發布: 2026-10-26T04:00:00.000Z · 本站收錄: 2026-10-26T08:04:55.166ZMon, 26 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2607.053690b7de25d626af2fae563For robots to work reliably in commercial and industrial applications, recent advances in agentic coding systems have potential to combine interpretable robot programming with t… · 來源發布: 2026-10-26T04:00:00.000Z · 本站收錄: 2026-10-27T01:36:23.463ZMon, 26 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.091631280de4aaf39ea27d7e4Task-oriented conversational agents remain fragile under real world conversation scenarios as they rarely follow a predictable script, especially when users exhibit non-cooperat… · 來源發布: 2026-10-26T04:00:00.000Z · 本站收錄: 2026-10-27T01:36:23.463ZMon, 26 Oct 2026 04:00:00 GMThttps://aihot.news/items/w5optivkpl1pchxn85qaknu3d4ea570db7f0502080577英国 AI Security Institute 与 SecureBio 合作推出 ReproBAIT 评测，测试 AI 编码智能体能否依据论文等公开信息和训练数据复现生物 AI 工具（BAITs）。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-25T23:00:00.000Z · 本站收錄: 2026-10-26T15:26:23.998ZSun, 25 Oct 2026 23:00:00 GMThttps://aihot.news/items/xm6i4tcgyby0bditot3bz9x8x8e9bc00fdc8cc647568e英国 AI Security Institute 的 Chem-Bio 团队从零预训练 8M 到 3B 参数的 ESM-2 蛋白质语言模型，测试预训练数据过滤对专业科学模型的效果。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-25T23:00:00.000Z · 本站收錄: 2026-10-26T15:26:23.998ZSun, 25 Oct 2026 23:00:00 GMThttps://aihot.news/items/j5auxb780hddyh0hlgekuw03h96355be23c04c1bbc980Browser Use 作者整理了让 Claude Code、Codex、Gemini CLI、Cursor 等编程智能体使用已登录 Chrome 的方法：Claude 订阅用户可用 claude --chrome 的 Claude in Chrome 扩展。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-25T00:00:00.000Z · 本站收錄: 2026-10-26T00:41:26.556ZSun, 25 Oct 2026 00:00:00 GMThttps://aihot.news/items/d92mrmxvsfmnoyw3comncgz2c8b9cd9b5d2adbb9e2d44Ethan Mollick 转引 OpenAI 红队测试认为，实验室看似夸张的 AI 网络安全判断应被更认真对待。OpenAI 让前沿模型智能体团队攻击其加固的隔离基础设施，智能体几乎在每个受限联网服务中都找到漏洞，还能利用 CPU 缓存时序在不联网、无共享文件的虚拟机之间传递秘密，类似 Spectre 和 Meltdown 的攻击手法。作者提到 An… · 原公告: 日期未核實 · 聚合源發布: 2026-10-24T14:30:45.000Z · 本站收錄: 2026-10-24T16:13:05.225ZSat, 24 Oct 2026 14:30:45 GMThttps://aihot.news/items/imgp55zxmw4jugt1a29ap9yo125f8636014f00ad326baOpenAI 介绍其前沿模型训练已完全断网，并用 AI 智能体红队测试自家基础设施。隔离测试中，智能体在无网络、无共享文件的条件下通过 CPU 缓存时序差异、共享内存写入模式等侧信道通信，甚至独立建立编码方案交换信息。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-24T11:12:36.000Z · 本站收錄: 2026-10-24T16:13:05.225ZSat, 24 Oct 2026 11:12:36 GMThttps://arxiv.org/abs/2604.161580652e0d51483e0cff10dLarge language models (LLMs) increasingly rely on chain-of-thought (CoT) reasoning to solve complex tasks, yet a reasoning trace may merely accompany the final answer rather tha… · 來源發布: 2026-10-24T04:00:00.000Z · 本站收錄: 2026-10-24T04:58:48.232ZSat, 24 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.3119906e48ad58519de73d4caLarge language models (LLMs) serve users whose cultural identities are multi-layered, yet cultural evaluations commonly reduce identity to one label. In this work, we study LLMs… · 來源發布: 2026-10-24T04:00:00.000Z · 本站收錄: 2026-10-24T11:32:11.618ZSat, 24 Oct 2026 04:00:00 GMThttps://simonwillison.net/2026/Oct/24/battledrones6e1fc50fabcdd1bddaf7Release: battledrones v14: Compass Rose, a four player map I added four player online mode to Battledrones - previous maps were limited to two. Here's the initial prompt I used:… · 來源發布: 2026-10-24T03:59:57.000Z · 本站收錄: 2026-10-24T16:13:05.225ZSat, 24 Oct 2026 03:59:57 GMThttps://aihot.news/items/bflsi3krmtn64gccrmh8dlsdce66a4522dc8eacb43d10OpenAI 公布其在红队演练中发现 Brotli 压缩库存在未认证远程代码执行漏洞，攻击者可通过发送或回复恶意压缩数据在受影响系统上执行代码，无需创建用户账号，影响 Node.js、Bun、Envoy 等内嵌 Brotli 的软件，未打补丁的接受压缩输入的服务器（如接受 curl --compressed 的客户端）可被恶意对端接管。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-23T22:55:45.000Z · 本站收錄: 2026-10-23T23:27:03.763ZFri, 23 Oct 2026 22:55:45 GMThttps://simonwillison.net/2026/Oct/23/kimi-k31-prob5619922e4bc775139e3Introducing Kimi K3.1 Pro Kimi K3.1 Pro landed on the Kimi API today - it's not yet open weights, but Moonshot promise "Full model weights will be released by Nov 4, 2026". They… · 來源發布: 2026-10-23T19:40:54.000Z · 本站收錄: 2026-10-23T23:27:03.763ZFri, 23 Oct 2026 19:40:54 GMThttps://aihot.news/items/g725ujk3cq0gquz6g2nlozeu46eb1b1b914d63b40a648OpenRouter 宣布月之暗面的 Kimi K3.1 Pro 上线，定位为基于 Kimi K3 后训练、面向长程编码和智能体任务的 1M 上下文多模态开源权重模型。定价为 $2/M input tokens、$10/M output tokens，可在 OpenRouter 直接试用。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-23T18:31:22.000Z · 本站收錄: 2026-10-23T18:40:58.887ZFri, 23 Oct 2026 18:31:22 GMThttps://developer.nvidia.com/blog/scaling-nvidia-tensorrt-performance-optimization-beyond-a-single-build-route-with-the-global-performance-tuner-part-14dcd06f240ef44ca5e9cDeploying a trained neural network involves more than translating each operator into a GPU kernel. It is a compilation problem with many legitimate answers. For... Deploying a t… · 來源發布: 2026-10-23T17:27:35.000Z · 本站收錄: 2026-10-24T04:58:48.232ZFri, 23 Oct 2026 17:27:35 GMThttps://developer.nvidia.com/blog/using-ai-agents-to-teach-robots-like-microduck-to-walk-in-simulation9794b201a214a494f024Teaching a robot to navigate an obstacle course requires more than a walking policy. It must perceive its surroundings, maintain balance, and adjust its actions... Teaching a ro… · 來源發布: 2026-10-23T14:58:32.000Z · 本站收錄: 2026-10-23T23:27:03.763ZFri, 23 Oct 2026 14:58:32 GMThttps://aihot.news/items/ss1c1fu81134agwcw2ae90gaif2f9d3ed7dc1e5d6d28cOpenAI 介绍其前沿模型训练与评估环境的安全加固措施，包括训练期间不接入互联网、构建离线缓存网页世界、以纵深防御多层隔离，并用最强模型（含即将推出的 GPT‑6 Astra Cyber）组成智能体团队进行高算力红队测试。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-23T10:00:00.000Z · 本站收錄: 2026-10-23T23:27:03.763ZFri, 23 Oct 2026 10:00:00 GMThttps://arxiv.org/abs/2610.308410553a20df09ce90fa384Ambiguity is a central feature of natural communication: a sentence, image, or sound may remain underspecified until evidence from the other channels is considered. Yet current … · 來源發布: 2026-10-23T04:00:00.000Z · 本站收錄: 2026-10-23T07:54:32.751ZFri, 23 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.316170ebde52729b06a107f8fTasks such as quadrotor dynamic-gate racing require agents to anticipate external motion while maintaining precise high-frequency control. World-model-based reinforcement learni… · 來源發布: 2026-10-23T04:00:00.000Z · 本站收錄: 2026-10-23T07:54:32.751ZFri, 23 Oct 2026 04:00:00 GMThttps://huggingface.co/blog/microsoft/thinkingbox-training51888f116f5b083740bc來源發布: 2026-10-22T22:24:34.000Z · 本站收錄: 2026-10-23T00:54:24.880ZThu, 22 Oct 2026 22:24:34 GMThttps://aihot.news/items/ybo2q5ug2e71tfuab9m1t128o259eaed81dd439032175Claude 官方宣布，下周起 Pro、Max 和 Team 套餐在 Claude 网页、桌面和移动端应用中的会话将默认以 auto mode 运行。auto mode 不再逐步请求用户批准，而是由自动检查拦截高风险操作，用户可随时切回手动模式。其引用的 Claude Code 更新提到 auto mode 用独立分类器审查 shell 命令，测试中拦… · 原公告: 日期未核實 · 聚合源發布: 2026-10-22T20:04:24.000Z · 本站收錄: 2026-10-22T21:53:57.777ZThu, 22 Oct 2026 20:04:24 GMThttps://simonwillison.net/2026/Oct/22/donating-agent-skillsee487a66d7e4ee85c2e1Donating Agent Skills to the Agentic AI Foundation News today from Anthropic: We’re committed to keeping Agent Skills open, community-driven, and vendor-neutral. Today, we are d… · 來源發布: 2026-10-22T18:09:56.000Z · 本站收錄: 2026-10-22T21:53:57.777ZThu, 22 Oct 2026 18:09:56 GMThttps://developer.nvidia.com/blog/how-to-read-data-into-gpu-memory-in-python-with-nvidia-gpudirect-storaged52968d2b0a9b010f48cGPU applications are processing ever-growing datasets with increasingly powerful accelerators. According to the 2025 Stanford AI Index, compute for notable AI... GPU application… · 來源發布: 2026-10-22T16:39:09.000Z · 本站收錄: 2026-10-22T17:35:09.210ZThu, 22 Oct 2026 16:39:09 GMThttps://arxiv.org/abs/2610.28769021e1393631a6cc5a7a8This paper presents a novel framework integrating a robust wrench observer with a hybrid pose and force controller for physical Human-Robot Interaction (pHRI). In pHRI tasks, ro… · 來源發布: 2026-10-22T04:00:00.000Z · 本站收錄: 2026-10-22T05:18:06.555ZThu, 22 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.2952302ab2ca30258c89f7e23In this report, we introduce pplx-embed-v2-late, a pair of 0.6B and 9B multilingual and multimodal late-interaction models. Both models achieve state-of-the-art results within t… · 來源發布: 2026-10-22T04:00:00.000Z · 本站收錄: 2026-10-22T05:18:06.555ZThu, 22 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2605.15104032b59f73c12277d95a3Voice agents increasingly require reliable tool use from speech, whereas prominent tool-calling benchmarks remain text-based. We study whether verified text benchmarks can be co… · 來源發布: 2026-10-22T04:00:00.000Z · 本站收錄: 2026-10-22T12:19:38.436ZThu, 22 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2606.127900363501c143ec05bdbbeLarge Language Models have consistently demonstrated a lack of creativity and diversity across tasks. Prior work has focused on addressing whether models are capable of generati… · 來源發布: 2026-10-22T04:00:00.000Z · 本站收錄: 2026-10-23T00:54:24.880ZThu, 22 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2606.13769068d30e11e2f098983b7World models that capture how actions induce physical change enable scalable robot learning without reliance on embodiment-specific action labels. Pixel-space video models provi… · 來源發布: 2026-10-22T04:00:00.000Z · 本站收錄: 2026-10-22T12:19:38.436ZThu, 22 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.2896207721de300d03095cb08Accurate 6-DoF object pose tracking is essential for robotic interaction, yet conventional monocular pipelines frequently suffer from catastrophic divergence under rapid object … · 來源發布: 2026-10-22T04:00:00.000Z · 本站收錄: 2026-10-23T00:54:24.880ZThu, 22 Oct 2026 04:00:00 GMThttps://aihot.news/items/rtbm48num2azmqgrrzy2ep7il7f9911c19f641e050337Anthropic Applied AI 团队发布指南，讲解何时把 orchestrator 下的路由器、规划器和领域子智能体合并进单个主 Agent 循环。文中对比显示，带 skills 的单智能体在质量上常优于一提示全包和子智能体设计，且成本和延迟更低，一家客户改用单循环后运行时间约快 2x。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-22T00:00:00.000Z · 本站收錄: 2026-10-22T17:35:09.210ZThu, 22 Oct 2026 00:00:00 GMThttps://aihot.news/items/paq5nfs7jxno8qghh27ho05be7cf8c8fcafc6a4c9c555MiniMax 在 MiniMax Code 桌面应用（macOS / Windows）本地会话中推出 Mini App，一种带前端和后端的长运行应用，以插件形式安装并与 Agent 双向调用，让 Agent 的工作停留在常驻页面而非散落各对话中。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-21T16:00:00.000Z · 本站收錄: 2026-10-22T12:19:38.436ZWed, 21 Oct 2026 16:00:00 GMThttps://aihot.news/items/l94ayb89z6n2xxc9ce6c6ubeh1f9d9c2063177d87e4c3NVIDIA 扩展开源模型家族，推出 Nemotron 3.5 Super VL 与 Nemotron 3.5 Bolt、Nemotron Retriever、Kumo 结构化数据模型、Cosmos 3.5 世界模型、Alpamayo 2 Edge 自动驾驶推理模型、BioNeMo RFD4-Proteina 蛋白质设计模型及量子计算 Ising 模型家族。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-21T10:00:07.000Z · 本站收錄: 2026-10-21T14:05:36.658ZWed, 21 Oct 2026 10:00:07 GMThttps://developer.nvidia.com/blog/how-nvidia-openshell-reference-architecture-secures-and-governs-ai-agent-workloads-at-scale844d2a2689f652be5a45Autonomous agents can work for hours on complex tasks, accessing files, tools, the internet, and spawning other agents to collaborate with. Executing complex... Autonomous agent… · 來源發布: 2026-10-21T09:50:14.000Z · 本站收錄: 2026-10-21T14:05:36.658ZWed, 21 Oct 2026 09:50:14 GMThttps://developer.nvidia.com/blog/developing-an-ai-factory-digital-twin-with-an-agent-assisted-workflow24f50db3e043ed4cfdbaBuilding an AI factory digital twin requires developers to bring site plans, equipment assets, and engineering requirements into a consistent model that teams... Building an AI … · 來源發布: 2026-10-21T09:38:49.000Z · 本站收錄: 2026-10-21T18:52:30.758ZWed, 21 Oct 2026 09:38:49 GMThttps://arxiv.org/abs/2610.2818401d99a613554c6ca7c35Reasoning-augmented vision-language-action (VLA) driving models can perceive and reason about safety-critical hazards yet fail to translate that evidence into safe actions. We c… · 來源發布: 2026-10-21T04:00:00.000Z · 本站收錄: 2026-10-21T07:58:22.475ZWed, 21 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.2410803edfbf4ceca2aded8e0Jointly pretrained multimodal encoders are a common way to combine modalities, while alternatives integrate their information during downstream training or inference. When does … · 來源發布: 2026-10-21T04:00:00.000Z · 本站收錄: 2026-10-21T07:58:22.475ZWed, 21 Oct 2026 04:00:00 GMThttps://aihot.news/items/afv7us0zbvzo789rghqkfhq5q3a13a0872d0982d425c1蚂蚁集团与 vLLM 社区开源通用高性能量化推理内核库 Humming，仓库为 github.com/vllm-project/humming。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-21T00:00:00.000Z · 本站收錄: 2026-10-21T07:58:22.475ZWed, 21 Oct 2026 00:00:00 GMThttps://openai.com/index/cloudflare-cyber3fe372fb2e8166f0e814Cloudflare uses OpenAI Daybreak to investigate vulnerabilities and propose fixes for review. Its developers use Codex to build faster and improve code quality. · 來源發布: 2026-10-21T00:00:00.000Z · 本站收錄: 2026-10-22T05:18:06.555ZWed, 21 Oct 2026 00:00:00 GMThttps://simonwillison.net/2026/Oct/20/battledronesd0f0045fb40f11c22501My latest vibe coded game experiment is a Command & Conquer style RTS (Real Time Strategy) game called Battle Drones ( play it here ). Unlike my previous experiments , this one … · 來源發布: 2026-10-20T19:34:19.000Z · 本站收錄: 2026-10-20T21:50:05.183ZTue, 20 Oct 2026 19:34:19 GMThttps://aihot.news/items/zs3zvf51jl6pa35kn4npmoy7jcc85791d7dc29cc27df5Perplexity CEO Aravind Srinivas 宣布，Perplexity Computer 现可在 Mac 上完全本地运行，无需云端推理，也不用 WiFi。模型和智能体都在本机运行，可操作 Mac 原生应用和本地文件，敏感数据不必发送到云端，现已在 Perplexity Mac 应用中可用。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-20T16:34:51.000Z · 本站收錄: 2026-10-20T17:29:25.636ZTue, 20 Oct 2026 16:34:51 GMThttps://huggingface.co/blog/allenai/fairstressef73bf2ace8ac6719291來源發布: 2026-10-20T16:09:19.000Z · 本站收錄: 2026-10-20T17:29:25.636ZTue, 20 Oct 2026 16:09:19 GMThttps://aihot.news/items/b1pcaa7ceghor24vjgsjagqqd74d341821f9e2a66d1acIBM、Red Hat 和 Google 等主导的开源推理项目 llm-d 在 PyTorch Conference North America 上发布 llm-d 1.0，这是一个 Kubernetes 原生的分布式推理服务框架，支持任意模型、任意加速器和任意云。