AI 情报文章归档

ML code should run where researchers choose, not where it was first written. Today, machine learning code mainly lives in two ecosystems, PyTorch on NVIDIA GPUs and JAX on Google TPUs. Migrating repositories between them means building and…

Interpretability, AI auditing and assessments of machine consciousness increasingly rest on mechanism-level quantities: functionals of a system's causal transition law that describe how it computes, not what it outputs. Experiments see tha…

To interpret a direction in a language model's residual stream, we can look for text that strongly activates it. But when such text is found by scanning a corpus, it is limited to what that corpus contains. We develop Maximally-Activating …

Lauren Tan 认为形式化方法只能证明代码符合特定假设下的 spec，而 spec 来自人的意图，无法被数学证明正确，建议对代码库做分层验证，结合类型约束、测试与形式化验证。

Adversarial Training (AT) is an important method for improving model robustness. We study the dynamics of deterministic Gradient Descent (GD) in AT. Most existing works in this direction focus on the small or vanishing step sizes regime, w…

On-policy distillation (OPD) trains a student on its own rollouts under a frozen teacher's per-token supervision. Does this dense signal reduce the training FLOPs needed to reach a target accuracy compared with reinforcement learning? We s…

作者曾在 ThinkPad T490 上按指南安装 macOS 26，用集成 Claude Code 和 OpenAI Codex 的 XCode 26 在一天内不写一行代码做出可用的 iOS 应用。

Reproducing machine learning experiments requires reconstructing procedures from information scattered across papers, released code, and supporting artifacts. For coding agents, reproduction success can therefore depend on their ability to…

Verne Robotics 发布开源权重世界动作模型 V0.5，可读取摄像头画面并输出下一秒机械臂、夹爪与移动底座的运行动作。该模型基于 Wan2.2 视频生成器、从超 170 万条演示中学习，不到 2 小时演示即可学会新工业任务。在 RoboTwin 2.0 随机杂乱、光照与纹理测试中，V0.5 从 81.0% 降至 73.4%，π0.5 则从 70.7% 降至 46.0%。

Claude Code Hooks: A Complete Guide to Automating Your AI Coding Workflow If you've been...

Coding with agents turned my job into one infinite loop of asking, reviewing and approving. I lost the fun part of solving problems. Here's what I got back.

The run said success. The dashboard showed three findings. The report read like a careful security...

A defender anticipating an attacker needs a probability distribution over the attacker's next action from information available before that action occurs. Two structures complicate this: the attacker's behavioral regime persists and can sw…

我们正在 Muse 应用中推出语音和表情虚拟形象，为你提供一种全新、自然的方式与你的个人智能体互动。 https://about.fb.com/news/2026/11/introducing-muse-ai-voice-expressive-avatars/

在 Connect 大会上，我们预览了 Muse Realtime Avatar，这是我们能将 Muse Realtime Voice 转化为富有表现力、可交互虚拟形象的实时具身技术。 今天，我们在美国面向 @Muse 应用推出该功能，为人们提供一种与个人智能体互动的更自然方式。 在我们的研究博客上了解更多关于 Muse Realtime Avatar 的信息：https://go.meta.me/47ZSNso

Microsoft Research 介绍 Proof-Driven Development Kit，用于解决 AI 智能体生成代码难以逐行审查的信任问题。该工具把开发者的自然语言意图转成 Verus 规范，沿用安全关键系统中的形式化方法，用数学证明尽可能验证智能体生成的代码是否符合规范，并明确标出证明无法覆盖的部分。研究人员在 MSR 博客中详细拆解了这一方法。

Slashing AI Agent LLM Bills by 75%: Sub-Millisecond Context Compression and Semantic Prompt...

Typed-decision models score options that the caller supplies at request time, so their accuracy depends on label wording that the model's developer does not control. We compare two releases of a 307M-parameter multilingual cross-encoder, R…

OpenAI 总裁 Greg Brockman 宣布与 JPMorganChase 达成战略合作，探索用先进 AI 支持其企业运营，包括帮助工程师构建、安全团队防御以及让员工和客户受益于更强 AI。合作特别聚焦网络安全领域，目标是保护金融系统；引用内容提到该行将使用 Codex 和 OpenAI 的 cyber 模型。

Meta 在 Muse 应用的移动、网页和 Mac 端推出此前于 Connect 2026 预览的实时语音对话功能，点击右上角电话图标即可开始。语音模式下可以边说边让 Muse 帮忙完善想法或排练面试、演讲；头像会配合语气做出表情和手势，用户可选仅语音、伴随头像或全屏显示，并有数百种声音可选。

Neural networks are likely to be among the most consequential artefacts in human history. As these systems become more capable and autonomous, their potential to cause societal harm also grows, making AI safety one of the defining challeng…

