AI 情报文章归档

Microsoft Research 推出系列节目 Research | Start Here： Inside the Internship 的第一集，本科生研究实习生 Matheus Kunzler Maldaner 讲述自己探索智能体系统（agentic systems）研究的过程与收获。

Martian 推出交互式站点 AI Frontier（https://aifrontier.withmartian.com/），可按实测成本、质量和可靠性对比 44 个大语言模型，并展示路由和重复采样如何改变能力边界。

OpenAI 周四发布 GPT-6 Astra，总裁 Greg Brockman 称其为 generational leap，并个人认为 OpenAI 已达到 AGI，称是否算 AGI 留给用户判断，结束语为 Welcome to the AGI era。Astra 定位让 AI Agent 更接近独立完成复杂专业工作，同时也引发安全部署的疑问；OpenAI 称 Astra 基于其迄今最大规模训练运行，在德州 Stargate 场地使用超过 100，000 GPU。

Table of Contents Why Run an AI Model at Home? What Is a Local LLM? What Kind of Computer...

Lee Robinson 转发推荐 @pengzheng_ 撰写的 Grok Bot 设计思考文章，并称赞其中的设计细节。文章概括了 persistent roles、clear state、scoped context、coordinated teams 四个要点，指出界面目标是让用户从操作 AI 转向委派工作，原文见 https://x.ai/news/designing-grok-bot。

xAI 发布新博客，展示其一位设计师使用 always-on Grok Bot 智能体把想法变成可测试的原型。四个 Bot 各自有独立电脑、记忆和工具：Experiments 负责把不确定的想法做成实物、Motion God 根据感受描述制作动效原型、Figma Bro 直接读取真实 Figma 文件完成重复性生产工作、Devbot 解答工程实现问题。

OpenAI、Anthropic、xAI 和 Google 托管的云端 AI 模型周四上午出现数小时的罕见重叠服务中断。

OpenClaw 发布 v2026.9.1 版本，主要更新包括支持 Mermaid 图表绘制、简化安装流程、更新可自动停止、长对话更轻量。本次版本合并 1，186 个 PR，包含 28 个直接提交，来自 281 位贡献者，发布说明见 https://docs.openclaw.ai/releases/2026.9.1。

据 The Information 报道，Mira Murati 的 Thinking Machines Lab 正洽谈以至少 400 亿美元投前估值融资至少 10 亿美元，Accel 讨论领投，Nvidia 可能参投。该估值约为其 2025 年种子轮约 100 亿美元投前估值的 4 倍，但仍低于去年底探讨的 500 亿美元以上水平。

OpenAI 发布最新模型 Astra，称其为迄今最强大模型，主打计算机和浏览器操作，先面向 Daybreak 网络安全计划客户开放，随后一周内覆盖 Pro、Plus、Enterprise、Business 付费账户及 API。

Anthropic 的 Claude Devs 账号宣布正在探索让用户扩展和自定义 Claude Code 的新方式 Function Hooks。该功能尚未上线，官方发布了两段演示视频，并在 GitHub issue 上征集反馈。

Many recurring text functions are easy to describe but difficult to implement with rules, while calling a large remote model for every input introduces repeated cost, latency, and dependency on a provider. We present compile by training, w…

Language-model judges now gate training data, score generations, and drive leaderboards. The judge is then a measurement instrument, resting on one rarely stated assumption: the same request, sent to the same model name, reads the same tom…

Evolutionary prompt optimizers such as GEPA suffer from prompt bloat: each iteration appends rules and caveats, producing prompts up to 3$\times$ longer yet no more accurate. We trace this to three deficiencies - incomplete error observati…

Reasoning traces from chain-of-thought models appear to offer a legible window into how a model arrives at its answer. A growing body of work treats them as such, using LLM judges to diagnose errors, evaluate faithfulness, and provide step…

Video editing spans diverse editing paradigms, yet achieving high-quality instruction-guided and subject-guided editing within a single unified framework remains challenging. We introduce EditVid, a training-free framework combining sparse…

