AI 雷達 / 週報
按原始发布周整理来源摘录与阅读线索，不自动生成个人观点或趋势结论。
重點訊號
本週動態 / 來源訊號
AI 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…
Epistemic 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…
Figma 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。
GitHub 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。
GitHub 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。
Large language model (LLM)-driven agent scaffold evolution improves agent workflows through iterative candidate generation and evaluation. Evaluation both guides subsequent sear…
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…
Anthropic 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。
Humans intuitively solve complex problems by flexibly shifting among reasoning modes: they plan, execute, revise intermediate goals, resolve ambiguity through associative judgme…
Task-oriented conversational agents remain fragile under real world conversation scenarios as they rarely follow a predictable script, especially when users exhibit non-cooperat…
本週動態 / 來源訊號
Philipp 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。
Cursor 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 …
As 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.…
Large 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…
How 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…
OpenRouter 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。
Boris 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。
GitHub 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…
Addy 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。
Game 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 …
As 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 …
Generating 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…
Ai2 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。
Supervised Fine-Tuning (SFT) of LLMs suffers from catastrophic forgetting, and prevailing mitigation strategies struggle to reconcile the plasticity-stability dilemma. To addres…
End-to-end full-duplex spoken dialogue models enable natural, real-time spoken interactions with interruptions, backchannels, and overlapping speech. Prior evaluations have asse…
Serving 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…
本週動態 / 來源訊號
Building 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…
Factory 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。
Microsoft 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。
I'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 …
OpenRouter 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。
GPT-6 Astra tops Retell’s voice agent benchmark, while Codex helps it reach a $1.6M annual revenue run rate per employee.
How Greptile uses GPT-6 Astra to test user workflows, Daybreak to strengthen security, and Codex to reduce rework.
See how OpenAI engineers use Codex Ultrafast to investigate incidents in minutes. Explore faster security analysis, code refactoring, and daily workflows.
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。
本週動態 / 來源訊號
OpenAI will honor a shared robots.txt signal to help website operators communicate their AI training preferences to participating providers.
Modern 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 …
Musculoskeletal robots combine compliant actuators with articulated skeletons, but exploring alternative configurations often requires repeated fabrication. We present MUSBRIC, …
As robots and intelligent systems increasingly interact with humans, they must anticipate long sequences of future actions from only brief observations. While purely neural mode…
Cooperative transport with heterogeneous legged robots requires coordinating different motion capabilities while supporting an unfixed shared payload. We present HA-DiPO (Hetero…
Dynamic and continuously evolving environments present significant challenges for robotic mapping and scene understanding, particularly on construction sites where moving agents…
Physical 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…
Existing approaches for robot learning in simulation often require substantial manual effort to construct scenes and specify tasks, limiting training scalability and data divers…
For robots to work reliably in commercial and industrial applications, recent advances in agentic coding systems have potential to combine interpretable robot programming with t…
本週動態 / 來源訊號
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/。
本週動態 / 來源訊號
vLLM 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…
值得追蹤