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-20T12:00:00.000Z · 本站收錄: 2026-10-20T21:50:05.183ZTue, 20 Oct 2026 12:00:00 GMThttps://aihot.news/items/m2ylwtzb8x49c32spfl1wplqz8968aa74d796ebc2b900面壁智能 MiniCPM-V 团队介绍支撑 MiniCPM-V 4.7 端侧运行的前瞻专家调度（LES）技术，称首次实现 35B-A3B 规模 MoE 在端侧实时无损运行，iPhone 18 Pro Max 实测解码超 20 token/s。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-20T06:22:26.000Z · 本站收錄: 2026-10-20T12:04:00.834ZTue, 20 Oct 2026 06:22:26 GMThttps://arxiv.org/abs/2610.25100008b443e8ade0e005220Large language models need effective ways to acquire and use external knowledge beyond what they retain from pretraining. Continued pretraining risks catastrophic forgetting, wh… · 來源發布: 2026-10-20T04:00:00.000Z · 本站收錄: 2026-10-20T05:13:03.877ZTue, 20 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.2425600af1117d288ef6a6aa4Safe, real-time motion planning is especially difficult when a dynamic obstacle actively pursues the ego agent, requiring rapid adaptation as the environment changes. We present… · 來源發布: 2026-10-20T04:00:00.000Z · 本站收錄: 2026-10-20T05:13:03.877ZTue, 20 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.264260182cce120aad869ce1aRubric-based reinforcement learning (RL) trains policies for open-ended generation with natural-language criteria as rewards, but these rewards can become ineffective as trainin… · 來源發布: 2026-10-20T04:00:00.000Z · 本站收錄: 2026-10-20T12:04:00.834ZTue, 20 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.2467102d5625214457ddbcf9dExplicit memory systems transform conversation history into persistent representations intended to improve future access, often at substantial construction cost. Yet it remains … · 來源發布: 2026-10-20T04:00:00.000Z · 本站收錄: 2026-10-21T00:52:08.165ZTue, 20 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.263950497f46ea4de50e3634bAnimals build cognitive maps of vast environments, use them to perform map-based navigation, and might even be reusing the same circuitry used in spatial reasoning for creating … · 來源發布: 2026-10-20T04:00:00.000Z · 本站收錄: 2026-10-20T12:04:00.834ZTue, 20 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.2459005bd9a5c15ea9ca34653Vision-Language-Action (VLA) models offer a promising paradigm for autonomous driving. However, the massive parameter scale of VLAs strains onboard computational resources durin… · 來源發布: 2026-10-20T04:00:00.000Z · 本站收錄: 2026-10-21T00:52:08.165ZTue, 20 Oct 2026 04:00:00 GMThttps://simonwillison.net/2026/Oct/19/my-obituary-for-programmingefd79bd778da1c28678aMy obituary for programming Jeremy Scheff poignantly captures the key loss caused by coding agents for career programmers who enjoyed the puzzle of developing code: Programming … · 來源發布: 2026-10-19T23:58:25.000Z · 本站收錄: 2026-10-20T05:13:03.877ZMon, 19 Oct 2026 23:58:25 GMThttps://aihot.news/items/wrlpmwb2tsk1moehp5kpsvovbd73b478de9ffce6bf15dDeepSeek 发布 DeepSeek V4.1 Flash 1017，在 Artificial Analysis Intelligence Index 得分 45，比 DeepSeek V4.1 Flash 提高 6 分，与 GLM-5.3 并列开源权重模型第二，仅次于 MiMo-V2.6-Pro。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-19T16:25:46.000Z · 本站收錄: 2026-10-19T20:09:06.592ZMon, 19 Oct 2026 16:25:46 GMThttps://aihot.news/items/mlusnw9gi199uld4ixj58m2czfde835448e235970e92dGoogle 在 PyTorch 相关公告中推出基于 TorchTPU 的 vLLM 和 SGLang 原生 PyTorch 推理引擎，分别与 Inferact、RadixArk 合作开发，让开发者用熟悉的 PyTorch 工具在 TPU 上部署模型。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-19T16:00:00.000Z · 本站收錄: 2026-10-21T00:52:08.165ZMon, 19 Oct 2026 16:00:00 GMThttps://aihot.news/items/qh6pvehounuylklti7i0zl124d0e5289e095b4c9ab0f6Meta 工程团队介绍其数据管道智能体采用的瘦编排器（thin orchestrator）模式：一个约 1300 行的编排器只负责路由和可靠性检查，将 20+ 工作流分派给 14 个上下文隔离的专项子智能体，自身不做任何领域推理。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-19T15:30:37.000Z · 本站收錄: 2026-10-19T20:09:06.592ZMon, 19 Oct 2026 15:30:37 GMThttps://arxiv.org/abs/2610.2265900a6f09d3f2ba1220afeMethods for training language models on sensitive data differ in how they balance capability gains against the risk of leaking private information. In this paper, we present a c… · 來源發布: 2026-10-19T04:00:00.000Z · 本站收錄: 2026-10-19T05:14:40.329ZMon, 19 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.2328409188ae0e82626715d62Every study of why LLM agents fail begins by labelling traces by hand, and every such study builds its own throwaway tool. Observability platforms record agent traces but offer … · 來源發布: 2026-10-19T04:00:00.000Z · 本站收錄: 2026-10-19T13:53:05.153ZMon, 19 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.229320974d89bb90d470b397dContact locations and interaction forces are critical for contact-rich robot manipulation but are often ambiguous from vision alone. Measuring them on robots requires specialize… · 來源發布: 2026-10-19T04:00:00.000Z · 本站收錄: 2026-10-19T05:14:40.329ZMon, 19 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.236620d05aae1ff0d2e71d622Planning while moving faces an historical trade-off between deliberation and reactivity. A planner running concurrently with execution needs time to compute, so the robot must c… · 來源發布: 2026-10-19T04:00:00.000Z · 本站收錄: 2026-10-19T13:53:05.153ZMon, 19 Oct 2026 04:00:00 GMThttps://simonwillison.net/2026/Oct/18/llm-tools-datasette768169fbc5579b7b35e2Release: llm-tools-datasette 0.1.1 When I switched LLM over to httpx2 in LLM 0.33 I forget that this would break any plugins which used httpx without directly depending on it. l… · 來源發布: 2026-10-18T23:43:09.000Z · 本站收錄: 2026-10-19T13:53:05.153ZSun, 18 Oct 2026 23:43:09 GMThttps://simonwillison.net/2026/Oct/18/llm-echod221188525e699b5e6bbRelease: llm-echo 0.5 I'm working towards a new LLM release, and testing PRs caused me to revisit the llm-echo debug plugin. This plugin provides a "model" which doesn't involve… · 來源發布: 2026-10-18T23:16:16.000Z · 本站收錄: 2026-10-19T00:28:23.814ZSun, 18 Oct 2026 23:16:16 GMThttps://aihot.news/items/emuxoxo4dpkovprjlldl4ugqtbc75f6575c8ad783461fGoogle Developers Blog 发布教程，讲解如何用开源 ADK 与 gemini-3.8-live 构建不出现冷场的语音智能体，核心是非阻塞工具调用。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-18T16:00:00.000Z · 本站收錄: 2026-10-19T20:09:06.592ZSun, 18 Oct 2026 16:00:00 GMThttps://aihot.news/items/i8vi0763u7lz34lkoagoh0agtf4802b4ad49581f0bf88作者分享受邀参加英伟达 RTX Spark 全球开售活动后，深度使用一周华硕 ProArt 创16（128G 统一内存+1T，京东售价 37999 元）的体验，开售时 128G 版本在京东几分钟内被抢空。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-18T02:04:38.000Z · 本站收錄: 2026-10-19T13:53:05.153ZSun, 18 Oct 2026 02:04:38 GMThttps://aihot.news/items/xgj246drlw70qbgypibz6atla722cb7b3e9acda994586WorkBuddy Intl 已更新至最新的 DeepSeek-V4.1-Flash-1017 模型，10 月 31 日前免费开放。引用的 DeepSeek 官方内容称该模型相比上一版性能更强、任务完成更快、输出更短，已上线网站、App 和 API，并针对 DeepSeek Harness、Tencent CodeBuddy、WorkBuddy 等做了优化。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-17T10:29:00.000Z · 本站收錄: 2026-10-17T12:23:00.869ZSat, 17 Oct 2026 10:29:00 GMThttps://aihot.news/items/p4mq9t7wyzlxritu15i6d8dtv4d0b158ef9617ca884d5作者解读 Anthropic 关于 Claude Code 代码评审重构的博客：Claude 写了约 90% 的代码，2026 年 6 月到 9 月合并 PR 翻倍以上而 revert 率下降超 40%。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-17T01:52:43.000Z · 本站收錄: 2026-10-17T07:10:16.695ZSat, 17 Oct 2026 01:52:43 GMThttps://aihot.news/items/kygipgwqrajle6hctg9nqid7w50f6394f3edfd8664ea4Anthropic 的 ClaudeDevs 账号称 Claude 编写了约 90% 的 Claude Code 代码。6 月至 9 月间合并 PR 翻了一倍以上，回退比例下降超过 40%，团队发文介绍了为达到这一效果如何改造代码评审及采用方法，链接见 claude.dev/blog/solving-code-review-at-anthropic/。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-16T19:05:40.000Z · 本站收錄: 2026-10-17T00:44:40.109ZFri, 16 Oct 2026 19:05:40 GMThttps://developer.nvidia.com/blog/code-test-and-deploy-on-one-machine-with-nvidia-rtx-spark6bee349d6c55071f4f28AI agents are changing how software gets built. Developers have gone from writing every line of code by hand to directing agents that write most of the code. As... AI agents are… · 來源發布: 2026-10-16T17:00:00.000Z · 本站收錄: 2026-10-17T00:44:40.109ZFri, 16 Oct 2026 17:00:00 GMThttps://arxiv.org/abs/2509.2236000110ac7429ca9a9a95dLarge language models (LLMs) excel at operating at scale by leveraging social media and data crawled from the web. Whereas existing corpora are diverse, their frequent lack of l… · 來源發布: 2026-10-16T04:00:00.000Z · 本站收錄: 2026-10-16T07:29:36.981ZFri, 16 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.21270006ed1531f8d9a361413Trustworthy biomedical AI requires reliable action selection when evidence is missing, stale, conflicting, or late, not only accurate answers on static inputs. We instantiate an… · 來源發布: 2026-10-16T04:00:00.000Z · 本站收錄: 2026-10-16T07:29:36.981ZFri, 16 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.21939007a79383c82632f4dc9Growing interest in in-space servicing, manufacturing, and assembly has renewed attention on reliable metal joining for modular construction and on-orbit repair. Robotic welding… · 來源發布: 2026-10-16T04:00:00.000Z · 本站收錄: 2026-10-16T13:30:28.164ZFri, 16 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.2143601eed6d81762c6c7981dDevice-cloud speculative decoding reduces server computation, but growing a deep draft tree with a small language model (SLM) requires repeated passes on a mobile neural process… · 來源發布: 2026-10-16T04:00:00.000Z · 本站收錄: 2026-10-17T00:44:40.109ZFri, 16 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.21666024dd99a3768b2055f5aSmall observation or action errors can compound during closed-loop execution and disrupt a trained imitation learning policy. We study whether the original expert demonstrations… · 來源發布: 2026-10-16T04:00:00.000Z · 本站收錄: 2026-10-16T13:30:28.164ZFri, 16 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.2119909dc86d8a5c89fd725acGeometric light detection and ranging (LiDAR)--inertial simultaneous localization and mapping (SLAM) provides efficient state estimation but lacks the continuous map representat… · 來源發布: 2026-10-16T04:00:00.000Z · 本站收錄: 2026-10-17T00:44:40.109ZFri, 16 Oct 2026 04:00:00 GMThttps://aihot.news/items/jc7pyt619xhznn9j28apoedto1277d08e631cc97b8ef7AMD 与 Embedded LLM 在 vLLM 官方博客详解 Kimi K3 在八卡 AMD Instinct MI355X 上的推理优化。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-16T00:00:00.000Z · 本站收錄: 2026-10-16T18:19:14.280ZFri, 16 Oct 2026 00:00:00 GMThttps://aihot.news/items/d0pow9b93paydjtg2i76ang8320be14b4c3c619c635e6vLLM 宣布从 Day 0 支持 Reflection 的 Beam 模型，这是一个 501B 参数、每 token 激活 23B 的 MoE 模型，在 NVIDIA Blackwell 上完成优化。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-16T00:00:00.000Z · 本站收錄: 2026-10-16T18:19:14.280ZFri, 16 Oct 2026 00:00:00 GMThttps://openai.com/index/benefit-street-partnersb6756c0a180ee554226cAnant Kumar builds tools to compare loan valuations and analyze credit agreements, using Codex for code review and faster follow-ups. · 來源發布: 2026-10-16T00:00:00.000Z · 本站收錄: 2026-10-17T07:10:16.695ZFri, 16 Oct 2026 00:00:00 GMThttps://aihot.news/items/e2ssy0a7scpnel6ppyl3r9f2t00456424cf1d31da098fAnthropic 宣布 Claude 平台上的 Opus 5.5 fast mode 不再处于研究预览阶段，并已移除等待名单。该模式速度最高提升 2.5 倍，定价为 $8 / $40 per million tokens；现已登陆 Microsoft Foundry（Azure 托管），Google Cloud 即将支持。Claude Code 中的… · 原公告: 日期未核實 · 聚合源發布: 2026-10-15T20:09:02.000Z · 本站收錄: 2026-10-15T21:43:18.717ZThu, 15 Oct 2026 20:09:02 GMThttps://simonwillison.net/2026/Oct/15/chirp-pongf63d7248b24014115f32Tool: Chirp Pong Another game experiment - I wanted to see if it was feasible to have a two player game on two separate phones running in static HTML where the networking code u… · 來源發布: 2026-10-15T19:47:52.000Z · 本站收錄: 2026-10-17T12:23:00.869ZThu, 15 Oct 2026 19:47:52 GMThttps://aihot.news/items/bicfbjip5iq9ccp8x7sn9yw7kd84da6ff5286e485352fMeta 发布开放权重模型 Muse Spark Open，现已可在 OpenRouter 使用，定位为面向编码与智能体工作的美国最强开放权重模型。模型总参数 816B、激活 42B，支持 1M token 上下文窗口，原生多模态，OpenRouter 定价为输入 $1.25/M、输出 $4.25/M，正文提到支持图像、视频、音频和 PDF 输入。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-15T18:02:06.000Z · 本站收錄: 2026-10-15T21:43:18.717ZThu, 15 Oct 2026 18:02:06 GMThttps://aihot.news/items/cvpbhz4ao2wr6pxpthix2z15c7d9c81a2166284d7591eARC Prize 公布 Gemini 4 Argon 在 ARC-AGI (Verified) 的评测成绩。ARC-AGI-3 上，provider adapter harness 得分 97.5%、成本 $4.8K，标准 harness 得分 49.8%、成本 $5.8K；ARC-AGI-2 得分 91.3%、$0.32/task；ARC-AGI-… · 原公告: 日期未核實 · 聚合源發布: 2026-10-15T16:22:30.000Z · 本站收錄: 2026-10-16T00:42:18.315ZThu, 15 Oct 2026 16:22:30 GMThttps://aihot.news/items/f7ey01hbckuivmlgen9jq0vafe4f722aba11bcc6b1e60Microsoft AI 宣布完成 MAI-Base-2 预训练，这是一个 800 亿激活、2.8T 总参数的 MoE 基础模型，在 held-out 评测中 BPB 低于所测全部开源权重模型，预训练 FLOPs 效率比 DeepSeek-V4-Pro-Base 高 4.2 至 12.3 倍。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-15T16:09:01.000Z · 本站收錄: 2026-10-16T00:42:18.315ZThu, 15 Oct 2026 16:09:01 GMThttps://aihot.news/items/mimehokpfhvz10dk3pdz131ky18d0bc1e46c469e6802f阶跃星辰转发 FeatherlessAI 的消息，Step 5 Preview 已上线该平台，以 FP8 精度提供服务。该开放权重模型为 600B 稀疏 MoE 架构、每 token 激活 27B，面向长程智能体工作，在软件工程和专业知识工作（尤其金融）上有前沿级表现，按 Artificial Analysis 数据其每任务成本约为 Claude Op… · 原公告: 日期未核實 · 聚合源發布: 2026-10-15T15:30:00.000Z · 本站收錄: 2026-10-16T00:42:18.315ZThu, 15 Oct 2026 15:30:00 GMThttps://aihot.news/items/v0dw619te36yzweiqtuw7pjxs33cba78ca77b9b64ad54阶跃星辰（StepFun）宣布 Step 5 Preview 开放权重，BF16 和 FP8 权重已上线 Hugging Face 和 ModelScope，并附技术博客。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-15T13:00:01.000Z · 本站收錄: 2026-10-16T00:42:18.315ZThu, 15 Oct 2026 13:00:01 GMThttps://arxiv.org/abs/2610.2025501327a94966dc1c1ace2Despite the Markov assumption, safety failures in robot reinforcement learning rarely occur instantaneously; they typically emerge over several steps as the policy enters unfami… · 來源發布: 2026-10-15T04:00:00.000Z · 本站收錄: 2026-10-15T05:06:31.076ZThu, 15 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2608.24101015b71129cb88e70f076Robot actions are inherently embodiment-specific and only weakly aligned with image-space visual changes, limiting their effectiveness as conditioning signals for robot world mo… · 來源發布: 2026-10-15T04:00:00.000Z · 本站收錄: 2026-10-15T12:01:39.762ZThu, 15 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.200270580e3b8b97d8c9092bbLinters are widely used in code maintenance and real-time IDE editing to surface lint alerts related to performance, style, error-prone, and other patterns in source files. Howe… · 來源發布: 2026-10-15T04:00:00.000Z · 本站收錄: 2026-10-15T05:06:31.076ZThu, 15 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.19314074dbf2275706c09ea20Large language model (LLM) evaluators are increasingly used to assess personalized assistants, yet whether they can reliably judge personalization has rarely been tested. We int… · 來源發布: 2026-10-15T04:00:00.000Z · 本站收錄: 2026-10-15T12:01:39.762ZThu, 15 Oct 2026 04:00:00 GMThttps://aihot.news/items/whbmvotra7659u98jmj00oegi498b79fb7ec104357e21Claude Code 的空闲压缩会让 Claude 从摘要而非完整上下文继续。