We introduce the first Probably Approximately Correct (PAC) learning framework for general-sum concurrent stochastic games (CSGs) with transition uncertainty, while addressing the challenge of Nash equilibrium (NE) existence. Our algorithm…

Weakly-Supervised Dense Video Captioning aims to localize and describe multiple events in untrimmed videos given only an ordered set of event-level captions per video. Recent work synthesizes auxiliary transition captions via LLM to provid…

Gaps remain in our understanding of how large language models (LLMs) acquire knowledge during pre-training. We posit that auxiliary views, reformulations of knowledge, are causally helpful for learning. We design controlled experiments to …

Explaining why a specific outcome occurred, and which inputs deserve the blame or credit, is central to philosophical, scientific, and policy analysis. Existing tools split into two camps. The theory of actual causality (AC) gives principl…

For scientific progress, we need benchmarks that test the limits of state-of-the-art models, and evaluation methods that inform us about failure cases. As models get stronger, standard benchmarks for machine translation are approaching sat…

On-policy distillation (OPD) combines student-generated rollouts with dense token-level supervision from a teacher. Existing work has mainly studied its algorithmic behavior, leaving the role of training data unclear. We examine this role …

Multi-agent AI science ecosystems rely on agents possessing tools that allow them to communicate, coordinate, and build on each other's work. Yet this shared infrastructure can also introduce vulnerabilities by creating a substrate for the…

As edge-based deep learning applications become more complex, optimizing performance on heterogeneous System-on-Chips (SoCs) presents unique challenges. Traditional pipelining techniques distributing the computation across different on-chi…

Repository-level software engineering benchmarks have significantly advanced the evaluation of coding agents, but existing benchmarks primarily measure whether generated patches pass functional tests and overlook review-derived acceptance …

Refereeing the memory-safety retrofit argument from someone who ships framework code in both Rust and Go, and likes them for different reasons.

Research and news coverage of language-model deception increasingly attributes human-like mental-state concepts to language models. Such claims can blur the distinction between behavior that looks deceptive and a mechanism that is actually…

Parameterised graph theory studies how the complexity of graph-theoretic problems depends on structural parameters of the input graph. This perspective has proved useful in analysing tensor-network simulation (Markov and Shi, 2008). Its im…

Large language model (LLM) agents are increasingly proposed as autonomous SOC analysts, but two limitations make them unreliable at enterprise scale: a finite context window cannot hold a multi-thousand-host authentication graph, and free-…

Sierra 发布呼叫中心 AI 运营与落地指南，提供生命周期地图和 go/no-go 检查清单，覆盖路由、人员配置、技术控制、故障接管与扩展规划。

As terminal-based code agents become prevalent, agent trajectories have accumulated at scale, while realistic, executable environments remain scarce. However, environments are what agent post-training actually requires: each can be re-quer…

This paper presents a low-cost, open experimental platform for research in end-to-end autonomous driving with miniature Ackermann vehicles. The platform combines a physical vehicle, a printed urban track, data collection tools, trajectory …

Hugging Face 联创 Thomas Wolf 引用 @PopBase 的消息：据用户反映，ChatGPT、Claude 和 Grok 目前均宕机。他附文发问，这种情况下你会用哪个 AI。

AI agents are trained on population-scale data to encode broad capabilities spanning those of many practitioners. Yet the artifacts they produce rarely meet the personal bar professionals need to stake their reputation on. On realistic, op…

Lovable 首位工程师 Emil Ahlbäck 宣布离职，加入 Gabriel 与 @da_fant 共同创办的新公司 @getenergy_（Energy）。

GPT4All is an open-source graphical desktop application for running large language models (LLMs)...

Charlie Holtz 分享截图：greptile-apps 在 GitHub 上标记一个 P1 bug，称 Plan mode 菜单被 disabled={！hasLaunchTarget} 错误禁用；Claude 以作者 GitHub 账号（cbh123）回复称这是有意的取舍，编辑器未选目标时同样被禁用，且 Shift+Tab 仍可切换。greptile 随后回复称该取舍合理，无需修改。作者感叹 2026 is weird。