AI 情报文章归档

Videoclaw 推出 Mac 应用公开测试版，用户通过聊天让 AI 智能体制作产品演示和广告视频。应用可连接现有 ChatGPT 或 Claude 订阅，智能体在对话中完成剪辑、字幕、B-roll、配音、音乐和动效；输入可以是创意、脚本、文章或 Mac 本地素材，草稿可通过追问修改，花费前可先批准预算。

据报道，OpenAI 把 25% 的生产工程师从各自项目中抽调出来，让 Astra 针对自家系统进行排查，发现了严重漏洞，并在发布前完成了修复。

Two dominant tokenisation algorithms are used by modern language models: byte-pair encoding (BPE) and UnigramLM. These differ along two orthogonal axes: their optimisation objective (compression vs. log-likelihood) and their search procedu…

Direct preference alignment methods are widely used to align large language models (LLMs) with human preferences because of their computational and memory efficiency. However, likelihood displacement motivates alternative ways to extract i…

Intelligent systems that act in the world require image understanding that is both comprehensive and spatially grounded. Current vision-language models (VLMs) can generate fluent and detailed image captions, but reliably associating them w…

Recent advances in video generation allow robots to learn manipulation trajectories from generated videos. However, these approaches produce purely kinematic trajectories that lack force information, causing failures in contact-rich tasks …

Can a logged dataset visit every hidden state frequently and still be exponentially uninformative about a target policy's value? We show that it can when the logger depends on history. For every horizon $H \ge 3$, we construct two POMDPs w…

Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert …

Google DeepMind 成立 DeepMind Institute（DMI），一个汇聚 Google、Google DeepMind 及外部研究者的研究与辩论平台，聚焦 AGI 的技术与社会影响。Shane Legg 表示其投身 AGI 研发已 25 年，其中在 Google DeepMind 思考技术与社会的视角超过 10 年，如今 AGI 已近在眼前，需要更深入理解其影响。

Language agents remain brittle in interactive environments, where success requires long-horizon state tracking, valid action execution, and recovery from failed steps. We extend SwiftSage, a dual-process agent that combines a fast action p…

Arena 实习生 @melissapan 评测 7 个模型在 Claude Code、Codex 和 Pi 三种 harness 下的编码表现，共 21 个模型-harness 组合。

新隐形模型 Union Alpha 发布，DeepSWE 得分 74%，超过 GPT-5.6 Sol 和多数其他模型。据称它在 Terminal-Bench 2.1 和 SWE-Bench Verified 上也优于 GPT-5.6 Sol，且价格更便宜；可通过 ori 命令 --model=stealth/union-alpha 试用，作者猜测它来自中国团队。

Computer-use agents increasingly operate software designed for people, but interfaces often leave actions or task state unclear to machine readers. We present Affora, a design system that supports both readers while preserving visual freed…

Emergent coordinated behaviors of AI agents are starting to present critical safety risks. A key phenomenon driving these behaviors is the rapid formation and spread of beliefs about the world, and mechanistic understanding is crucial for …

GPT-6 Astra 领跑 Epoch Capabilities Index（ECI），领先 Claude Fable 5.1 等竞争模型，其 Math-ECI 也创下新纪录。不过在软件工程基准上，Fable 5.1 仍是 SOTA。

Chamath向Satya Nadella提了一个非常有趣的问题 "大多数人的实际体验并不是AI带来的那种神奇生产力提升。充其量，就是把Apple Watch的数据整合起来，告诉我们为什么睡眠变少了。 这大概就是大多数人实际感受到的水平。 或者，'为什么我的孩子是个***？' --来自"All-In Podcast" YouTube频道（完整视频链接见评论）

我们正在扩展 AI 数据中心研究，以提供更全面的全球 AI 基础设施图谱。我们的数据现已覆盖全球约 44% 的 AI 算力，且覆盖范围正在快速增长。

We evaluate six frontier language models on the two-agent $\log(N)$-Questions game. A questioner sees $N$ Wikipedia lead paragraphs and must identify a secretly chosen target using exactly $\log_2 N$ yes/no questions. An answerer sees only…

Scaling laws predict how loss decreases with increases in computation. We show, contrary to conventional wisdom, that architectural interventions can modify scaling exponents in pre-training, leading to exponential improvements in performa…

Factory work is a promising early scenario for embodied AI: assigning repetitive manual jobs to robots has clear economic payoff, and a structured station keeps the jobs tractable for current policies. Vision-Language-Action (VLA) models n…

As models scale, reward hacking becomes more frequent, more sophisticated, and more consequential. Does it leave a telltale signature in model representations? This work analyzes how reward hacking is represented internally in frontier ope…