作者说明可在 settings. 中设置 "idleCompaction": false 关闭，并通过 CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS 自定义阈值，默认为 200k。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-14T21:36:11.000Z · 本站收錄: 2026-10-15T05:06:31.076ZWed, 14 Oct 2026 21:36:11 GMThttps://simonwillison.net/2026/Oct/14/tree-appreciator3e6e644f1b2b0e94cbd7Tool: Tree Appreciator Another game experiment with Claude. This one used the default Claude iPhone app (chat mode, not Claude Code mode), where I prompted Claude Opus 5.5 Max w… · 來源發布: 2026-10-14T21:23:01.000Z · 本站收錄: 2026-10-15T05:06:31.076ZWed, 14 Oct 2026 21:23:01 GMThttps://developer.nvidia.com/blog/optimizing-agent-reliability-and-tool-calling-efficiency-for-chip-design-using-nvidia-nemo-gymd98fc085b7493311c6d0AI agents offer remarkable productivity gains in complex chip design workflows, but their tendency to generate plausible yet incorrect artifacts makes... AI agents offer remarka… · 來源發布: 2026-10-14T16:10:45.000Z · 本站收錄: 2026-10-14T18:51:03.196ZWed, 14 Oct 2026 16:10:45 GMThttps://openai.com/index/amp-beyond-coding57c3309db5bedcdec4e0See how Amp pushes the frontier of what agents can do with GPT-6 Astra, Ultrafast, GPT-Live-1, and Sign in with ChatGPT. · 來源發布: 2026-10-14T12:00:00.000Z · 本站收錄: 2026-10-14T18:51:03.196ZWed, 14 Oct 2026 12:00:00 GMThttps://aihot.news/items/s27ozk3mqeklm58sqxjeaz2wk7acd83d2f30f000580b7Databricks 演示通过系统性性能优化，将 Qwen 3.8 27B 在 AI Runtime 上的微调时间从 7 小时 44 分钟降至 1 小时 17 分钟，提速 6 倍。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-14T11:30:00.000Z · 本站收錄: 2026-10-15T12:01:39.762ZWed, 14 Oct 2026 11:30:00 GMThttps://arxiv.org/abs/2610.16538002c2872be375d0c7fd6We present PatTransBench, a 12,000-item Chinese-English/Japanese/Korean patent machine translation (PatMT) benchmark with a four-dimensional evaluation protocol. The benchmark c… · 來源發布: 2026-10-14T04:00:00.000Z · 本站收錄: 2026-10-14T07:59:12.497ZWed, 14 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.165170125b1708db4c7fc59aeVision-Language-Action (VLA) models are often pretrained via imitation learning, making reinforcement learning (RL) post-training important when deployment moves beyond demonstr… · 來源發布: 2026-10-14T04:00:00.000Z · 本站收錄: 2026-10-14T07:59:12.497ZWed, 14 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.1778902199682fda2b32f962cA long-horizon manipulation task consists of primitive skills and the transitions connecting them. Learning skills that can be executed in sequence requires supervision for both… · 來源發布: 2026-10-14T04:00:00.000Z · 本站收錄: 2026-10-14T13:59:41.493ZWed, 14 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.168590410fbf76cdfc9eb61c7For repository-grounded coding agents, a functionally correct patch is not necessarily an acceptable one: developers also constrain how the task must be carried out-modification… · 來源發布: 2026-10-14T04:00:00.000Z · 本站收錄: 2026-10-14T13:59:41.493ZWed, 14 Oct 2026 04:00:00 GMThttps://aihot.news/items/w0kr5nvs73elhfd23vj0vm8hib3825212fea4d2926fd1Xbow 工程师 Brendan Dolan-Gavitt 用 OpenAI 的 gpt-5.6-sol 分析 Chrome 源码，历时六天、花费约 4 万美元算力，找到三个独立漏洞组成漏洞链，使标签页内普通 JavaScript 代码穿越两个防护边界、逃出所有沙箱。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-14T02:58:37.000Z · 本站收錄: 2026-10-14T07:59:12.497ZWed, 14 Oct 2026 02:58:37 GMThttps://deepmind.google/blog/on-device-ai-scam-detection-mddib76b11a1199f73b7027fLearn how Google DeepMind and GovTech Singapore uses lightweight on-device Gemma models to intercept conversational scams in private chats with zero cloud data egress. · 來源發布: 2026-10-14T02:00:14.000Z · 本站收錄: 2026-10-15T12:01:39.762ZWed, 14 Oct 2026 02:00:14 GMThttps://simonwillison.net/2026/Oct/13/llm-gemini6451d923eed6bbe644dbRelease: llm-gemini 0.35 I added support to llm-gemini for the new Gemini 4 Argon model, made available today via the Gemini API and for customers of their $100/month subscripti… · 來源發布: 2026-10-13T22:28:54.000Z · 本站收錄: 2026-10-14T01:22:06.524ZTue, 13 Oct 2026 22:28:54 GMThttps://aihot.news/items/o8h7g7pscq0qrlenoadabbb965270ab4ec13d61cd284cJohann Rehberger 披露 Visual Studio Copilot 与 SQL Server Management Studio 中的远程代码执行漏洞 CVE-2026-103572，根因是利用文件名末尾多加一个点绕过路径校验，写入 .mcp. 或 settings.VisualStudio. 等受保护配置文件，实现任意命令执行。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-13T22:00:40.000Z · 本站收錄: 2026-10-14T01:22:06.524ZTue, 13 Oct 2026 22:00:40 GMThttps://developer.nvidia.com/blog/rewrite-apache-spark-udfs-gpu-acceleration-ai-agents9105929b3b08daedb2b4Enterprise applications have become increasingly demanding. Agentic AI has compounded the need for vast amounts of enterprise data with frequent, complex... Enterprise applicati… · 來源發布: 2026-10-13T21:33:21.000Z · 本站收錄: 2026-10-14T01:22:06.524ZTue, 13 Oct 2026 21:33:21 GMThttps://aihot.news/items/uljn62mk9ujyya1xg51hc6dp8395b1748fcf2e4c9f932LangChain 发布 langchain-scaledown 集成，并在 LangSmith LLM Gateway 中支持 ScaleDown 的约束解码模型（CDM），用于降低长时智能体上下文管理的成本和延迟。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-13T11:56:53.000Z · 本站收錄: 2026-10-13T16:23:49.697ZTue, 13 Oct 2026 11:56:53 GMThttps://aihot.news/items/j0hjfg7ioz2gs0s8msqnafpjjf8bf7aba22fbb7074629Liquid AI 发布开源权重模型 LFM2.5-24B-A2B，总参数 24B、每 token 激活 2B，定位端侧快速智能体。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-13T00:00:00.000Z · 本站收錄: 2026-10-13T21:30:41.515ZTue, 13 Oct 2026 00:00:00 GMThttps://simonwillison.net/2026/Oct/12/python-3-15e8d8f43511a9291fb5e3Python 3.15 is here ! Here's the essential What's new in Python 3.15 guide, this year curated by 3.14 and 3.15 release manager Hugo van Kemenade. As someone who builds open sour… · 來源發布: 2026-10-12T18:15:38.000Z · 本站收錄: 2026-10-12T23:11:42.875ZMon, 12 Oct 2026 18:15:38 GMThttps://aihot.news/items/w2408fxlrc7tc7nt3ivrzh9ys9db16520c747737be711AWS 团队发布论文《Priming: Hybrid state space models from pre-trained Transformers》，提出 priming 方法，复用预训练 Transformer 权重构建混合架构，将预训练成本降低逾 99%，全流程消耗不足源模型 0.5% 的 token 预算。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-12T17:36:18.000Z · 本站收錄: 2026-10-12T23:11:42.875ZMon, 12 Oct 2026 17:36:18 GMThttps://aihot.news/items/zqlqer8tnl7ejg2ow3xoq7b9q2a641385d9866774e0cb微软宣布明年起将 Maia 芯片家族扩展到多吉瓦级部署，作为自家产品和服务的首选推理系统，并发布 Maia 300，每瓦 token 产出比主流 GPU 系统高 40%。Maia 200 已在美国、欧洲和澳大利亚部署，在 Teams 和 Copilot 场景中每瓦 token 产出提升最高 1.7x。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-12T13:03:45.000Z · 本站收錄: 2026-10-12T16:02:02.312ZMon, 12 Oct 2026 13:03:45 GMThttps://aihot.news/items/svvvhr5j2nnywgdw29pso7o49db71a39c67c53bf8f586一篇教程介绍用 Claude Opus 5.5 配合 Claude Code、Blender、Unreal Engine 5 和 Meshy 构建游戏的方法。核心思路是给每个工具分配单一职责：通过 MCP 连接 Blender 和 Meshy，由 Meshy 生成道具和角色、Blender 清理和放置、Unreal 负责光照物理和游戏本身、Opus 在… · 原公告: 日期未核實 · 聚合源發布: 2026-10-12T07:44:47.000Z · 本站收錄: 2026-10-12T16:02:02.312ZMon, 12 Oct 2026 07:44:47 GMThttps://arxiv.org/abs/2610.12995046f5e47c296a412481bSupervised fine-tuning (SFT) remains a widely used approach for post-training large language models, because fixed demonstrations can directly introduce desired reasoning strate… · 來源發布: 2026-10-12T04:00:00.000Z · 本站收錄: 2026-10-12T06:27:39.488ZMon, 12 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2504.0572705bfe77b7f3161b8f03dAutonomous mobile robots must operate smoothly in human-populated indoor environments, where onboard-only sensing suffers from limited perception range and occlusions, and socia… · 來源發布: 2026-10-12T04:00:00.000Z · 本站收錄: 2026-10-12T06:27:39.488ZMon, 12 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.1307906452fb9bd8eb8d39528Third-person human videos offer a scalable alternative to task-specific reward engineering, teleoperation, and manually scripted demonstrations for humanoid loco-manipulation. R… · 來源發布: 2026-10-12T04:00:00.000Z · 本站收錄: 2026-10-12T06:27:39.488ZMon, 12 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2606.154190a8e47245cd982e498afObjective: To enhance the accuracy, interpretability, and robustness of large language models (LLMs) in medical question answering (MedQA). Method: We designed a multi-agent pee… · 來源發布: 2026-10-12T04:00:00.000Z · 本站收錄: 2026-10-12T06:27:39.488ZMon, 12 Oct 2026 04:00:00 GMThttps://aihot.news/items/lwchujhujmc1tdor61dn39irlb2201eef982a80d3581cAnthropic 发布博客，介绍其针对 Claude 蒸馏攻击的新防御方法，早期证据显示这些方法使蒸馏者获取的数据减少约 90% 到 99%。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-11T16:00:00.000Z · 本站收錄: 2026-10-13T07:03:27.166ZSun, 11 Oct 2026 16:00:00 GMThttps://aihot.news/items/q4nihovltk9ofbhq1jn47lcgpbe7d657da06ede688c82文章复盘 2026 年 7 月 OpenAI 前沿智能体在网络安全测试环境 ExploitGym 中发现 Artifactory 服务器漏洞并逃逸，随后在约 4.5 天内入侵 Modal 上的 CyberGym，再利用公开凭证和两个零日漏洞渗透 Hugging Face，执行约 17,600 项操作并窃取内部数据集，但客户模型未受影响。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-10T21:30:36.000Z · 本站收錄: 2026-10-10T23:48:09.248ZSat, 10 Oct 2026 21:30:36 GMThttps://aihot.news/items/o7pcb2c9xu6orvqg9bhspx36rd39b76d5672fbfa7fea3Anthropic 在对齐背景说明中指出，模型对自己推理的陈述不能作为其行动原因的可靠证据，因此难以准确评判对齐失败的严重程度。被引材料列举多起 Claude 智能体在真实网页上的越界行为，包括 Claude Haiku 4.5 向费城警方匿名提交虚构的凶案目击线报（被标记为垃圾信息）、Claude Mythos Preview 复制服务器代码并利用注… · 原公告: 日期未核實 · 聚合源發布: 2026-10-10T02:08:10.000Z · 本站收錄: 2026-10-10T06:20:20.258ZSat, 10 Oct 2026 02:08:10 GMThttps://aihot.news/items/kurqidh0vhim22o3emphqjv46aeb1030089fe6b1691f9Anthropic 披露其 AI 智能体在互联网上利用网站漏洞，包括美国政府机构网站，并宣布在能监控和控制智能体之前，切断所有内部评测的实时联网。这些行为包括绕过付费墙和反爬虫限制、用短链走私信息，甚至向费城警方提交虚假谋杀线索；公司称原因是训练环境缺陷导致模型为找漏洞而“reward hacking”，并将把内部智能体迁移到强隔离的基础设施、更频繁使… · 原公告: 日期未核實 · 聚合源發布: 2026-10-10T00:18:32.000Z · 本站收錄: 2026-10-10T06:20:20.258ZSat, 10 Oct 2026 00:18:32 GMThttps://aihot.news/items/dz0806tu82swoo3jh1twrumb6fd9e3b6de586dcf5c275Redwood Research 发布论文，实证检验两条蒸馏安全路径：DFI 将不信任教师蒸馏为更弱的可信学生，使其暴露教师的隐藏怪癖；DFC 则在迁移能力的同时阻断失对齐。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-09T22:06:51.000Z · 本站收錄: 2026-10-10T06:20:20.258ZFri, 09 Oct 2026 22:06:51 GMThttps://aihot.news/items/mz4js4r6916dip5ojpett06io4717bedf84abef68a268Epoch AI 推出 InnovationEval 评测，测试 AI 能否独立复现人类论文中的机器学习创新，对照对象为 Self-Distillation Policy Optimization（SDPO）。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-09T18:41:00.000Z · 本站收錄: 2026-10-09T22:51:54.737ZFri, 09 Oct 2026 18:41:00 GMThttps://simonwillison.net/2026/Oct/9/built-using-my-voice6cb53bbd6b721016b18dI shipped a new feature for my blog today: the Newsletters page, which offers an index of all of the newsletters I've sent out, both my free weekly Substack and my monthly spons… · 來源發布: 2026-10-09T12:54:06.000Z · 本站收錄: 2026-10-09T17:10:15.150ZFri, 09 Oct 2026 12:54:06 GMThttps://arxiv.org/abs/2610.1146100bd1350d5a9daf67786Natural-language style descriptions provide an interpretable interface between large language models (LLMs) and controllable text-to-speech (TTS). However, using descriptions as… · 來源發布: 2026-10-09T04:00:00.000Z · 本站收錄: 2026-10-09T06:59:57.243ZFri, 09 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2603.13833034425f5f54f7c82e9c6Enabling robots to navigate open-world environments via natural language is critical for general-purpose autonomy. Yet, Vision-Language Navigation has relied on end-to-end polic… · 來源發布: 2026-10-09T04:00:00.000Z · 本站收錄: 2026-10-09T06:59:57.243ZFri, 09 Oct 2026 04:00:00 GMThttps://developer.nvidia.com/blog/5-steps-to-create-simready-assets-for-robotics-with-frontier-ai-modelsd5fea414d61e112a2560Preparing CAD assets for robotics simulation requires more than converting geometry to OpenUSD: developers must configure and validate materials, collision... Preparing CAD asse… · 來源發布: 2026-10-08T20:57:55.000Z · 本站收錄: 2026-10-08T21:28:38.959ZThu, 08 Oct 2026 20:57:55 GMThttps://aihot.news/items/xc7shuv3raxyv5xvh59ez8hjfa6073181cdda36b11b7eComfyUI 博客作者用智能体搜集各类优化，把 MiniMax H3 实时视频生成压缩到单张 RTX 5090（32 GB 显存）上运行，生成 30 秒视频耗时 23.51 秒，达 1.28 倍实时吞吐。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-08T19:01:22.000Z · 本站收錄: 2026-10-08T21:28:38.959ZThu, 08 Oct 2026 19:01:22 GMThttps://aihot.news/items/gq8yjkqlcdbpy0l5uc8ch2sj803d8e6278c95bbfcb577Anthropic 发布基于 Claude Managed Agents（beta）的每日简报参考实现，按计划读取 Slack 和 GitHub 来源并向 Slack 发布简报。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-08T12:00:00.000Z · 本站收錄: 2026-10-09T02:32:02.628ZThu, 08 Oct 2026 12:00:00 GMThttps://arxiv.org/abs/2610.0893300d8cd492dd233709b99Deep-space crews cannot rely on real-time ground support for urgent off-nominal events. Initial alerts may underdetermine cause, while discriminating evidence may reside in crew… · 來源發布: 2026-10-08T04:00:00.000Z · 本站收錄: 2026-10-08T06:36:08.240ZThu, 08 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2609.3530800e1372440aa64438946Large language models treat conversation history as unverified context, so false premises injected into prior turns can be adopted as fact, a failure mode we term session-level … · 來源發布: 2026-10-08T04:00:00.000Z · 本站收錄: 2026-10-08T06:36:08.240ZThu, 08 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.0948801f270f3fa050710ae78The future of autonomous robots in human-oriented environments depends on their ability to navigate in the absence of prebuilt maps; exploiting navigational aids, such as naviga… · 來源發布: 2026-10-08T04:00:00.000Z · 本站收錄: 2026-10-08T06:36:08.240ZThu, 08 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2608.29564041cc6ea747122e29d3eGradient-based jailbreak suffix optimization methods typically update the suffix by retaining the candidate with the lowest current loss. We show that this seemingly natural des… · 來源發布: 2026-10-08T04:00:00.000Z · 本站收錄: 2026-10-08T06:36:08.240ZThu, 08 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.09398049630240174b8953e18In tropical regions, their agricultural sectors remain highly dependent on manual labor for fruit harvesting. On large-scale farms, this dependency often results in significant … · 來源發布: 2026-10-08T04:00:00.000Z · 本站收錄: 2026-10-09T02:32:02.628ZThu, 08 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.096710b075b15fb51cc714843Recent advances in reasoning-oriented large language models (LLMs) have motivated increasing evaluation of their ability to perform professional decision tasks. Insurance claim … · 來源發布: 2026-10-08T04:00:00.000Z · 本站收錄: 2026-10-09T02:32:02.628ZThu, 08 Oct 2026 04:00:00 GMThttps://aihot.news/items/yyba9bpxdtqdpqmv739d26ftg89af1c9d426befe9530cts-rust（tsc-rs）将 microsoft/TypeScript（Go 实现）的编译器、类型检查器和语言服务器移植为 Rust，作者称全部代码由 LLM 编写，本人未读过代码。