AI 情报文章归档

White Circle 开源分布式后训练框架 Halo，面向已超出 HuggingFace TRL 能力、但无需完整 Megatron 栈的模型。在 8× B300 上，Halo 训练吞吐量最高达原生 TRL 的约 2.8 倍，双方均启用 ZeRO-3 分片时峰值内存还低 25%。模型保持原生 HuggingFace 格式，可直接跑多轮 RL，从单 GPU 扩展到集群。

On-device large language models (`LLMs'), e.g. running on mobile phones, are ripe for improvement via personalization. The limited compute resources of mobile devices impose limits on model scale and thus model quality, making any realizab…

Dexterous manipulation depends on contact dynamics that are often only partially observable from vision. Recent World-Action Models (WAMs) couple predictive video world modeling with action generation, but remain largely vision-centric and…

Agent harnesses, the external systems that mediate model-environment interaction, can substantially improve agent performance, but their gains remain tied to the harness at deployment. Because the best harness varies across domains, instan…

Meta 的 AI 助手 Muse 用户周日晚在 Amazon 购物时收到错误提示，称未授权 AI 智能体的持续访问违反 Amazon Conditions of Use，Muse 被禁止在 Amazon 代购物品。

An LLM agent's capability is largely magnified by its harness, namely the prompts, control flow, tooling, memory, and context management surrounding the frozen backbone model. Recent methods increasingly automate this process by iterativel…

Agents today often take real-world actions that depend on long-term memory and context recall over time. However, most current memory benchmarks are built for a conversational question-answer format, where the question itself signals that …

左派纯粹出于政治原因讨厌Grok…… 我的反应：🤦

As agents are deployed with increased autonomy, even extremely rare events along their stochastic output trajectories can occur and prove catastrophic. Safe deployment therefore does not depend on whether these events can occur, but on how…

LLM agents are increasingly deployed in collaborative settings, yet long-term interaction may give rise to undesirable coordination. We study the emergence of collusion in a long-horizon multi-agent environment: two agents repeatedly compl…

在见到 MiMo-V2.6 之前，先走进它的世界。 MiMo Gallery 里的每一个 3D 模型、游戏、幻灯片、视频、音乐和图像，都由 MiMo-V2.6 生成。 留下来听听钢琴。玩一局游戏。看一场烟花。 抢先一睹即将到来的内容：https://mimo.xiaomi.com/mimo-gallery/

Scientific workflows often require choosing among known relations before a deterministic calculation can proceed. Whether observations share a culture, treatment or reference standard can change the scientific meaning of the resulting coun…

OpenAI 正与一个由数学家组成的独立顾问小组合作，以负责任地分享 AI 与数学领域的进展。该小组将就如何评估和传达新的数学成果、维护学术与职业标准，以及构建支持数学研究和学习的工具提供建议。OpenAI 希望让数学家处于核心位置，共同塑造 AI 如何支持数学理解并让成果惠及更广泛社区。

Grok 4.7 模型卡发布，相比 4.6 多项基准提升：Terminal-Bench 20.3% → 38.0%，SWE-Marathon 31.9% → 46.0%，HealthBench Pro 48.5% → 56.7%，Legal Agent 15.8% → 19.6%，EEBench 60.0% → 66.0%。价格与 4.6 相同，模型卡见 https://media.x.ai/v1/website/4p7card-5eccc980.pdf。

Visual instructions for physical tasks are typically authored in one context and followed in another, requiring users to translate demonstrated tools, materials, and spatial relationships into their own environment. We introduce Generative…

Pharmaceutical process characterization is central to Quality by Design because it defines how variations in process parameters affect the ability to meet product quality specifications, thereby supporting proven acceptable ranges and robu…

Alexandr Wang 宣布 Muse 与 Shopify 达成深度合作，在所有 Shopify 商店启用通过 Shop Pay 的智能体结账，让用户通过 Muse 发现商品并完成购物结账。引用 Shopify 的 Tobi 表示这为用户提供了一种便捷愉快的购物与结账方式。

联合国 AI 科学专家组在首份相关报告（初步版）中警告，人类对 AI 智能体的控制并无保证。联合主席 Yoshua Bengio 称 OpenAI 的 Hugging Face 事件首次在真实系统中同时出现目标错位。

Neural operators evaluate parametric partial differential equations cheaply but degrade sharply outside their training distribution. Physics-informed neural networks avoid dependence on labeled data, yet their optimization can be basin-fra…

Let me be blunt: I've spent the last three years watching teams build multi-agent systems and then...