LLM-based medical assistants that follow a patient across visits need a memory that keeps the patient's clinical state correct as it changes. We examine the errors of five memory systems on longitudinal medical dialogue and find three recu…

Google 论文测试 RAG 系统的检索层投毒，发现攻击关键在于让伪造文档复制用户查询文本，是否附带隐藏攻击指令对检索结果影响很小。

Chat remains the default surface for many large language model (LLM) tasks, even as agentic tools such as OpenCode excel at coding and automation. Yet the model sees the current dialogue but not the workspace around it. We explore meta acc…

Vision-Language-Action models (VLAs) enable general-purpose embodied intelligence, but their performance is brittle under distribution shift, where unanticipated failures can damage equipment or endanger people. Deploying VLAs in real-worl…

Discovering continuous symmetries directly from differential-equation data can reveal geometric structure for model reduction and scientific machine learning without symbolic determining equations or a closed-form representation of the gov…

Molecular representations such as fingerprints and learned embeddings play a central role in cheminformatics and small-molecule machine learning. They serve as the primary input for a diverse array of tasks, including clustering, dataset s…

Coding agents are among the most economically consequential deployments of LLMs to date, making reliable evaluation increasingly important. Yet even benchmarks that are carefully validated at release can become stale. As models improve, th…

Tim O'Reilly 在 O'Reilly Radar 撰文，结合 Foo Camp 讨论与历史类比，主张当 LLM 让代码生成变廉价时，软件经济中可分享、可交易的知识产物正从源代码转向提示词、规格说明、Agent Skill、MCP server、agent trace 和 eval 等"context bundles"。

Low-rank adaptation (LoRA) is commonly used in both supervised fine-tuning (SFT) and reinforcement learning (RL), often under similar optimizer settings. Yet, how the different dynamics of the two stages manifest in the low-rank increment …

The growing use of AI in mathematics raises a question: can we detect AI involvement in proofs, separately from AI assistance with writing? Historical arXiv papers, recent papers disclosing substantive AI use, and manuscripts released by O…

Greg Brockman 转发并称赞团队的网络安全模型工作。被引用内容称，该模型除发现 948 个 Linux 漏洞外，还在 Firefox 中发现 111 个渲染器 RCE，其中 81 个为新漏洞，另有 30 个是开发者已知但被认为不可利用的漏洞的可利用利用程序。团队正用该技术扫描更多开源项目以维护互联网基础设施，并表示这只是开始。

Note This is a shortened version of the original writing on kig.re — my personal blog. Here is...

Google Gemini 宣布 Gemini Omni 1.2 Flash 视频生成模型即日起向所有 Google AI 订阅者推送，Gemini 应用内支持多轮对话式剪辑和更稳定的画面。Google AI Ultra 订阅者可单次生成并编辑最长 15 秒的视频；该模型由 Google DeepMind 更新，可先以 360p 或 720p 草稿（$0.07/s，成本低 30%）再放大至 1080p 或 4K，支持 3s 到 15s 任意时长、更自然的多人对话口型同步和更…

带作用域权限的电子邮件正成为 AI 智能体最好的接口之一，因为上下文已经写好了。 @viktor_com 这个 AI 智能体过去需要安装 Slack 或 Microsoft Teams，现在有了一个公开的电子邮件地址。 由于所有人共享这个地址，发件人身份就承担了访问控制：智能体只服从来自该用户且通过认证的邮件。

OpenAI 联合创始人兼总裁 Greg Brockman 在《华尔街日报》科技大会上表示，公司 2027 年可以做好 IPO 准备。他称第三季度年化收入环比增长 70%，企业业务翻倍，目前企业收入已占多数；此前企业收入仅约 30%。他承认 OpenAI 曾在 agentic coding 领域缺位，Claude Code 已抢占市场，并提到与 J.P. Morgan Chase 的近期合作。

Run untrusted LLM-generated code in a hardware-isolated sandbox with the network locked to an...

Anthropic 为 Claude 推出 Personal Finance 功能，无广告和赞助推荐，用于帮助用户理解个人财务，不用于提供投资建议。该功能正逐步向美国 Pro、Max 和 Ultra 计划用户开放。

Claude 个人理财功能现已推出，美国地区 beta 版。 连接你选择的账户，Claude 就能用日常语言回答你的财务问题，基于你自己的余额和交易记录。访问权限为只读，Claude 无法移动、管理或投资你的资金。

Claude 上线个人理财功能 Personal Finance，用户连接账户后，Claude 可解释消费、账单和余额情况。该功能为只读访问，用户可自选账户并随时解绑，目前处于 beta 阶段，正逐步向美国 Pro、Max 和 Ultra 套餐用户开放。