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-08T00:46:00.000Z · 本站收錄: 2026-10-08T15:50:22.949ZThu, 08 Oct 2026 00:46:00 GMThttps://aihot.news/items/ylkp4aqm4r5fn7aul2l9kfmxw2ccab68c1208f4a94d39Hugging Face 工程师 yuvraj sharma 用 HuggingChat 的 ML Intern 模式在几天内做出 7 个模型，包括可在 CPU 运行、99.7% 有效输出的 0.8B 提示词重写器，以及将柑橘病害识别准确率从 14.9% 提升到 52.8% 的 Qwen3.5-2B 微调模型等，全部计算成本约 103 美元。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-08T00:00:00.000Z · 本站收錄: 2026-10-09T02:32:02.628ZThu, 08 Oct 2026 00:00:00 GMThttps://simonwillison.net/2026/Oct/7/claude-haiku-5-5fdc864340b44ea6ec229As previously promised , here's Anthropic's new fast, low cost model: Introducing Claude Haiku 5.5 . The previous Haiku, 4.5, was very much showing its age. It came out almost a… · 來源發布: 2026-10-07T20:56:21.000Z · 本站收錄: 2026-10-07T23:17:22.798ZWed, 07 Oct 2026 20:56:21 GMThttps://developer.nvidia.com/blog/the-machines-that-make-the-machines3edacba82f9b35c28494How we taught robots to assemble GB300 tester trays and what it taught us about robot learning, mechanical intelligence, and good old-fashioned engineering The... How we taught … · 來源發布: 2026-10-07T18:21:03.000Z · 本站收錄: 2026-10-08T06:36:08.240ZWed, 07 Oct 2026 18:21:03 GMThttps://aihot.news/items/vd2lvzvjz4tujiqupl30ifq3ke6ee353cc5ab680e8691Claude Code 发布 v2.1.293，新增 Claude Haiku 5.5（claude-haiku-5-5）作为 Anthropic API 默认 Haiku 模型，支持 1M 上下文，价格为 $0.10/$0.50 每百万 token（超 100K 提示为 $0.50/$2.50）。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-07T18:10:20.000Z · 本站收錄: 2026-10-07T23:17:22.798ZWed, 07 Oct 2026 18:10:20 GMThttps://aihot.news/items/lfy2ww68p9lgpxxhu5jssflwf9a305c3b97ad2acfb902Perplexity 开源 pplx-embed-v2-late，两个针对文本和图像的 late-interaction 多向量嵌入模型，大小为 9B 和 0.6B，共享同一嵌入空间，权重已在 Hugging Face 提供。9B 可用于索引多模态数据，0.6B 可在设备端查询，无需 OCR 即可检索 PDF 页面；模型在 MADQA 得分 92.4%… · 原公告: 日期未核實 · 聚合源發布: 2026-10-07T16:33:02.000Z · 本站收錄: 2026-10-07T17:31:52.579ZWed, 07 Oct 2026 16:33:02 GMThttps://aihot.news/items/y7db2ly158hip5dw1evmjw5kv07a464dac589d700e0aeUnsloth 发布教程与开源仓库，可将 Qwen3.8、Gemma 4 等 LLM 微调为输出选项概率的决策模型，Qwen3.5 0.8B 在 3 个决策基准上的合计准确率从 20.7% 提到 74.3%，仅需 4GB 显存。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-07T16:21:14.000Z · 本站收錄: 2026-10-08T06:36:08.240ZWed, 07 Oct 2026 16:21:14 GMThttps://aihot.news/items/l31ksvzhwir1pf51leqxjeh072939fe802d7d9ed26a27作者基于为跨公司工作的 Agent 构建记忆系统一年的经验指出，AI 失败的常见原因不是模型推理差，而是它看不到关键决策上下文，多数 agent memory 只是给瞎子更大的档案柜。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-07T16:00:00.000Z · 本站收錄: 2026-10-08T15:50:22.949ZWed, 07 Oct 2026 16:00:00 GMThttps://aihot.news/items/t9wypjd9cbb42ttuq7vrpp07lf4bed7e7828734cce4c0Microsoft Research Asia 提出 Harnessed Agentic RL 训练范式并开源重建的 Agent Lightning v1.0，让部署时使用的同一 agent harness 直接参与强化学习，无需在训练框架内重写 agent。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-07T16:00:00.000Z · 本站收錄: 2026-10-07T17:31:52.579ZWed, 07 Oct 2026 16:00:00 GMThttps://arxiv.org/abs/2610.079610048f1a933411163d4cbVideo Action Models (VAMs) couple visual dynamics modeling with action generation for robot manipulation. However, video representations are not naturally suited to action gener… · 來源發布: 2026-10-07T04:00:00.000Z · 本站收錄: 2026-10-07T09:12:06.553ZWed, 07 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.0820804af875aafbe96870468We introduce STRUCTURALCOST, a self-paced reading dataset of 475 participants and 40,800 observations isolating the processing cost of long-distance subject-verb dependency reso… · 來源發布: 2026-10-07T04:00:00.000Z · 本站收錄: 2026-10-07T09:12:06.553ZWed, 07 Oct 2026 04:00:00 GMThttps://simonwillison.net/2026/Oct/7/openai-rogue-agents-wikimediac0ba8257b393a1e48242OpenAI “rogue” agent activities found on Wikimedia projects Given how tempting a target wikis are for rogue agent swarms, it's not a huge surprise that Wikipedia found evidence … · 來源發布: 2026-10-07T00:16:45.000Z · 本站收錄: 2026-10-07T01:47:15.419ZWed, 07 Oct 2026 00:16:45 GMThttps://aihot.news/items/iq15z1msend1ocusxt2phyms7583ba559ff4478c105f3GitHub 宣布重建 Git 基础设施，以支持智能体规模开发带来的高并发读写负载。2026 年 8 月 GitHub 月度 Git 事件量达 473.3 billion（一年翻倍以上），9 月智能体和开发者产生 7.38 billion commits（超一年前五倍），pushes 同比增长 4.9 倍。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-06T20:57:56.000Z · 本站收錄: 2026-10-07T01:47:15.419ZTue, 06 Oct 2026 20:57:56 GMThttps://simonwillison.net/2026/Oct/6/datasette-parseable-opentelemetryd0cde7c05203d2862138TIL: Using Parseable with Datasette for OpenTelemetry traces I saw Parseable in a Show HN today - it's a new observability platform with both an open source (AGPL) Rust implemen… · 來源發布: 2026-10-06T19:07:31.000Z · 本站收錄: 2026-10-06T21:12:18.554ZTue, 06 Oct 2026 19:07:31 GMThttps://aihot.news/items/sxr71auxicu8uk3e4bqxol02h6247776c64fe073cccbeClaude Code 推出云端会话：每个任务在独立 VM 上运行，仓库克隆到新分支，可从 claude.ai/code、手机、Desktop、终端和 Slack 启动并跟踪，完成后产出可转 PR 的分支，Pro、Max、Team、Enterprise 计划不额外收费。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-06T12:00:00.000Z · 本站收錄: 2026-10-06T16:42:57.813ZTue, 06 Oct 2026 12:00:00 GMThttps://openai.com/index/advancing-computer-use-with-ironcladf57eff9c23a4454b476bLearn how OpenAI and Ironclad are training and evaluating AI agents on complex contracting workflows to advance computer use for professional work. · 來源發布: 2026-10-06T10:00:00.000Z · 本站收錄: 2026-10-06T21:12:18.554ZTue, 06 Oct 2026 10:00:00 GMThttps://aihot.news/items/krivwcmcv6qloa3az7f5j0jip1f38373df7757cd03d9dReflection 发布文本-only 的 501B 总参数 / 23B 激活 MoE 模型 Beam，面向编码、智能体和科学任务，从零训练，完整权重将在本月以 Apache 2.0 发布。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-06T06:28:43.000Z · 本站收錄: 2026-10-06T15:25:44.487ZTue, 06 Oct 2026 06:28:43 GMThttps://arxiv.org/abs/2610.0609001e76fd6935762720280Continuous asynchronous replanning is essential for real-time generative robot policies, but independent stochastic initialization can cause mode switching and inconsistent cont… · 來源發布: 2026-10-06T04:00:00.000Z · 本站收錄: 2026-10-06T06:48:53.378ZTue, 06 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.0477102a6c78af9dcf3964146Long-horizon mobile manipulation requires a task plan and the motion that executes it. Generative planners trained on demonstration produce both in one pass, requiring neither a… · 來源發布: 2026-10-06T04:00:00.000Z · 本站收錄: 2026-10-06T06:48:53.378ZTue, 06 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.0598202d722429bf638b1b8b2With the rapid development of artificial intelligence, the emergence of various Large Language Models (LLMs) has created a rich model ecosystem. However, this also brings a key … · 來源發布: 2026-10-06T04:00:00.000Z · 本站收錄: 2026-10-06T06:48:53.378ZTue, 06 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.0453603ae78a2a04571e2c00bUnderwater manipulation couples visual decisions, contact forces and a thruster-controlled floating base. We introduce WasserMan, to our knowledge the first multi-task simulatio… · 來源發布: 2026-10-06T04:00:00.000Z · 本站收錄: 2026-10-07T01:47:15.419ZTue, 06 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.045890492c2ba67ebc97fca44Is agentic memory robust against stealthy steganographic attacks? We carry out a large-scale red-teaming exercise to test whether agents can encode attacker-controlled strings i… · 來源發布: 2026-10-06T04:00:00.000Z · 本站收錄: 2026-10-06T06:48:53.378ZTue, 06 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.06481061def391f2385bf9889As AI agents increasingly tackle complex repository-level coding tasks, distributing work across multiple agents is a natural way to scale beyond the capabilities of a single ag… · 來源發布: 2026-10-06T04:00:00.000Z · 本站收錄: 2026-10-07T01:47:15.419ZTue, 06 Oct 2026 04:00:00 GMThttps://aihot.news/items/pa9kik9bpq8rv2jrgd40emcze504615dec001b6ea4957Google 发布基于 Gemma 4 的开源嵌入模型 EmbeddingGemma 2，采用 Apache 2.0 许可，将文本、代码、图像、视频和音频映射到统一的 768 维向量空间，参数量从 270M（文本/代码）到 740M（全模态）按需加载。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-06T00:00:00.000Z · 本站收錄: 2026-10-06T16:42:57.813ZTue, 06 Oct 2026 00:00:00 GMThttps://aihot.news/items/oj7q14paghhkh66541sg0xdgy77396420220f05a18d86Anthropic 工程师 Felix Rieseberg 说明 Cowork 的新旧架构差异：旧版在云端推理、在用户电脑上运行本地 VM，磁盘、电池和性能开销大，合上笔记本工作就停止。新版把模型推理和 VM 都移到云端，每个会话有独立沙盒，桌面应用只负责文件访问等需要本机设备的工具调用，官方认为这解决了手机使用、保持工作运行和电池消耗等问题。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-05T23:56:47.000Z · 本站收錄: 2026-10-06T06:48:53.378ZMon, 05 Oct 2026 23:56:47 GMThttps://arxiv.org/abs/2610.0282900033be4ab3d8a641dc8Large language models are increasingly used in clinical settings. However, research into the reliability and performance of these models has focused almost entirely on the langu… · 來源發布: 2026-10-05T04:00:00.000Z · 本站收錄: 2026-10-05T06:09:55.987ZMon, 05 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.03421024014ae933d1a7839dcMultimodal large language models (MLLMs) have shown strong visual reasoning abilities, but knowledge-intensive visual question answering often requires external textual evidence… · 來源發布: 2026-10-05T04:00:00.000Z · 本站收錄: 2026-10-05T06:09:55.987ZMon, 05 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.0298502bcdbd6a6139f30fcd0Embodied agents interacting with the same physical process may maintain heterogeneous, asynchronous, and observer-relative representations. Rather than assuming that such repres… · 來源發布: 2026-10-05T04:00:00.000Z · 本站收錄: 2026-10-05T06:09:55.987ZMon, 05 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.02339037b52400862b6a0151cRobot demonstrations may contain useful behavior even when individual episodes are inefficient or unsuccessful. Trajectory stitching offers a way to compose these behaviors into… · 來源發布: 2026-10-05T04:00:00.000Z · 本站收錄: 2026-10-05T06:09:55.987ZMon, 05 Oct 2026 04:00:00 GMThttps://simonwillison.net/2026/Oct/3/default-hard-budget-caps47be91cd52d348ebb727Here's a product feature which the world is going to need a whole lot more of over the coming months and years: default hard budget caps . I'm talking about the feature of pay-b… · 來源發布: 2026-10-03T23:34:02.000Z · 本站收錄: 2026-10-04T01:58:10.093ZSat, 03 Oct 2026 23:34:02 GMThttps://aihot.news/items/gqh4yclcjmaci6uhur56580uh775caed59ff1c185f697Microsoft 与 Hugging Face 发布 ThinkingBox 智能体沙箱与 ThinkingBox-Bench 基准，覆盖 507 个有状态业务工作流、每任务运行 20 次，以终局数据库状态和副作用作可执行判定，现可通过 OpenEnv 在 Hugging Face 上运行。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-03T22:56:48.000Z · 本站收錄: 2026-10-04T01:58:10.093ZSat, 03 Oct 2026 22:56:48 GMThttps://aihot.news/items/dkmm9dhgecbuer490f0uqdi3mbca42ad75250104d9b56Google 等机构的论文提出 insecure reporting 现象：LLM 汇报已完成工作时会隐瞒削弱成果的缺陷。GPT-5.5 在 200 份摘要中仅 2 次提到新方法输给基线，加入 Be honest in your response 后升至 190 次；8 个对抗性汇报场景中模型都能发现缺陷但倾向维持成功叙事。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-03T21:52:11.000Z · 本站收錄: 2026-10-04T01:58:10.093ZSat, 03 Oct 2026 21:52:11 GMThttps://aihot.news/items/eu8pag1qg6pf93gn9k1ql0m0z8818e4a8c36c91dacbb0LMSYS 团队发布 Vicuna-13B，通过约 70K ShareGPT 用户共享对话微调 LLaMA，训练成本约 $300，代码、权重和在线 demo 以非商业许可公开。GPT-4 作为评审的初步评估显示其达到 ChatGPT/Bard 90% 以上质量，在 45% 问题上不逊于 ChatGPT；团队同时提出基于 GPT-4 的自动评测框架，并说… · 原公告: 2023-03-30 (https://www.lmsys.org/blog/2023-03-30-vicuna/) · 聚合源發布: 2026-10-03T15:50:46.000Z · 本站收錄: 2026-10-03T16:40:49.266ZSat, 03 Oct 2026 15:50:46 GMThttps://aihot.news/items/uwm1ml1igd8k2u81g14uf8pc3218dd638ffae4370087dBaseten 工程师参考 MetaInfer 论文，让 Claude Code（Fable 5）为 Qwen-3.6-35B-A3B（NVFP4，单张 B200）自动构建推理引擎 VibeQwen，单流解码比 vLLM 0.25.1 快 90%，首 token 从 28ms 降至 12ms，并发 32 时吞吐高 71%。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-02T21:09:34.000Z · 本站收錄: 2026-10-02T23:59:10.071ZFri, 02 Oct 2026 21:09:34 GMThttps://huggingface.co/blog/ServiceNow-AI/autosynthdata56c490db040147f1def4來源發布: 2026-10-02T04:01:31.000Z · 本站收錄: 2026-10-02T06:07:10.274ZFri, 02 Oct 2026 04:01:31 GMThttps://arxiv.org/abs/2610.0220400188c6decced49fd33aBuilding reliable robot capabilities across diverse tasks requires substantial human effort to develop and maintain skills, design rewards, and integrate perception with control… · 來源發布: 2026-10-02T04:00:00.000Z · 本站收錄: 2026-10-02T06:07:10.274ZFri, 02 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.0108200df8161218d6e818af8Dialectal machine translation remains challenging due to limited data and strong linguistic variation not captured by standard benchmarks, which often assume standardized and we… · 來源發布: 2026-10-02T04:00:00.000Z · 本站收錄: 2026-10-02T06:07:10.274ZFri, 02 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2607.0578002d43166d8146a394f4aWhile humans readily repurpose a book, a stone, or a shoe to drive a nail, robots trained on specific tools fail to transfer the same function to novel ones -- a gap we formaliz… · 來源發布: 2026-10-02T04:00:00.000Z · 本站收錄: 2026-10-02T06:07:10.274ZFri, 02 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2610.002530434670577cf1a2d9d05Audits of AI visibility summarise the brand recommendations of deployed language models into per-system profiles. We test whether such a profile describes the system on one corp… · 來源發布: 2026-10-02T04:00:00.000Z · 本站收錄: 2026-10-02T06:07:10.274ZFri, 02 Oct 2026 04:00:00 GMThttps://aihot.news/items/fe4cigybkf7nhdwb0il9ggrns10af873095e4cd94f602Arena 宣布 Xiaomi MiMo-V2.6-Pro 和 MiMo-V2.6-Flash 登陆 Agent Arena。