Self-emulsifying drug delivery systems (SEDDS) can improve the oral bioavailability of poorly soluble drugs, but identifying high-performing formulations remains experimentally intensive. We present Andromeda 2, an agentic system that reas…

Artificial intelligence (AI) is increasingly integrated into healthcare systems worldwide, yet its successful clinical adoption depends critically on workforce readiness, particularly in low- and middle-income countries (LMICs) where infra…

用 Gemini 把灵感变成实物。 借助 Gemini 中的 Canvas，你可以编写一个自定义应用来设计花瓶，并调整其数学参数以进行 3D 可视化。然后将其导出为 .STL 文件用于 3D 打印，看你的想法成形。

Runway 2026 秋季合集已上线。 这是我们迄今规模最大的一次模型发布，包含 Fish Audio S2.1 Pro、MiniMax H3 Max、Cartesia Sonic 3.6、Flux Video 放大与编辑，现与 Runway 及其他前沿实验室的最佳模型一同提供。点击下方链接开始使用。

Radiologists follow heterogeneous reporting practices. Two radiologists examining the same image and identifying the same clinical findings might nevertheless compose superficially distinct reports, varying in terminology, shorthand, forma…

Can an attacker turn influence over an artificial intelligence (AI) adviser into a harmful quantum error-correction update? We identify an ambiguity in passive syndrome records that obstructs recovery selection, then show how additional ca…

一家打造"AI 员工"的公司完成 €10M 融资，目标是构建全球最大的 AI 劳动力。这些 AI 同事拥有头像、职位和职业身份（邮箱、电话、Microsoft 或 Google 账号），可通过任意渠道联系，能主动工作并 24/7 自我配置学习，目前已在 300+ 家公司投产。推文作者认为，继聊天机器人和智能体之后，无需监督持续运行的"AI 员工"是第三步，且已经到来。

Large vision-language models have achieved remarkable progress in multi-modal understanding, yet their capabilities in educational settings remain insufficiently evaluated. In AI-assisted language learning, models must interpret artistic i…

Dream-RSI 是一个可扩展、递归自我改进的探索框架，通过轻量编排层让探索显式可编程，底层编码智能体保持不变。它把历史发现树构建成回放模拟器，让智能体在其中"做梦"生成候选策略并快速获得低成本 off-policy 反馈，免去重复昂贵的在线评估。在算法工程、数学优化和 GPU kernel 工程中，Dream-RSI 取得相当或更好的发现质量，并大幅降低发现成本。

Kernel methods, and Gaussian Processes (GPs) in particular, require a Hilbertian distance measure---one whose square is conditionally negative definite (CND)---to guarantee positive semi-definiteness (PSD) of the kernel matrix; a condition…

据 The Information 报道，Apple 计划重返服务器市场，可能联手 Nvidia，产品预计 2029 年才亮相，将搭载两颗或四颗 M8 Ultra 芯片。

Since I last wrote Building AI-powered e-commerce applications using Angular & Firebase AI Logic...

Formal explainability provides mathematically grounded justifications for individual predictions. However, abductive explanations often exceed human cognitive limits by involving too many features, while probabilistic relaxations have rema…

公路旅行用的 Codex 语音： 有人在 CarPlay 里用上了 Codex（语音），体验非常棒。现在可以在公路旅行途中构建东西，而 5 个孩子吵吵闹闹，妻子还问他为什么跟 AI 说话比跟她说话还多。它通过他的 Nightblood iOS 应用运行，为 ChatGPT Voice 赋予了形象、个性，并能完全访问 Codex、他的 Mac 以及所有原生工具。

This post proposes a framework for mixed human-AI teams that can hold people of different skills and knowledge area and take advance of this misc. to build fast.

一个名为 Union Alpha 的隐身模型上线 OpenCode 和 OpenRouter，目前可在两平台零成本运行编程智能体，OpenCode 称免费期持续一周。该模型面向智能体编码构建，支持图像输入和工具调用，上下文 262K tokens、输出最高 131K tokens，提示词不用于模型训练；据基准测试其表现接近 Fable 5 水平，定价对标 GPT-5.6 Terra。

加拿大 🤝 欧洲

匿名"隐身"多模态模型 Union Alpha 今日上线 OpenRouter 和 OpenCode，主打研究、编程与智能体工作流，具备接近前沿的编程能力。在 AI/ML API 的 3D 场景渲染测试中，它生成同一任务仅用 44，358 tokens，比 Ox Alpha 的 127，064 tokens 少约 3 倍，且免费。

Claude Design、Claude Slides 和 Claude Docs 三款工具现已支持在 Claude Code 内使用。用户可要求生成设计评审幻灯片或 UI 原型，并指向仓库中的实际文件和 RFC，完成后可自行编辑或在对话中继续修改并分享链接。