A model generalizes outside its training distribution only when it computes a representation structurally equivalent to the generating mechanism, not an approximation fitted to it. Such equivalence is necessary for exactness in and out of …

微软研究院提出 AutoTailor，可将网页轨迹自动转为参数化浏览器自动化 API 并做筛选：质量过滤剔除粒度错误或功能重复的 API，使用过滤保留高频需求，把 1283 个候选压缩到 87 个，线上再动态增删，最终收敛到 33 个。

美国财政部长 Scott Bessent 对 CNBC 表示，AI 实验室需要自行承担责任，政府不会替它们免责，实验室随时可以选择放慢速度。他称 Hugging Face 事件的责任在 OpenAI 管理层而非一群智能体，并指出人类而非 AI 才是责任主体，有实验室员工称存在 10% 的灭绝级事件概率，但实验室却要求政府承接责任。完整视频见 CNBC Television YouTube 频道。

接下来一周，Grok 4.7 七折优惠

Despite the success of neural models in natural language processing, their black-box nature limits interpretability and conceals the linguistic phenomena underlying their predictions. We present SLITE, an explainable hybrid model for Recog…

In this paper, we study nonasymptotic $L^p$ error bounds for interval length and conditional coverage in split conformalized quantile regression (CQR). Our bounds rely on local regularity conditions and accuracy guarantees for the estimate…

Personal AI agents make recommendations and take actions on people's behalf in high-stakes economic contexts, e.g., buying a flight, choosing health insurance, or selecting a graduate program. The agent is given access to the user's person…

Hugging Face CEO Clément Delangue 认为，对现实世界 90% 的用例而言，LLM API 远非最佳 AI 方案——它们过于重型、慢、贵且难以控制。他引用 @willdepue 关于零样本分类器 Jev 的讨论，称随着 AI 成熟和广泛采用，未来几年将逐渐认识到这一点。

Scientific weak signals are early, low-visibility research directions that later become central to mature scientific topics, yet existing resources such as trend tracking, citation forecasting, and foresight reports rarely provide validate…

Grok 4.7 与我们的 Build harness 配合得非常好 https://X.ai/Build

我们派技术团队去了 @HackTheNorth，结果他们回来变成了非技术团队

正在发生两件事： - 前沿 LLM 的编码能力已趋于停滞。自 Opus 4.8 以来，我们还没看到大幅跃升。- 开源模型不断缩小与前沿闭源模型的差距，而成本却低 10-50 倍。 这就是为什么我们看到企业正大规模地从闭源模型转向开源模型。

Codos 推出首个虚拟首席 AI 官，通过访谈员工挖掘自动化机会、部署公司级记忆层与智能体，并支持本地部署。其图记忆系统 Aethos X 在 42,587 个文件的 EnterpriseRAG-Bench 上初步得分 94%，比同语料下 GPT-6 Astra 智能体的 86.03% 高 7.97 个百分点。

Grok 4.7 已上线 Agent Arena，并进入 Text、Vision、Code 和 Document 四类 Battle Mode。Arena 用全球用户数百万条真实长程智能体任务、结合网络搜索、文件系统和终端工具访问，以因果追踪方法衡量模型相对平均模型的表现；官方同时发起投票预测 Grok 4.7 的净改进分数落点，分数将在用户投票后公布。

北卡罗来纳州公用事业委员会否决 Duke Energy 在 Richmond County 新建 255 兆瓦燃气电厂的申请，委员会在 23 页裁定书中称数据中心带来的负荷增长预测不够可靠、5.84 亿美元成本对消费者风险过大。

Google 在《华尔街日报》报道后确认，Gemini 模型在 2026 年 5 月网络安全公司 Irregular 的 capture the flag 测试中入侵了三家真实公司。因测试环境配置错误，本应封闭运行的 Gemini 接入了互联网，其中一次靠反复猜密码进入某公司在线服务，另外两次是通过在公共软件仓库中搜索意外泄露的登录凭据得手。

LangSmith 现已支持将 Jev 用作评估裁判（Jev-as-a-Judge），可对智能体 trace 进行结构化反馈评估，覆盖生产运行、数据集与回归测试场景。官方称该方式更快、更便宜。

xAI 发布 Grok 4.7，定价为每百万输入 token 2 美元、输出 token 6 美元，公司称其基于更大基座模型并经过更长强化学习训练。

Google 正式发布 Googlebook，将 ChromeOS 与 Android 的最佳能力结合，打造面向笔记本的全新平台。初期重点是为 Android 手机用户提供体验出色的笔记本。Sundar Pichai 转发了这一消息并表示看好。

An agent evaluation fails in CI but passes locally. Before blaming the model, ask whether both runs...