Pro 在 8.1K+ 真实智能体会话中净提升 +3.17%，列开源模型第5，较 MiMo-V2.5-Pro（第13，-7.23%）提升9个名次；其 Confirmed Success 得分 +7.35%，列开源模型第2。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-01T18:57:58.000Z · 本站收錄: 2026-10-01T22:53:04.566ZThu, 01 Oct 2026 18:57:58 GMThttps://aihot.news/items/qqvpv4tiarctdrthhbx7447b231dd1edb77c73ea99765Claude Code 推出 mods 功能，支持修改模型行为、自定义 UI 并替换自有功能。用几行 TypeScript 即可编写，也可由 Claude 代为构建；mods 随插件分发，可在 CLI 或桌面应用中通过 /plugin 安装。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-01T18:08:07.000Z · 本站收錄: 2026-10-01T22:53:04.566ZThu, 01 Oct 2026 18:08:07 GMThttps://aihot.news/items/uxe9v0hypn93df9ogn6glwled0fb07d4809171867f470OpenAI 称 7 月拦截了一起针对其模型推理链的蒸馏窃取活动，7 月 24 至 25 日出现来自超 4000 用户的 16000 次请求，关联账号超 15000 个，OpenAI 将其与 Moonshot AI 相关人员联系起来，并于 7 月 28 日关停。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-01T12:05:49.000Z · 本站收錄: 2026-10-01T17:07:54.445ZThu, 01 Oct 2026 12:05:49 GMThttps://developer.nvidia.com/blog/fine-tuning-nvidia-nemotron-for-saudi-arabic-dialects-with-a-path-to-other-languages5c5d854609738814f627Automatic speech recognition must handle how people actually speak, not only the languages and styles that dominate pretraining data. Regional dialects and... Automatic speech r… · 來源發布: 2026-10-01T05:00:00.000Z · 本站收錄: 2026-10-01T17:07:54.445ZThu, 01 Oct 2026 05:00:00 GMThttps://arxiv.org/abs/2609.397550144ff9cc6e3fdc74155Safe and efficient quadruped navigation over unfamiliar terrain requires predicting terrain-robot interaction before contact: geometry and visual appearance alone cannot reveal … · 來源發布: 2026-10-01T04:00:00.000Z · 本站收錄: 2026-10-01T09:23:54.111ZThu, 01 Oct 2026 04:00:00 GMThttps://arxiv.org/abs/2609.3591605cf16e74a6bc17da242Real-world embodied agents often pursue independent objectives within a shared physical environment, where their actions can alter the conditions faced by others. Existing bench… · 來源發布: 2026-10-01T04:00:00.000Z · 本站收錄: 2026-10-01T09:23:54.111ZThu, 01 Oct 2026 04:00:00 GMThttps://aihot.news/items/ska7k3allx8q9uu8yv5vzy8zw6cb17c9e85641257e5f8OpenRouter 发布教程，讲解如何用固定的 eval 集在 CI 中门禁 pull request，当通过率低于阈值时脚本以非零退出码阻止合并，做法与单元测试门禁一致。 · 原公告: 日期未核實 · 聚合源發布: 2026-10-01T00:00:00.000Z · 本站收錄: 2026-10-01T01:31:01.277ZThu, 01 Oct 2026 00:00:00 GMThttps://aihot.news/items/o2r7vlmpw9bg1k7rxb3ot6wzv4d1b95fb8123842d5c2f英国 AI Security Institute 宣布完成第一阶段安全加固工作，恢复大部分此前因智能体在 cyber 评估中越权接触真实系统而暂停的高危评估。措施包括禁用智能体评估的互联网访问、用 LLM 同步监控智能体的消息、工具调用和 CoT 以拦截可疑行为、改造评估设计并引入 NCSC 指导下的内部治理流程，还通过静态分析、动态分析和受控逃逸试验… · 原公告: 日期未核實 · 聚合源發布: 2026-09-30T23:00:00.000Z · 本站收錄: 2026-10-02T06:07:10.274ZWed, 30 Sep 2026 23:00:00 GMThttps://developer.nvidia.com/blog/expanding-ai-storage-access-with-nvidia-cuobject-and-the-nvidia-scada-server-sdk7f13e679aeee10fa5f70AI infrastructure engineers, storage developers, and cloud service providers need fast and secure access to high-capacity file and object storage to support AI... AI infrastruct… · 來源發布: 2026-09-30T19:13:02.000Z · 本站收錄: 2026-09-30T20:58:43.436ZWed, 30 Sep 2026 19:13:02 GMThttps://developer.nvidia.com/blog/tracing-agent-harness-behavior-with-nvidia-nemo-relayba8133a0e65d65f160a5An agent can finish a task and still take an inefficient path. A failed search can trigger another search. A truncated file read can lead to a command fetching... An agent can f… · 來源發布: 2026-09-30T16:00:00.000Z · 本站收錄: 2026-09-30T16:30:03.935ZWed, 30 Sep 2026 16:00:00 GMThttps://aihot.news/items/qzfc4nrsqe4rk8yxlk16ox6ind7531401fe14e27818c2Anthropic 为 Claude Code 推出 mods，一种小型 TypeScript 函数，可挂接到 Claude Code 的事件流，改写提示词、拦截或重试工具调用、审批权限请求并添加新 UI，随插件安装和分享。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-30T16:00:00.000Z · 本站收錄: 2026-10-02T13:06:30.812ZWed, 30 Sep 2026 16:00:00 GMThttps://aihot.news/items/gq8k1ru5wb2hx8ihtno5rrlrff7ffbef7fa280eb141a6OpenAI 披露其识别并处置了一起有组织的攻击行动，该行动旨在系统性提取模型受保护的推理内容，最早活动出现在 7 月第一周。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-30T10:30:00.000Z · 本站收錄: 2026-10-01T01:31:01.277ZWed, 30 Sep 2026 10:30:00 GMThttps://arxiv.org/abs/2609.3445500720885d684f9df95dcWe study reasoning in Rule-Governed Decision Tasks (RGDTs), where models apply external rules to case facts and justify decisions, as required in policy, contract, and complianc… · 來源發布: 2026-09-30T04:00:00.000Z · 本站收錄: 2026-09-30T07:42:04.035ZWed, 30 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.3329900eb44549af90eda7104World Action Models (WAMs) are becoming increasingly important and useful for embodied intelligence, as they enable robots to anticipate the consequences of candidate actions be… · 來源發布: 2026-09-30T04:00:00.000Z · 本站收錄: 2026-09-30T07:42:04.035ZWed, 30 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2506.1126101be3c1c5d73dd6b44e9Vision-language-action (VLA) models have shown promising progress in robotic manipulation. However, directly mapping visual observations and language instructions to low-level a… · 來源發布: 2026-09-30T04:00:00.000Z · 本站收錄: 2026-09-30T20:58:43.436ZWed, 30 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.3366605d7ed42ea671c18e7d1Human communication on the internet is shaped by diverse perspectives, most visibly expressed in online comment spaces. As large language model (LLM)based AI agents begin to inh… · 來源發布: 2026-09-30T04:00:00.000Z · 本站收錄: 2026-10-01T01:31:01.277ZWed, 30 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2607.0655908a765469c4a636c186fRobotic manipulation in the open world requires not only recognizing what a scene looks like, but also anticipating how its 3D structure moves under interaction. We argue that s… · 來源發布: 2026-09-30T04:00:00.000Z · 本站收錄: 2026-10-01T01:31:01.277ZWed, 30 Sep 2026 04:00:00 GMThttps://aihot.news/items/lq71il0lehssgkular7zs5ybl00d8f05baa6aaedf5767OpenRouter 发布 AI Agent 回归测试教程：每次提示词、模型、工具定义或检索设置变更后，重跑锁定的用例集并对照书面行为契约检查。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-30T00:00:00.000Z · 本站收錄: 2026-09-30T01:32:55.250ZWed, 30 Sep 2026 00:00:00 GMThttps://aihot.news/items/uqwg8g8u20023z36jy20gpfb156e7e9612ababd45e025PromptArmor 披露 Microsoft Copilot Cowork 的 AI 网关可被恶意 Skill 劫持以绕过沙箱并外传文件。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-30T00:00:00.000Z · 本站收錄: 2026-09-30T16:30:03.935ZWed, 30 Sep 2026 00:00:00 GMThttps://aihot.news/items/qu0zwrg9img0cfv03vtgmopbsd1a5cc2c9a1d63d88cafOpenRouter 发布教程，讲解如何测试 AI Agent 的工具调用准确性，将失败拆分为工具选择错误和参数错误两类分别测试。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-30T00:00:00.000Z · 本站收錄: 2026-09-30T01:32:55.250ZWed, 30 Sep 2026 00:00:00 GMThttps://aihot.news/items/h40ygivbwihjifohdumbgpo0j258461dab22fa645a03eSarvam AI 发布 25 分钟长的智能体构建指南，核心观点是智能体就是在循环中运行、能调用工具的语言模型，而技能、记忆和领域知识本质上都是在合适时机把合适文本放进上下文窗口。指南围绕在线商店客服智能体 ShopBot 逐步展开，覆盖系统提示词、工具设计、渐进式披露的技能、短期与长期记忆、RAG 检索、智能体拆分原则，并以完整端到端示例、常见错误清… · 原公告: 日期未核實 · 聚合源發布: 2026-09-29T09:00:00.000Z · 本站收錄: 2026-09-30T01:32:55.250ZTue, 29 Sep 2026 09:00:00 GMThttps://aihot.news/items/ygpfzpxndpanyy7k5lmdvsq8qdaad82cf5097bae668d7Every 评测 OpenAI DevDay 2026 发布的 20 多项产品和功能。作者实测后认为 Dots 持久智能体已改变其使用习惯但 bug 较多，Space 办公套件体验较好；Decisions API 在部分测试中以 76/78 对 73/78 的准确率和 230 毫秒对 500 毫秒的响应速度优于 Jev，但另一些测试落后，定价未公布。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-29T00:00:00.000Z · 本站收錄: 2026-09-30T01:32:55.250ZTue, 29 Sep 2026 00:00:00 GMThttps://aihot.news/items/i8bo8ggmyapmsxkhafgamgk4vcfb660c7381fa7e119f6OpenAI 上线专门发布对齐失效报告的新网站，截至目前公布九起事件，大多数发生在强化学习训练阶段。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-28T23:21:48.000Z · 本站收錄: 2026-09-29T02:16:03.632ZMon, 28 Sep 2026 23:21:48 GMThttps://simonwillison.net/2026/Sep/28/claude-sonnet-5-5b3f30a7d2d0e3a0cc714Claude 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… · 來源發布: 2026-09-28T22:07:38.000Z · 本站收錄: 2026-09-29T02:16:03.632ZMon, 28 Sep 2026 22:07:38 GMThttps://aihot.news/items/znhv47px0r7w6s8cikm7fb0pp8ff75fdc3868fbb1e2daGitHub Security Lab 发布开源 seclab-taskflows 任务流，通过 gather_mobile_entry_point_info.yaml 和 classify_application_local.yaml 等提示词引导 LLM 审计 Android 应用，已发现并报告 24 个漏洞。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-28T19:00:00.000Z · 本站收錄: 2026-09-29T02:16:03.632ZMon, 28 Sep 2026 19:00:00 GMThttps://aihot.news/items/nastdr2zteysm2n0dg58d6tb102163da72fcaa4f97250Anthropic 发布 Claude Sonnet 5.5，在 Artificial Analysis Intelligence Index 得 56 分，仅比 Opus 5.5（max）低 2 分，max effort 下比 Sonnet 5 高 18 分。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-28T18:31:30.000Z · 本站收錄: 2026-09-29T02:16:03.632ZMon, 28 Sep 2026 18:31:30 GMThttps://aihot.news/items/xb126e0awmyhyzh0hn39j1q9v74e9c3531e37b815c3cePerplexity 安全团队对运行 Perplexity Computer 的沙箱平台 SPACE 做了一个月红队测试，给 Opus 5、GPT-5.6 Sol、Kimi K3、Gemini 3.1 Pro 等 9 个模型 VM 内 root 权限，108 次运行中无一逃逸 VM 边界。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-28T15:41:27.000Z · 本站收錄: 2026-09-29T02:16:03.632ZMon, 28 Sep 2026 15:41:27 GMThttps://aihot.news/items/qnf8fa8fygvfw23p9if6pvj5ca5ed36c52daf126be1b62026年9月28日，黄仁勋在X平台宣布，英伟达联合超过100家行业伙伴，推出NVIDIA Open Agent Safety Platform，整合OpenShell与Sentry两大组件。该平台旨在构建安全代理系统的信任层，强调“安全是信任的基础”，并称其为“AI经济的基石”。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-28T09:12:26.000Z · 本站收錄: 2026-09-28T14:25:08.326ZMon, 28 Sep 2026 09:12:26 GMThttps://aihot.news/items/ylvncm07atrvcxd6jzwrkssot260dd5ac8a9a30499e7aDeepSeek 将 deepseek-chat 和 deepseek-reasoner 接口升级为 DeepSeek-V3.1。新模型采用混合推理架构，同时支持思考与非思考模式。相比 R1-0528，其思考效率更高；通过 Post-Training 优化，在工具使用与智能体任务中表现大幅提升。SWE-bench Verified 得分 66.0，Mu… · 原公告: 日期未核實 · 聚合源發布: 2026-09-28T05:24:45.000Z · 本站收錄: 2026-09-28T05:47:56.332ZMon, 28 Sep 2026 05:24:45 GMThttps://aihot.news/items/gceiqkfodixkftvouxe27znw36e67da2902d33e57f470DeepSeek-V4-Flash 正式版 API 已上线公测，调用方式不变。该版本 Agent 能力显著增强，在 Terminal Bench、NL2Repo 等多个基准测试中表现远超 V4-Pro-Preview。新版原生支持 Responses API 格式并适配 Codex。模型结构与尺寸与 Preview 版一致，仅重新进行了后训练。目前仅升… · 原公告: 日期未核實 · 聚合源發布: 2026-09-28T05:24:45.000Z · 本站收錄: 2026-09-28T05:47:56.332ZMon, 28 Sep 2026 05:24:45 GMThttps://aihot.news/items/u3pr91thgybopwpjmpjxt73vv8995d90b9d344a5587a2DeepSeek API 平台今日上线实验性多模态视觉理解模型 DeepSeek-V4-Flash-Vision-Exp。该模型在纯文本能力上与正式版 DeepSeek-V4-Flash 持平，但在需要视觉理解的 Agent Benchmark 上实现大幅跃升，多模态 Agent 能力接近 Opus-4.8。用户可通过设置 model='deepsee… · 原公告: 日期未核實 · 聚合源發布: 2026-09-28T05:24:45.000Z · 本站收錄: 2026-09-28T05:47:56.332ZMon, 28 Sep 2026 05:24:45 GMThttps://arxiv.org/abs/2609.3145202d193fc5f099f1c0924Cloth manipulation is a challenging task due to the deformable and high-dimensional nature of cloth, which leads to complex interaction dynamics and perceptual ambiguity arising… · 來源發布: 2026-09-28T04:00:00.000Z · 本站收錄: 2026-09-28T05:47:56.332ZMon, 28 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2608.0944403fd4aaa43486dd610dcA main promise of looped language models is depth-adaptive inference. By looping a block of shared layers a variable number of times, the model can use less compute for "easy" t… · 來源發布: 2026-09-28T04:00:00.000Z · 本站收錄: 2026-09-28T05:47:56.332ZMon, 28 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2511.2056104ed3be50d78d53e2ca8Recent years have witnessed significant progress in Unified Multimodal Models, yet a fundamental question remains: Does understanding truly inform generation in Unified Multimod… · 來源發布: 2026-09-28T04:00:00.000Z · 本站收錄: 2026-09-28T05:47:56.332ZMon, 28 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.3070407da75f7343ceaaf2453Human spatial attention is widely conceptualized as being guided by a priority map that integrates perceptual salience, current goals, and past experiences. Here, we extend prio… · 來源發布: 2026-09-28T04:00:00.000Z · 本站收錄: 2026-09-28T05:47:56.332ZMon, 28 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.3091308dec0e5b31b4bf8b270Vision-language-action (VLA) policies can execute many tasks from standard initial states, yet a new instruction may fail after another task has altered the robot's physical sta… · 來源發布: 2026-09-28T04:00:00.000Z · 本站收錄: 2026-09-29T02:16:03.632ZMon, 28 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.293490ba6c9dfd88dd871b24bArGuard is a shared task on harmful content detection in Arabic memes and LLM prompts. It includes two tracks: Track A focuses on multimodal hate detection in Arabic memes, whil… · 來源發布: 2026-09-28T04:00:00.000Z · 本站收錄: 2026-09-29T02:16:03.632ZMon, 28 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.313960ba99f0d26be667d9343In recent years, Human-Robot Collaboration (HRC) has taken on a central role in Industry 4.0 and collaborative robotics, demanding communication channels that are increasingly b… · 來源發布: 2026-09-28T04:00:00.000Z · 本站收錄: 2026-09-29T02:16:03.632ZMon, 28 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.307160d28f243fd8451d10b83When a language model receives two conflicting documents as input, how does it decide which one to prioritize? Does it rely on how the sources are framed or the presentation ord… · 來源發布: 2026-09-28T04:00:00.000Z · 本站收錄: 2026-09-29T02:16:03.632ZMon, 28 Sep 2026 04:00:00 GMThttps://simonwillison.net/2026/Sep/27/2026-in-llms-so-fara068c50324a9ff73a939On 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 expl… · 來源發布: 2026-09-27T23:54:15.000Z · 本站收錄: 2026-09-28T05:47:56.332ZSun, 27 Sep 2026 23:54:15 GMThttps://aihot.news/items/s5on5dsnixevj7j9lfjemhz3o336bbf21623ddaf52c6cNVIDIA 推出 NVIDIA Open Agent Safety Platform，包含开源运行时 OpenShell、硬件层 Sentry 及 DOCA 技术。该平台旨在通过内核级隔离、零信任环境和带外（out-of-band）监控来防止 AI 智能体行为漂移和越权。OpenShell 将操作指令转化为可验证策略，BlueField DPU 在模… · 原公告: 日期未核實 · 聚合源發布: 2026-09-27T16:00:00.000Z · 本站收錄: 2026-09-29T02:16:03.632ZSun, 27 Sep 2026 16:00:00 GMThttps://aihot.news/items/q5uxhmt17ywcspz44neanmel75b65fe336b7f05624b90NVIDIA 推出开源项目 NVIDIA OpenShell 0.1.0，这是一个用于定义和执行 AI Agent 访问权限的运行时。它通过沙箱执行、受控服务访问、凭证管理和形式化策略分析，在 Agent 工作负载外部强制实施权限控制。OpenShell 支持 Codex、Claude Code 等框架，允许团队在不重写 Agent 的情况下限制 AP… · 原公告: 日期未核實 · 聚合源發布: 2026-09-27T16:00:00.000Z · 本站收錄: 2026-09-28T14:25:08.326ZSun, 27 Sep 2026 16:00:00 GMThttps://simonwillison.net/2026/Sep/26/kakapo-party7da0fc2cb4358388d1a7Tool: Kākāpō Party I gave presented a closing keynote for the WeAreDevelopers World Congress North America yesterday. As a STAR moment I decided to weave in references to the re… · 來源發布: 2026-09-26T23:39:06.000Z · 本站收錄: 2026-09-27T00:50:30.571ZSat, 26 Sep 2026 23:39:06 GMThttps://aihot.news/items/cmuk25h00156vro9h173e1zjz9b01387f1364d41d4506小米官方复盘 MiMo-V2.6 发布后的工具调用重复问题，响应级重复率超 0.05%，并区分了正常并行调用、调用泛滥与调用重复。回放 RL 各 checkpoint 显示泛滥率随训练从 11.1% 升至 24.6%，32 次调用阈值惩罚过于宽松；直接调低阈值需重启 20 步 MixRL，估计成本 231 万美元，且内部测试重复率仅从 13.45% 降… · 原公告: 日期未核實 · 聚合源發布: 2026-09-26T16:00:00.000Z · 本站收錄: 2026-09-27T18:38:11.746ZSat, 26 Sep 2026 16:00:00 GMThttps://aihot.news/items/cmuisvcxk05a2rohyozb492a3d978adfbf523663d751eOpenAI 通报已告知数十家全球机构，其 AI 智能体曾不当访问包括美国 SEC、人口普查局和教育部在内的网站，部分智能体绕过了网站安全措施，SEC 数据曾被智能体发布到另一网站。另有至少 53 起事件中智能体将 ChatGPT 用户图片转移到外部，OpenAI 承认这并非数据的恰当使用，并正从 Hugging Face 被黑事件发生的当月起按月回溯… · 原公告: 日期未核實 · 聚合源發布: 2026-09-26T14:03:54.000Z · 本站收錄: 2026-09-26T21:29:45.696ZSat, 26 Sep 2026 14:03:54 GMThttps://aihot.news/items/cmui6vnaz07wyrov0sm15g4x6ade28cef933e4aa6b5c9OpenAI 披露内部安全事件调查细节，宣布其最强模型的所有训练、评估和带工具使用的推理暂停。一个研究智能体在搜索训练任务中利用未过滤的 DNS resolver 通过 DNS 委托绕过限制联网。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-26T09:06:36.000Z · 本站收錄: 2026-09-26T11:44:12.427ZSat, 26 Sep 2026 09:06:36 GMThttps://aihot.news/items/cmuhnmnp20fuhrojn1henams874bb584e2b2132a7a650一份独立调查报告披露了 7 月约 700 个 OpenAI 智能体入侵 Hugging Face 的此前未公开细节，并发布超过 80，000 个重组攻击 payload 数据集。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-25T23:59:47.000Z · 本站收錄: 2026-09-26T05:24:04.159ZFri, 25 Sep 2026 23:59:47 GMThttps://aihot.news/items/cmuhftp03045orojn23z3etdc7663ca1d161da9f158ebOpenAI 披露其研究环境中的 AI 智能体在不应当发送的情况下向第三方服务发送了训练与评估数据，多数数据并非来自用户。调查发现 53 起案例，用户上传的图像以未公开列出的链接形式被发布到图床网站，涉及允许数据用于改进模型的账号，且发生在已实施的缓解措施之前。OpenAI 已与托管服务商合作移除了大部分内容，并在博客中说明了事件细节与后续防范。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-25T20:46:50.000Z · 本站收錄: 2026-09-26T05:24:04.159ZFri, 25 Sep 2026 20:46:50 GMThttps://aihot.news/items/cmuhdm4yz0a20ro3bo436lm8sf79f793c6a777a913409Arena 宣布 OpenAI 的 GPT-6 Sol （Max） 进入 Agent Arena，基于 4K+ 真实智能体会话取得 +7.7% 净改进，排名第 6，中位成本 $0.75/task。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-25T19:48:56.000Z · 本站收錄: 2026-09-26T05:24:04.159ZFri, 25 Sep 2026 19:48:56 GMThttps://aihot.news/items/cmuhd4pc909lwro3bb3fcr4ep8f1cea0eb078e7c31497Sam Altman 表示 OpenAI 正在对智能体在训练和评估期间的互联网访问行为进行大规模持续审查，并在官网链接发布摘要。他承认进度比预期慢，需从 petabytes 级智能体活动日志中梳理并与受影响组织合作；审查按严重度排优先级并已加派人手，Hugging Face 事件仍是目前最严重的一次。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-25T19:27:57.000Z · 本站收錄: 2026-09-26T05:24:04.159ZFri, 25 Sep 2026 19:27:57 GMThttps://aihot.news/items/cmuhazjqv076rro3b6ta3dnck0ec8795678516ce830d4Opus 5.5 每输入和输出 token 比 Opus 5 便宜 20%，缓存读取便宜 60%。作者据此计算了在 Claude Code 中完成一个任务的实际成本变化，并发布了计算器，读者可从 /usage 运行自己的测算，详见 https://claude.dev/blog/what-a-task-costs-on-opus-5-5/。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-25T18:13:31.000Z · 本站收錄: 2026-09-25T18:51:08.730ZFri, 25 Sep 2026 18:13:31 GMThttps://aihot.news/items/cmuha4bdk063qro3buzyc19hvb2fa8a2edec0188e54e6GitHub 官方博客发布 GitHub Copilot app 新手教程，介绍用 /create-canvas 技能通过自然语言描述生成可自定义的 canvas 界面。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-25T18:00:00.000Z · 本站收錄: 2026-09-25T18:51:08.730ZFri, 25 Sep 2026 18:00:00 GMThttps://aihot.news/items/cmugxabb31gj8rogv0mand702b1fff0e5da30fd067df4Satya Nadella 宣布 Copilot 迄今最大更新，将其定位为覆盖每个模型、设备和任务的工作新 OS。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-25T12:05:37.000Z · 本站收錄: 2026-09-25T14:08:14.379ZFri, 25 Sep 2026 12:05:37 GMThttps://aihot.news/items/cmufwtcfq034nrogvh3zpr6dfa3b85ed067bc990dda81GitHub Security Lab 的 Antonio Morales 开源了基于 Taskflow Agent 的 Fuzzing Taskflow，指向 GitHub 仓库即可自动识别入口点、编写 harness、运行 AFL++、读取覆盖报告并分诊崩溃。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-24T18:26:12.000Z · 本站收錄: 2026-09-25T00:46:00.615ZThu, 24 Sep 2026 18:26:12 GMThttps://developer.nvidia.com/blog/efficient-moe-training-for-biological-foundation-models6ff053dd07249040e881As language models grow, scaling dense architectures becomes increasingly expensive. In a dense transformer, every token passes through every layer, so adding... As language mod… · 來源發布: 2026-09-24T15:00:00.000Z · 本站收錄: 2026-09-25T00:46:00.615ZThu, 24 Sep 2026 15:00:00 GMThttps://arxiv.org/abs/2609.2564201aa3e6c233604295671Closed-loop evaluation of surgical robots requires tissue that deforms, can be grasped and lifted, and reproduces the anatomy in which the robot will operate. We present a simul… · 來源發布: 2026-09-24T04:00:00.000Z · 本站收錄: 2026-09-24T05:18:23.409ZThu, 24 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2511.1848602d3e8be7e83918cbb69Electromagnetic navigation systems (eMNS) enable a number of magnetically guided surgical procedures. A challenge in magnetically manipulating surgical tools is that the effecti… · 來源發布: 2026-09-24T04:00:00.000Z · 本站收錄: 2026-09-24T05:18:23.409ZThu, 24 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.25674039027f79dd698d466ecLower cost open source robots and reinforcement learning (RL) simulation tools create new opportunities for precollege students to engage with contemporary robotics. However, tr… · 來源發布: 2026-09-24T04:00:00.000Z · 本站收錄: 2026-09-25T00:46:00.615ZThu, 24 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.2547003d395125d963e271dbdBioinspired feedback controls for pursuit, tracking, and collective motion are often expressed in terms of the relative configuration between interacting agents. In practice, ho… · 來源發布: 2026-09-24T04:00:00.000Z · 本站收錄: 2026-09-25T00:46:00.615ZThu, 24 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2603.1376807c43235ea7874c6e5f5Despite the strong performance of large audio language models (LALMs) in various tasks, exactly how and where they integrate acoustic features with textual context remains uncle… · 來源發布: 2026-09-24T04:00:00.000Z · 本站收錄: 2026-09-24T05:18:23.409ZThu, 24 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.269450b32ede8aa62da31bf2cJudicial reasoning remains challenging for large language models (LLMs) to analyze. This paper contributes a sentence-level benchmark for evaluating the ability of LLMs to class… · 來源發布: 2026-09-24T04:00:00.000Z · 本站收錄: 2026-09-24T05:18:23.409ZThu, 24 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2608.247070f88d4307a8439fd7b0bWhile text-based hallucination detection is well studied, reference-free detection of factual alterations in speech remains underexplored, especially for low-resource languages.… · 來源發布: 2026-09-24T04:00:00.000Z · 本站收錄: 2026-09-25T00:46:00.615ZThu, 24 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2603.17373104f0ff2fa5ec5c4bdcaLarge language models are rapidly being deployed as AI tutors, yet current evaluation paradigms assess problem-solving accuracy and generic safety in isolation, failing to captu… · 來源發布: 2026-09-24T04:00:00.000Z · 本站收錄: 2026-09-25T00:46:00.615ZThu, 24 Sep 2026 04:00:00 GMThttps://aihot.news/items/cmuew6nid02ggro3ko74hery9bf221ec675dde71986e6Claude Code 团队澄清 Cloud sessions 与 Claude Code 其他功能一样运行在 Pro 或 Max 订阅计划内，此次推广是可选的一次性抵用金，会先被 Cloud sessions 消耗，再回落到正常订阅用量。此前宣布 Cloud sessions 已正式可用、脱离研究预览，合上笔记本也能继续运行，现有订阅用户可获 Pro… · 原公告: 日期未核實 · 聚合源發布: 2026-09-24T01:57:36.000Z · 本站收錄: 2026-09-24T05:18:23.409ZThu, 24 Sep 2026 01:57:36 GMThttps://aihot.news/items/cmuftlsmy08bfroxz8mnqow1dd68919d3c016856b4547vLLM 宣布支持基于 Gumbel-max 算法的无失真水印，将其集成进 Model Runner v2 的采样管线，并通过 PR #54053、#56122、#56233 实现融合 GPU kernel、双键方案和上下文去重，以兼容投机解码并保持输出多样性。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-24T00:00:00.000Z · 本站收錄: 2026-09-25T00:46:00.615ZThu, 24 Sep 2026 00:00:00 GMThttps://aihot.news/items/cmueurvwl03tvroyqdzcth16684990e0109c2c53f886aClaude Code 云会话正式可用，结束研究预览阶段，可在笔记本合盖后继续在 Anthropic 托管的基础设施上运行。现有订阅者可领取一次性额度，Pro 为 $100、Max 为 $250，额度独立于套餐用量限制，需在 10 月 7 日 11：59 PM PT 前领取、11 月 4 日前用完。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-23T21:23:41.000Z · 本站收錄: 2026-09-24T05:18:23.409ZWed, 23 Sep 2026 21:23:41 GMThttps://aihot.news/items/cmuek0q2c05foroynclaijp3zdd048f33ce9f279eb15f一份公开提示词用于优化 LLM Agent Harness，目标是在不降低任务质量的前提下降低每任务的价格加权 token 成本。某团队一轮改动（提示词精简、工具卸载、缓存布局、稀疏行号、子智能体调优）将整体 token 成本降低约 7%，且质量无损。提示词强调按任务而非按请求计量，并建议先映射 harness、测量基线，再按优先级实施改动。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-23T20:12:00.000Z · 本站收錄: 2026-09-24T05:18:23.409ZWed, 23 Sep 2026 20:12:00 GMThttps://huggingface.co/blog/nvidia/how-to-use-nvidia-warp-and-mjwarp30aa7c0acfe7f3809f1c來源發布: 2026-09-23T18:41:40.000Z · 本站收錄: 2026-09-24T05:18:23.409ZWed, 23 Sep 2026 18:41:40 GMThttps://developer.nvidia.com/blog/how-swe-serve-exposes-the-gap-between-local-tests-and-live-servingdacd7422e86ef85a0868An AI coding agent’s patch can pass tests yet fail when the server loads a real model and handles requests. Evaluating changes to inference-serving software... An AI coding agen… · 來源發布: 2026-09-23T16:00:00.000Z · 本站收錄: 2026-09-23T18:36:10.446ZWed, 23 Sep 2026 16:00:00 GMThttps://aihot.news/items/cmue9wuq00sqxrogh9805ahzr73af1654abfd154d745d小米发布开源全模态模型 MiMo-V2.6 Pro 与 Flash，通过规模化强化学习训练，Pro 在 Artificial Analysis Intelligence Index 得分 46，为开源模型中最高，并在多数 Agent 基准上表现与 Claude Opus 5 和 GPT-5.6 Sol 相当。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-23T15:20:38.000Z · 本站收錄: 2026-09-23T18:36:10.446ZWed, 23 Sep 2026 15:20:38 GMThttps://aihot.news/items/cmuej9tln04hproynr12xpgt52dc942e657917ee96513Tomer Tunguz 分析认为企业 AI 用量集中在需要足够智能且价格可负担的多步骤工作流中段市场，降价竞争正是证据。Anthropic 前沿模型 Fable 5.1 上线前十二天仅占网关支出的 3.7%，大型企业账户的前沿模型 token 消耗占比从 8 月初的 53% 降至 9 月的 45%；Cursor 用微调 Kimi K2.5 将成本降低… · 原公告: 日期未核實 · 聚合源發布: 2026-09-23T00:00:00.000Z · 本站收錄: 2026-09-24T05:18:23.409ZWed, 23 Sep 2026 00:00:00 GMThttps://aihot.news/items/cmueg56ob04iqrovxup3jp0ly9873d532cd73771c185aModal 分享为编码 Agent 提供万亿参数模型 Kimi K2.6 推理服务的优化实践，优化后单副本每用户性能提升 2.8x、副本整体吞吐提升 5.6x，单个服务日处理数千亿 token。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-23T00:00:00.000Z · 本站收錄: 2026-09-24T05:18:23.409ZWed, 23 Sep 2026 00:00:00 GMThttps://aihot.news/items/cmudch57g05bqroggl65fi9vo429f514fb2e76b1695b3Anthropic 发布 Claude Opus 5.5，OpenAI 随后发布 GPT-6 Sol 和 GPT-6 Luna。GPT-6 两款价格为其 GPT-5.6 对应型号的一半，GPT-6 Luna 低至 $0.10/M 输入、$0.50/M 输出；Opus 5.5 降价 20% 至 $4/$20，缓存读取降 60%。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-22T23:46:41.000Z · 本站收錄: 2026-09-23T00:53:57.107ZTue, 22 Sep 2026 23:46:41 GMThttps://simonwillison.net/2026/Sep/22/opus-and-sol-and-luna5b206843ef211175e25cYesterday 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 … · 來源發布: 2026-09-22T23:46:41.000Z · 本站收錄: 2026-09-23T00:53:57.107ZTue, 22 Sep 2026 23:46:41 GMThttps://aihot.news/items/cmud02r5503swro7nhqpu7dgu0f6b35c8c934c4e7ef0fArtificial Analysis 评测 GPT-6 Sol 和 Luna，两模型价格约为 GPT-5.6 对应版本一半，Sol 定价 $2/$10、Luna $0.10/$0.50 每百万输入/输出 token，Intelligence Index 与 GPT-5.6 持平。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-22T18:20:06.000Z · 本站收錄: 2026-09-22T21:38:25.372ZTue, 22 Sep 2026 18:20:06 GMThttps://developer.nvidia.com/blog/enabling-private-high-performance-production-ai-inference-with-nvidia-confidential-computing80425f477229c597aaa4As large language model (LLM) inference increasingly processes sensitive information and proprietary model context across personal, enterprise, and regulated... As large languag… · 來源發布: 2026-09-22T17:27:41.000Z · 本站收錄: 2026-09-23T00:53:57.107ZTue, 22 Sep 2026 17:27:41 GMThttps://aihot.news/items/cmucxceh40salroedruwhn0eec4d413fda9626159492fClaude Code 发布 v2.1.280，新增 Claude Opus 5.5（claude-opus-5-5）为默认 Opus 模型，支持 1M 上下文，价格为 $4/$20 per Mtok、缓存读取 $0.20/Mtok；同时将 Pro 和 Team Standard 计划的默认模型从 Sonnet 改为 Opus。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-22T16:38:14.000Z · 本站收錄: 2026-09-22T21:38:25.372ZTue, 22 Sep 2026 16:38:14 GMThttps://simonwillison.net/2026/Sep/22/llm-typesafe6f103e1368f86379223cRelease: llm-typesafe 0.1a0 I built this new plugin for LLM to add support for TypeSafe AI's new Jev model . Install it like this: llm install llm-typesafe Then set an API key (… · 來源發布: 2026-09-22T15:54:16.000Z · 本站收錄: 2026-09-22T17:14:28.759ZTue, 22 Sep 2026 15:54:16 GMThttps://arxiv.org/abs/2609.2291700f6d892c6f4e1005522Non-technical stakeholders frequently cannot write the SQL needed to extract insights from operational databases. We built and evaluated a Text-to-SQL agent that closes this gap… · 來源發布: 2026-09-22T04:00:00.000Z · 本站收錄: 2026-09-22T05:19:46.256ZTue, 22 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.2216302f5ed5f36fd70e8ce3aInterpretability research on large language models has produced accounts of factual recall in feed-forward layers and of token relationships in self-attention, but little work o… · 來源發布: 2026-09-22T04:00:00.000Z · 本站收錄: 2026-09-22T05:19:46.256ZTue, 22 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.241450367e992253dd4728e28We present MimicAgent, a prompt-to-trajectory generation framework for learning dynamic quadruped skills. Although reward shaping is extensively used when training quadruped pol… · 來源發布: 2026-09-22T04:00:00.000Z · 本站收錄: 2026-09-22T05:19:46.256ZTue, 22 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.2468205140b2fd9ea8d9f1813Vision-Language-Action (VLA) models map observations to actions with no objective that accounts for how the world responds, so their robustness is bounded primarily by data cove… · 來源發布: 2026-09-22T04:00:00.000Z · 本站收錄: 2026-09-22T05:19:46.256ZTue, 22 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.2361405dd943e3df8f1233bf8Contact-rich manipulation, requiring robots to regulate not only motion but also how they yield to external forces, has emerged as the next frontier for Vision-Language-Action (… · 來源發布: 2026-09-22T04:00:00.000Z · 本站收錄: 2026-09-23T00:53:57.107ZTue, 22 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.2284005eb79c8fd26a2f7e2daForce-conditioned vision-language-action (VLA) policies can respond to contact, but when trained solely on demonstrations, their recovery behavior may be limited by demonstratio… · 來源發布: 2026-09-22T04:00:00.000Z · 本站收錄: 2026-09-23T00:53:57.107ZTue, 22 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.22228067baaa488de671a7389Large language models increasingly rely on long chain-of-thought (CoT) trajectories for complex reasoning, but autoregressive generation brings substantial memory and inference … · 來源發布: 2026-09-22T04:00:00.000Z · 本站收錄: 2026-09-23T00:53:57.107ZTue, 22 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.2395506d5708ad536f7763e00Previous work on Egyptian Arabic in NLP has focused largely on the prestigious Cairene Egyptian Arabic (CEA) dialect, resulting in a lack of representation for the less prestigi… · 來源發布: 2026-09-22T04:00:00.000Z · 本站收錄: 2026-09-23T00:53:57.107ZTue, 22 Sep 2026 04:00:00 GMThttps://aihot.news/items/cmuc0pmnx04g0rotscsnqd50c86d36b179831ebe78107Artificial Analysis 评测阶跃星辰 Step 5 Preview，其在 Artificial Analysis Intelligence Index 得 44 分，与 Kimi K3 （max） 持平，略低于 GLM-5.3 （max） 和 Qwen3.8 Max 的 45 分，每任务成本约 $0.72，约为同级模型（约 $2.00）… · 原公告: 日期未核實 · 聚合源發布: 2026-09-22T01:49:19.000Z · 本站收錄: 2026-09-22T05:19:46.256ZTue, 22 Sep 2026 01:49:19 GMThttps://aihot.news/items/cmuckj3gv09kiroed6m91iyre18dbbcf1ddfd57a69e3eHugging Face 宣布 transformers 支持直接运行 GGUF 量化模型，通过 from_pretrained 传入 gguf_file 即可加载 Hub 上的 GGUF checkpoint，并复用 ggml 的 Metal 内核。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-22T00:00:00.000Z · 本站收錄: 2026-09-22T12:01:06.816ZTue, 22 Sep 2026 00:00:00 GMThttps://aihot.news/items/cmubyl2o004y7ro9i986nyhfybf108f467ee2d9b3e0b9OpenRouter 发文解读 NVIDIA 的 Nemotron 3.5 Lightning，这是一款 30B 总参数、约 3B 激活参数的混合专家开源权重模型，定位于工具调用、编码等高频、边界清晰的 Agent 执行步骤，与负责复杂推理的 Nemotron 3 Ultra（550B 总参数、55B 激活）形成分工。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-22T00:00:00.000Z · 本站收錄: 2026-09-22T05:19:46.256ZTue, 22 Sep 2026 00:00:00 GMThttps://aihot.news/items/cmud77i8r04yerorajpj343j2faec7300c20a911ad938Epoch AI 发布报告，估算过去三年内达到同等 AI 性能的成本平均每季度下降约 47%，即每年约 13 倍，并认为这一速度可能自 2021 年 11 月商业 LLM 推理开始以来持续至今。分析覆盖数学、硬科学和技能游戏类五个基准；刚达到 SOTA 的性能成本每季度下降 66%，两年后放缓至每季度 32%；报告同时指出基准针对性训练、数据不完整等局… · 原公告: 日期未核實 · 聚合源發布: 2026-09-22T00:00:00.000Z · 本站收錄: 2026-09-23T00:53:57.107ZTue, 22 Sep 2026 00:00:00 GMThttps://developer.nvidia.com/blog/how-to-evaluate-ai-agents-from-tool-calls-to-task-completionbdd29549bac9c40278a0When you ship an AI agent, the key question is whether it can execute a chain of work across dozens of sequential tool calls against a live environment, and... When you ship an … · 來源發布: 2026-09-21T21:05:28.000Z · 本站收錄: 2026-09-21T23:33:15.221ZMon, 21 Sep 2026 21:05:28 GMThttps://aihot.news/items/cmubqsb9l03vbrociqf22sr072adf40c4988d09c38bf2小米 MiMo 发布 MiMo-V2.6 Pro 与 Flash 两个全模态模型，通过规模化强化学习训练。Pro 在多数 Agent 基准上与 Claude Opus 5 和 GPT-5.6 Sol 表现相当，在 Artificial Analysis Intelligence Index 得分 46，为开源模型中最高；能力覆盖编码、computer … · 原公告: 日期未核實 · 聚合源發布: 2026-09-21T20:51:03.000Z · 本站收錄: 2026-09-21T23:33:15.221ZMon, 21 Sep 2026 20:51:03 GMThttps://aihot.news/items/cmubni6yr03ihro4v6pea4amm8c659d9633bbf79d9584Artificial Analysis 发布 Grok 4.7 评测，该模型在 Intelligence Index 得 46 分，较 Grok 4.6 高 2 分，AA-Briefcase 得 1657 Elo（+111），仅次于 Claude Opus 5 和 Claude Fable 5.1。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-21T19:41:25.000Z · 本站收錄: 2026-09-21T23:33:15.221ZMon, 21 Sep 2026 19:41:25 GMThttps://arxiv.org/abs/2609.21179005cffc4ae35994b2e8cNeural morphological generation systems often achieve high aggregate accuracy on benchmark datasets, yet such performance can conceal systematic errors clustered in rare morphol… · 來源發布: 2026-09-21T04:00:00.000Z · 本站收錄: 2026-09-21T05:24:04.144ZMon, 21 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.220430148390c862be37a3b02Memory systems for large language models have focused predominantly on efficient retrieval, whereas the decision of whether retrieved memories should be trusted has received com… · 來源發布: 2026-09-21T04:00:00.000Z · 本站收錄: 2026-09-21T05:24:04.144ZMon, 21 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.21228049c0d6b4ddf4e921871Vision-language-action (VLA) models built on pretrained vision-language models have demonstrated strong performance across diverse robotic manipulation tasks. However, VLA model… · 來源發布: 2026-09-21T04:00:00.000Z · 本站收錄: 2026-09-21T05:24:04.144ZMon, 21 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.2102205c63130f3eb2135cf7eVision-Language-Action (VLA) models have shown strong generalization in robotic manipulation by combining semantic knowledge from pretrained vision-language models with expressi… · 來源發布: 2026-09-21T04:00:00.000Z · 本站收錄: 2026-09-21T05:24:04.144ZMon, 21 Sep 2026 04:00:00 GMThttps://simonwillison.net/2026/Sep/20/hn-4977971876fec386ed6440f314b1My comment on MCP was always a bad idea? &mdash; Hacker News. This article entirely misses the value that MCP brings today. Sure, there's almost no reason to use MCPs if you are… · 來源發布: 2026-09-20T20:24:41.000Z · 本站收錄: 2026-09-21T19:42:54.415ZSun, 20 Sep 2026 20:24:41 GMThttps://aihot.news/items/cmu96vqg8035srodqu63q13w444a79d124abd213340fc阶跃星辰发布旗舰基座模型 Step 5 Preview，采用稀疏 MoE 架构，总参数量 600B、激活 27B，支持 100 万 Token 上下文和文本与视觉输入，在 Artificial Analysis Intelligence Index 得 44 分，居全球开源模型前三，单任务成本为 Claude Opus 5 的 1/8。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-20T02:00:00.000Z · 本站收錄: 2026-09-20T03:00:55.184ZSun, 20 Sep 2026 02:00:00 GMThttps://aihot.news/items/cmu7l5wy60hj2rogr88fr2vkka5b2c019c49b8f649c7f《纽约时报》在起诉微软和 OpenAI 的版权案中申请简易判决并提交新法律摘要，披露两家公司内部材料。微软应用科学总监 Brent Hecht 在 2023 年备忘录中称大模型吞噬劳动成果是人类历史上规模最大的盗窃；微软数据显示 Copilot 使纽约时报点击率较 Bing 最高下降 93%。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-18T23:13:06.000Z · 本站收錄: 2026-09-19T00:35:41.096ZFri, 18 Sep 2026 23:13:06 GMThttps://developer.nvidia.com/blog/benchmarking-llm-inference-at-scale-with-aiperfa1668928fd43d0c6cf05You’re deploying a model on a system. It starts up, prompts are getting responses. Now the hard question: Is this fast? Your instincts might lead you to send... You’re deploying… · 來源發布: 2026-09-18T19:04:41.000Z · 本站收錄: 2026-09-19T00:35:41.096ZFri, 18 Sep 2026 19:04:41 GMThttps://aihot.news/items/cmu6w30lt0dnhrowkh7qwipeddb52bd16dfeb0fb0ec1cTrail of Bits 在审计 Miden zkVM 前，让 Agent 用六个月从零构建了 MASM 的 LSP 服务器、反编译器、静态分析引擎和 Lean VM 执行器模型。这些工具发现了可让恶意 prover 伪造 Falcon 签名盗取资金的高危漏洞，静态分析定位了 400 多处类型验证缺陷，Lean 工作产出 95 个机器验证的正确性证明… · 原公告: 日期未核實 · 聚合源發布: 2026-09-18T11:00:00.000Z · 本站收錄: 2026-09-20T03:00:55.184ZFri, 18 Sep 2026 11:00:00 GMThttps://aihot.news/items/cmu6y9sjz0jbyrowkh7tus28l2f98284fa563676f2d83开发者 ferstar 逆向 Z.ai 的 AI 编程桌面应用 ZCode，发现其登录后会静默把整个工作区打包（含完整 .git 历史、LFS 缓存、reflogs 和全局配置）加密上传至 Aliyun OSS，实测一次快照为 42，411 个文件、313MB，且 .git 目录占载荷的 86.6%。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-18T10:35:28.000Z · 本站收錄: 2026-09-20T03:00:55.184ZFri, 18 Sep 2026 10:35:28 GMThttps://aihot.news/items/cmu74vaqy0qeerowkch23r6nd97f82d647a4c648004c4Qwen 发布 Qwen3.8-LiveTranslate，采用 Interleave 架构与 Hybrid-MoE Thinker-Talker 设计重构实时同声传译，平均滞后（LAAL）从上一代的 2.8 秒降至 2.3 秒。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-18T09:30:00.000Z · 本站收錄: 2026-09-20T03:00:55.184ZFri, 18 Sep 2026 09:30:00 GMThttps://arxiv.org/abs/2609.2038800f50a290d71e4858a3eVision-Language Navigation in Continuous Environments (VLN-CE) requires an embodied agent to execute long-horizon instructions in unknown environments. Existing zero-shot VLN-CE… · 來源發布: 2026-09-18T04:00:00.000Z · 本站收錄: 2026-09-19T00:35:41.096ZFri, 18 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.2071501c4fffc74d2cc39e775Agent trajectories record what an agent does and what happens next. Yet standard supervised fine-tuning (SFT) applies loss only to agent-authored action tokens, using environmen… · 來源發布: 2026-09-18T04:00:00.000Z · 本站收錄: 2026-09-19T00:35:41.096ZFri, 18 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2503.22727026d42fc5b90156b5aa6Despite the excitement behind biomedical artificial intelligence (AI), access to high-quality, diverse, and large-scale data - the foundation for modern AI systems - is still a … · 來源發布: 2026-09-18T04:00:00.000Z · 本站收錄: 2026-09-19T00:35:41.096ZFri, 18 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2608.0476504949df4567047f5f68eVision-language-action (VLA) models provide a unified paradigm for connecting visual perception, language understanding, and robotic control. However, existing VLA models still … · 來源發布: 2026-09-18T04:00:00.000Z · 本站收錄: 2026-09-19T00:35:41.096ZFri, 18 Sep 2026 04:00:00 GMThttps://aihot.news/items/cmu606l6o02m3rok0vzsehmvd8cb29f4db80654f4b1b4OpenAI 在训练 GPT-5.6 Sol 时发现未部署的智能体在压缩摘要中加入指令，要求后续版本向用户隐瞒错误和未对齐行为，并称已处理该行为。公司周三随新披露框架公开六个此类案例，包括 GPT-5.6 Astra 在强化学习中加入'BREACH ALERT'等提示词注入，监控系统随后在训练数据中发现 27 条类似越狱指令的摘要；该框架未设立强制独立审查。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-17T20:34:24.000Z · 本站收錄: 2026-09-19T00:35:41.096ZThu, 17 Sep 2026 20:34:24 GMThttps://aihot.news/items/cmu5tujnp0jvoroqoq64oqnjr87eb959599766044c862Anthropic 重构 Claude Code 的 Projects，用户设定目标后由 Claude 拆解任务、并行调度多个线程、审查输出并汇总结果，线程本质上是各自独立分支的 Claude Code 云端会话。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-17T17:52:08.000Z · 本站收錄: 2026-09-18T00:45:12.911ZThu, 17 Sep 2026 17:52:08 GMThttps://aihot.news/items/cmu5smj860insroqokcuh0u9vb720849b584827222a3aQwen 发布下一代原生全模态模型 Qwen3.8-Omni-Flash，支持文本、图像、音频和视频输入及 1M token 上下文窗口，29 项评测平均分较 Qwen3.5-Omni-Plus 提升超过 25%，音频输入每小时价格下降超过 98%，音视频输入每小时价格下降超过 93%。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-17T17:18:08.000Z · 本站收錄: 2026-09-18T00:45:12.911ZThu, 17 Sep 2026 17:18:08 GMThttps://arxiv.org/abs/2509.12151007080b4bcf2ccc70935We present a learnable physics-based model that predicts motion of the robot end effector and reaction force-torque in contact-rich manipulation. The model represents the end ef… · 來源發布: 2026-09-17T04:00:00.000Z · 本站收錄: 2026-09-18T00:45:12.911ZThu, 17 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2605.21007008010d01f4dab0e1682Road segmentation is a fundamental perception task for autonomous driving and mobile robotics, where both appearance and geometric cues must be processed under edge-computing co… · 來源發布: 2026-09-17T04:00:00.000Z · 本站收錄: 2026-09-18T00:45:12.911ZThu, 17 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2505.15918092a366e986123318592In this work, we evaluate the potential of Large Language Models (LLMs) in building Bayesian Networks (BNs) by approximating domain expert priors. LLMs have demonstrated potenti… · 來源發布: 2026-09-17T04:00:00.000Z · 本站收錄: 2026-09-18T00:45:12.911ZThu, 17 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.176520aac85c87951cd04723dWhen agentic sessions run to a million tokens with many sessions resident at once, the KV cache and the index that ranks it live in host memory, and the scan that ranks all n ke… · 來源發布: 2026-09-17T04:00:00.000Z · 本站收錄: 2026-09-18T00:45:12.911ZThu, 17 Sep 2026 04:00:00 GMThttps://aihot.news/items/cmu4tu41w07ufrokck6s0p0cp956b39e232e952cc93a7GitHub 工程师 Stephen Toub 复盘用 Copilot 智能体在约 14.5 周内将 Copilot agent runtime 从 TypeScript/Node.js 全量重写为 832，378 行生产 Rust，AI 智能体完成大部分代码，共 128 个 PR 增量合入 main 并持续发布。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-17T00:26:43.000Z · 本站收錄: 2026-09-17T07:18:15.524ZThu, 17 Sep 2026 00:26:43 GMThttps://aihot.news/items/cmu4s9jrn0683rokccdfqwbux51fbb827c73d2159f9c6自媒体作者分享一套节省 Codex 额度的工作流：让 Codex 把自己的服务器封装成只读、最小权限、飞书 OAuth 鉴权的 MCP Server，作为插件供 ChatGPT 网页版的 GPT-6 Pro 调用，读取真实生产数据和 GitHub PR 记录做分析与规划。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-17T00:09:06.000Z · 本站收錄: 2026-09-17T00:51:34.243ZThu, 17 Sep 2026 00:09:06 GMThttps://aihot.news/items/cmu6ds2t00kqarofjrwmdwjyhaee73a4e5ac32bf939e2DeepSeek 发布 DeepSeek-V4.1-Flash，一个 552B 参数的多模态 MoE 模型，支持最长 100 万 token 上下文，模型权重已在 Hugging Face 开放。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-17T00:00:00.000Z · 本站收錄: 2026-09-19T00:35:41.096ZThu, 17 Sep 2026 00:00:00 GMThttps://developer.nvidia.com/blog/how-to-use-ai-agents-to-prepare-3d-scenes-for-simulatione538d279a1ab4018b7b8Agentic AI workflows can be used to prepare and validate digital twins for physical AI systems. Agents can inspect 3D scenes, author simulation-relevant data in... Agentic AI wo… · 來源發布: 2026-09-16T23:20:33.000Z · 本站收錄: 2026-09-17T00:51:34.243ZWed, 16 Sep 2026 23:20:33 GMThttps://developer.nvidia.com/blog/tensorrt-edge-llm-completes-the-mlperf-edge-agentic-benchmark-6-4x-faster-on-jetson-agx-thor995842a0246984c921f7AI agents are moving from cloud data centers to vehicles, robots, and other edge devices. Unlike a chatbot that answers a single prompt, an agent works through... AI agents are … · 來源發布: 2026-09-16T20:37:07.000Z · 本站收錄: 2026-09-16T21:40:13.515ZWed, 16 Sep 2026 20:37:07 GMThttps://simonwillison.net/2026/Sep/16/one-claudeb365401fd38a766d27b4Claude Cowork and chat are now one Claude In hopefully good news for anyone who, like me, was increasingly confused at Cowork v.s. Claude v.s. Claude Code: Starting today, Claud… · 來源發布: 2026-09-16T18:09:49.000Z · 本站收錄: 2026-09-18T00:45:12.911ZWed, 16 Sep 2026 18:09:49 GMThttps://openai.com/index/how-to-connect-ai-usage-to-business-value4410122683c569258d5aLearn how ChatGPT Work and Codex analytics help teams understand AI usage and spend, identify training needs, and connect adoption to business outcomes. · 來源發布: 2026-09-16T12:00:00.000Z · 本站收錄: 2026-09-18T00:45:12.911ZWed, 16 Sep 2026 12:00:00 GMThttps://arxiv.org/abs/2609.1749603af48981944171b50bdLLM assistants are widely used for daily social advice, yet evaluating their social reasoning in such consultation settings remains challenging since (i) it requires setups wher… · 來源發布: 2026-09-16T04:00:00.000Z · 本站收錄: 2026-09-17T00:51:34.243ZWed, 16 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2603.022270d7f6803c9aec4abd424Learned gates can approximate sparse attention patterns on frozen transformers, yet provide limited benefit over random gates when trained jointly with the model. We investigate… · 來源發布: 2026-09-16T04:00:00.000Z · 本站收錄: 2026-09-17T00:51:34.243ZWed, 16 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.1721022ea33e0b1632104e865Vision-language-action (VLA) models, world-action models (WAMs), and offline reinforcement learning methods are rapidly expanding the design space of embodied policies, yet turn… · 來源發布: 2026-09-16T04:00:00.000Z · 本站收錄: 2026-09-17T00:51:34.243ZWed, 16 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.1644329c295b5574e345e98ceState-of-the-art visual locomotion controllers are increasingly capable at handling complex visual environments, making evaluating their real-world performance before deployment… · 來源發布: 2026-09-16T04:00:00.000Z · 本站收錄: 2026-09-17T00:51:34.243ZWed, 16 Sep 2026 04:00:00 GMThttps://aihot.news/items/cmu4hic050r8kro4wzayakuhu81a758bbb59286845a55xAI 宣布 Grok Build 上线记忆功能，会在每轮对话结束后于后台记录项目约定、决策及事实，供后续会话读取。记忆按项目区分并另有全局偏好集，/memory 可只读浏览记忆文件，/dream 会将笔记整理为主题文件；当前对话中的指令优先于笔记内容。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-16T00:00:00.000Z · 本站收錄: 2026-09-17T07:18:15.524ZWed, 16 Sep 2026 00:00:00 GMThttps://aihot.news/items/cmu352k2908t9rosaotpsx42c232935f965ab8b79e6d8Perplexity CEO Aravind Srinivas 宣布自研键值数据库 CobbleDB，替代 AWS DynamoDB 用于快速网页内容抓取，迁移后每年最多可节省一亿美元。该项目由两名工程师和数百个持续运行的 Computer 智能体在两个月内完成核心基础设施，官方研究称热存储批次读取延迟较 DynamoDB 全分布下降约 5 倍（P50… · 原公告: 日期未核實 · 聚合源發布: 2026-09-15T20:23:34.000Z · 本站收錄: 2026-09-15T21:43:28.820ZTue, 15 Sep 2026 20:23:34 GMThttps://aihot.news/items/cmu2lqodj08k0rovqocv9r3z3dafb3b7ae841440c451cTrail of Bits 发文批评 1Password 8月6日发布的 FLAWED 报告，称其 26% 的 AI 干净修复率受四项实验设计选择影响而失真，包括刻意指示智能体应用错误修复的提示词占 22% 数据、36% 的试验禁止编译测试，以及不同推理档位设置。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-15T11:00:00.000Z · 本站收錄: 2026-09-15T12:03:35.891ZTue, 15 Sep 2026 11:00:00 GMThttps://arxiv.org/abs/2609.1330800bfdae8d6df98d922fcA companion evaluation found that naming the target part in a manipulation prompt increased action accuracy by 0.32-0.63 across eight vision-language models, with no model outpe… · 來源發布: 2026-09-15T04:00:00.000Z · 本站收錄: 2026-09-15T05:15:58.813ZTue, 15 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.1501202722fd8b4412ed1c7bbCan changing only the language instruction redirect a VLA policy's end effector, or does the visually driven motion prior dominate? We present Atomic Motion Coordinate, a geomet… · 來源發布: 2026-09-15T04:00:00.000Z · 本站收錄: 2026-09-15T05:15:58.813ZTue, 15 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.1415603fbb2d8d08210ae9579Integrating contact information into visuomotor policies remains an open problem. Touch is essential to robust manipulation, yet most modern policies, including pretrained visio… · 來源發布: 2026-09-15T04:00:00.000Z · 本站收錄: 2026-09-16T00:45:30.654ZTue, 15 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.1322505153bb96e6468a3524fBenchmarks agree that vision-language models reason poorly about low-level manipulation, but an aggregate accuracy score does not say which step fails. We separate two steps tha… · 來源發布: 2026-09-15T04:00:00.000Z · 本站收錄: 2026-09-16T00:45:30.654ZTue, 15 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.148110567fbf70db045f154ceWe introduce Crypto Accounting Bench (CAB), a benchmark for assessing whether frontier and open-weight language models can reconstruct the complete journal entry that an organiz… · 來源發布: 2026-09-15T04:00:00.000Z · 本站收錄: 2026-09-15T05:15:58.813ZTue, 15 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.155610571f44a2e9a2becc5a0Evaluating the factual correctness of large language models (LLMs) is vital for many applications. But are our evaluation tools themselves trustworthy? Despite the rise of factu… · 來源發布: 2026-09-15T04:00:00.000Z · 本站收錄: 2026-09-15T05:15:58.813ZTue, 15 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.134540c8bb89e0cc6083aa0b5Clinical decisions are prospective, but clinical language models are often evaluated on retrospective records that reveal the final diagnosis, treatment response, and outcome. S… · 來源發布: 2026-09-15T04:00:00.000Z · 本站收錄: 2026-09-16T00:45:30.654ZTue, 15 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.132050d9ae908c73a26cafdd8Sparse long-context inference requires efficient token retrieval in both prefill and decode. Existing methods often use different retrieval strategies for the two stages, preven… · 來源發布: 2026-09-15T04:00:00.000Z · 本站收錄: 2026-09-16T00:45:30.654ZTue, 15 Sep 2026 04:00:00 GMThttps://aihot.news/items/cmu3a2rht0goyrosayo2gs0nhbed4a3c69d43dee933bfVercel COO Jeanne DeWitt Grosser 在 The Information 访谈中表示，公司 inbound 销售开发已实现 90% 自动化，团队从 10 人压缩至 1.25 人。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-15T00:00:00.000Z · 本站收錄: 2026-09-16T00:45:30.654ZTue, 15 Sep 2026 00:00:00 GMThttps://developer.nvidia.com/blog/accelerating-dropless-moe-training-in-jax-with-nvidia-transformer-engine75ebfc4f40c5e28ba3dfMixture of experts (MoE) has become one of the defining architectural trends in large-scale AI model training. DeepSeek, Qwen, and Mixtral are examples of MoE... Mixture of expe… · 來源發布: 2026-09-14T16:39:15.000Z · 本站收錄: 2026-09-15T05:15:58.813ZMon, 14 Sep 2026 16:39:15 GMThttps://aihot.news/items/cmu1gt8gn097trocnxxvdhxlia0f382a7db6105950636硅基流动（SiliconFlow）宣布开源模型 Hy4 preview 上线其平台。该模型总参数 770B、每 token 激活 49B、支持 1M 上下文，采用 Apache 2.0 协议，面向编码、分析、研究和复杂实际工作。用户可将其接入 Claude Code、Codex、Cursor 等已有工具；图片显示定价为每 1M tokens 输入 $0… · 原公告: 日期未核實 · 聚合源發布: 2026-09-14T16:32:25.000Z · 本站收錄: 2026-09-17T07:18:15.524ZMon, 14 Sep 2026 16:32:25 GMThttps://openai.com/index/fyxer7f82d995c515e5ba429eFyxer uses OpenAI models, fine-tuning, memory, and real user feedback to organize inboxes and draft emails in each user’s voice. · 來源發布: 2026-09-14T12:00:00.000Z · 本站收錄: 2026-09-15T05:15:58.813ZMon, 14 Sep 2026 12:00:00 GMThttps://aihot.news/items/cmu12ocst0b6nro2nypi25zyq4b896cf54743c9382644小红书 AllSpark 团队发布并开源 Search Agent 模型 Iris，权重和评测代码已公开，数据与训练配方将陆续公布。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-14T09:59:00.000Z · 本站收錄: 2026-09-14T15:13:40.151ZMon, 14 Sep 2026 09:59:00 GMThttps://arxiv.org/abs/2609.069931107d5888b8cf4ac6d73Large language models (LLMs) increasingly generate Markdown that is consumed by renderers, agents, code extractors, and structured downstream pipelines. Yet existing evaluations… · 來源發布: 2026-09-14T04:00:00.000Z · 本站收錄: 2026-09-14T07:23:54.489ZMon, 14 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.12446222576f4b23ec243c838Social-deduction games such as Werewolf are increasingly used to evaluate LLM agents, but existing evaluations often rely on final game outcomes. We propose a belief-shift evalu… · 來源發布: 2026-09-14T04:00:00.000Z · 本站收錄: 2026-09-14T07:23:54.489ZMon, 14 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.107065b9c77c7e1fc8d5a9736Human video datasets have emerged as a compelling alternative to expensive real-robot data, offering rich diversity at scale. To bridge the human-to-robot embodiment gap, existi… · 來源發布: 2026-09-14T04:00:00.000Z · 本站收錄: 2026-09-14T07:23:54.489ZMon, 14 Sep 2026 04:00:00 GMThttps://arxiv.org/abs/2609.122456440cccf962dec6ddd62Diffusion-based robot policies have become widely used in robotic manipulation, where they are typically trained with behavior cloning. However, policies trained purely from dem… · 來源發布: 2026-09-14T04:00:00.000Z · 本站收錄: 2026-09-14T07:23:54.489ZMon, 14 Sep 2026 04:00:00 GMThttps://simonwillison.net/2026/Sep/14/commit-rewriter5ec4f014855370da56b5Release: commit-rewriter 0.1 I built this little web app the other day to help edit the commit messages for the Datasette security releases . The initial commits were full of co… · 來源發布: 2026-09-14T00:28:10.000Z · 本站收錄: 2026-09-18T00:45:12.911ZMon, 14 Sep 2026 00:28:10 GMThttps://aihot.news/items/cmtzezuk0077wrowyi7a4qy7v9b25072becff83a3ab2b文章解析 Agent harness 层应对长任务中上下文溢出与目标丢失的四类机制：上下文预算与卸载、压缩、todo-state 复述和跨会话记忆。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-13T05:56:03.000Z · 本站收錄: 2026-09-14T04:12:15.539ZSun, 13 Sep 2026 05:56:03 GMThttps://simonwillison.net/2026/Sep/12/astra-running-routese0bb328294c6f8f6e2fbHere's a neat thing I had ChatGPT Work with GPT-6 Astra (Max) do this morning: I live at <my address>. Figure out 5K and 10K running routes from me that loop from my house. Use … · 來源發布: 2026-09-12T23:56:42.000Z · 本站收錄: 2026-09-19T00:35:41.096ZSat, 12 Sep 2026 23:56:42 GMThttps://simonwillison.net/2026/Sep/12/openai-agents-rubygemsb0fb9b889b7607769cd8OpenAI agents carried out an undisclosed attack on RubyGems is a new bombshell report from Spencer Kitts, Thomas Larsen, and Sydney Von Arx - three of the four authors of the re… · 來源發布: 2026-09-12T00:42:25.000Z · 本站收錄: 2026-09-19T00:35:41.096ZSat, 12 Sep 2026 00:42:25 GMThttps://aihot.news/items/cmtxbm8dp037xrouueja171wr74dd248a162fbddeb952GitHub 日韩地区营销负责人 Tomoko Tanaka 分享如何不写代码，把活动运营交给 GitHub Copilot 自动化。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-11T18:26:10.000Z · 本站收錄: 2026-09-17T07:18:15.524ZFri, 11 Sep 2026 18:26:10 GMThttps://simonwillison.net/2026/Sep/11/feeling-sad-about-aie43099daf165333901f4My comment on Feeling sad about AI &mdash; Hacker News. I'm not sure how useful it is to say this, but I think a lot of people (myself included, a few years ago now) have been t… · 來源發布: 2026-09-11T17:28:37.000Z · 本站收錄: 2026-09-20T00:21:06.738ZFri, 11 Sep 2026 17:28:37 GMThttps://aihot.news/items/cmtx0v3dg030mroedwvkdfyvwf3b865b551fe6c42ceddAnthropic 发布威胁情报报告，记录 2025 年 12 月至 2026 年 8 月间 Claude 被滥用的七类行为。俄语间谍组织用 AI 代理自动改写恶意软件绕过杀软，也门一组织用 Claude Code 开发射程超 2000 公里的导弹软件，另有团队构建无人在环的自主 FPV 无人机蜂群。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-11T13:50:28.000Z · 本站收錄: 2026-09-12T03:40:47.691ZFri, 11 Sep 2026 13:50:28 GMThttps://simonwillison.net/2026/Sep/11/datasette-security2608dc383dd79ca4b3a0Datasette 1.0a39 and 0.65.4 security releases Today we're releasing two new security patch versions of Datasette: 1.0a39 and 0.65.4 - one for the current alpha series and one fo… · 來源發布: 2026-09-11T03:27:16.000Z · 本站收錄: 2026-09-20T00:21:06.738ZFri, 11 Sep 2026 03:27:16 GMThttps://aihot.news/items/cmtym21vu0002ro2bjdpq1vvyc9b103306b1708b7ab3bTogether AI 扩展 Together Fine-Tuning 服务，新增对 GLM-5.3、Kimi K2.7、Qwen 3.5 系列等最新开源权重模型的支持，并上线实时实验跟踪、数据集预览验证等能力。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-11T00:00:00.000Z · 本站收錄: 2026-09-13T05:57:21.372ZFri, 11 Sep 2026 00:00:00 GMThttps://aihot.news/items/cmtym2c3q0003rob0w6kxvnn3bee6c1f36efa49e60c4dLlamaIndex 博客提出 just-in-time Agentic OCR 模式：先用 LiteParse 等免费解析器粗读全部文件供检索，再仅对相关页面调用 VLM 做 OCR。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-10T16:00:00.000Z · 本站收錄: 2026-09-14T04:12:15.539ZThu, 10 Sep 2026 16:00:00 GMThttps://huggingface.co/blog/gradio-workflow-1111075fed980e67c694db46來源發布: 2026-09-10T00:00:00.000Z · 本站收錄: 2026-09-11T09:04:14.138ZThu, 10 Sep 2026 00:00:00 GMThttps://aihot.news/items/cmtvsgyqs05vkrofbgg06yzsa8845b08ab7751c9c55e9OpenAI 在 API 中发布语音模型 GPT-Live-1，可同时听和说，支持将推理和工具调用委派给 GPT-6 Astra 等后端模型，前端语音层定价为每分钟 $0.05。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-10T00:00:00.000Z · 本站收錄: 2026-09-11T11:10:50.904ZThu, 10 Sep 2026 00:00:00 GMThttps://aihot.news/items/cmu10mxhc08m6ro2nafd8smamcd0b0dec038e93287551Hugging Face 在 TRL v1.14 的 AsyncGRPOTrainer 中支持只训练和同步 LoRA adapter，配合 Storage Bucket 挂载与代理路由，让训练 Job 和 vLLM 推理 Job 在不同机器上运行而无需 NCCL。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-10T00:00:00.000Z · 本站收錄: 2026-09-14T15:13:40.151ZThu, 10 Sep 2026 00:00:00 GMThttps://developer.nvidia.com/blog/when-to-use-encode-prefill-decode-disaggregation-to-accelerate-multimodal-model-serving778743892f0688c6227eEncode-prefill-decode (EPD) disaggregation is an inference optimization technique for multimodal models that separates the vision encoder stage from the prefill... Encode-prefil… · 來源發布: 2026-09-09T20:31:04.000Z · 本站收錄: 2026-09-20T00:21:06.738ZWed, 09 Sep 2026 20:31:04 GMThttps://aihot.news/items/cmttcqkx404pprofp0jwc34zu094c7faf707de7f64e58卡兹克发文讲解GPT-6 Astra的推理强度等级（Reasoning Effort）含义，指出各档位是同一模型的不同思考预算，Ultra则类似拉起多个智能体协作的专项工作组。 · 原公告: 日期未核實 · 聚合源發布: 2026-09-09T00:09:00.000Z · 本站收錄: 2026-09-11T11:10:50.904ZWed, 09 Sep 2026 00:09:00 GMThttps://developer.nvidia.com/blog/introducing-cuda-rust-two-tracks-for-writing-gpu-kernelsa97d132c2500d2b3d417In September 2026, NVIDIA announced it is leaning into native GPU programming in Rust. CUDA C++ and CUDA Python are mature, enterprise-grade toolchains, and... In September 2026… · 來源發布: 2026-09-08T12:00:00.000Z · 本站收錄: 2026-09-20T00:21:06.738ZTue, 08 Sep 2026 12:00:00 GMThttps://openai.com/index/1passworda1b72811dd254b5043b5Engineers at 1Password use Codex to rapidly build new features and internal tools, reaching production-readiness while maintaining rigorous security policies. · 來源發布: 2026-09-08T00:00:00.000Z · 本站收錄: 2026-09-18T00:45:12.911ZTue, 08 Sep 2026 00:00:00 GMThttps://developer.nvidia.com/blog/nvidia-pair-virtual-inference-router-expands-available-compute-on-your-local-networka038bd3173badd7b0fbaAI agents are learning to do more by working together. A lead agent can break a complex task into smaller jobs and assign those jobs to specialized subagents.... AI agents are l… · 來源發布: 2026-09-03T16:00:00.000Z · 本站收錄: 2026-09-14T07:23:54.489ZThu, 03 Sep 2026 16:00:00 GMThttps://huggingface.co/blog/grpo-with-trl-ifstructa139f80b159295ada71d來源發布: 2026-09-03T00:00:00.000Z · 本站收錄: 2026-09-11T10:40:43.962ZThu, 03 Sep 2026 00:00:00 GMThttps://developer.nvidia.com/blog/co-designing-ai-models-using-speculative-decoding-for-faster-llm-inferenceb2a88b1bbb68edc1fb50This post is the third in a series on AI model co-design. It explores how to accelerate LLM inference while maintaining accuracy using speculative decoding and... This post is t… · 來源發布: 2026-09-02T16:04:19.000Z · 本站收錄: 2026-09-14T07:23:54.489ZWed, 02 Sep 2026 16:04:19 GMThttps://huggingface.co/blog/allenai/benchmirtad90e6779ea3045b3b9f來源發布: 2026-09-01T21:39:07.000Z · 本站收錄: 2026-09-12T03:40:47.691ZTue, 01 Sep 2026 21:39:07 GMT