AI 情报文章归档

Vision-Language Models (VLMs) exhibit remarkable zero-shot generalization, yet they often encode unwanted or hazardous stylistic domains such as idealized textbook diagrams in medical AI or cartoon vehicles in autonomous driving. Approxima…

Contemporary large language model (LLM) chat systems treat conversation history as an immutable sequence of turns that defines the model's working context. However, user intent in real interactions is not static: it evolves through correct…

Time-series carry structure simultaneously at multiple scales (fine-grained variation, mid-range motifs, and global properties) and downstream tasks operate at correspondingly different scales. Most existing self-supervised learning approa…

Humyn Labs 推出 BRIDGE ASR 2.0，用真实双人对话评测 23 个语音识别模型，覆盖 18 种印度语言及西班牙语、葡萄牙语、越南语。其核心指标 code-switch F1 检验句中混用的英文词是否保持英文，排行榜支持按语言和指标筛选，方法论与评测数据公开。

Large video diffusion models offer expressive priors for embodied prediction and learning, yet their many-step sampling remains costly for interactive downstream use. Distribution Matching Distillation (DMD) enables few-step video generati…

美团 LongCat 发布 LongCat-2.5-Preview，作者转发时指出定价与之前一样。该模型 1.6T 参数、约 48B 激活、1M-token 上下文窗口，原生多模态，面向长程任务，覆盖终端、浏览器、GUI、电子表格和设计工具等场景。API 入口 https://longcat.ai/platform/ ，对话入口 https://longcat.ai/chat/ 。

Retrieval-augmented generation (RAG) is often used to address outdated knowledge by providing external evidence. But retrieval helps only when that evidence is still valid. We identify a temporal alignment failure, stale-document poisoning…

Whether an information extraction pipeline should process page images or parsed text depends on the document, and the answer flips across the layout spectrum. We study this trade-off under a constraint that rules out (closed) cloud service…

Have you ever asked AI to do something very specific and it did almost everything right? Except the...

Morphology--control co-design jointly optimizes an agent's body structure and control policy as an integrated embodied system. However, existing methods typically model morphology design and control with separate networks coupled only indi…

🌕中秋快乐🌕

Meta 发布个人助理 Muse 并推出配套掌上设备 Muse Charm，计划假期前上市，个人智能体开始成为大众消费产品。Shopify 与 Muse 达成智能体结账合作，股价涨超 10%，但 Amazon 宣布屏蔽 Muse 购物；智能体创业公司 Instinct 刚完成 25 亿美元估值融资，据报正寻求以 100 亿美元估值再融资。

Perplexity Research 发布后训练研究，将拒绝采样微调与提示引导自蒸馏（OPSD）结合，用 GLM 5.2 驱动的 Perplexity Computer 真实用户会话（含失败会话）训练模型。

Ordinal acne severity grading requires distinguishing visually similar neighboring grades while jointly weighing holistic facial appearance and localized lesion burden - evidence that most existing approaches collapse into a single opaque …

The Information 报道称，特斯拉近几个月已将 Optimus 人形机器人产量扩大到原来的约 10 倍，上个月每周可生产数百台，而第二季度测试时每周仅几十台，但手部结构、自动化设备和供应商限制等规模化生产难题仍未解决。

Continual spatio-temporal forecasting supports traffic management and environmental monitoring under evolving dynamics and expanding sensor networks. However, conventional graph-based continual learning methods tie forecasting representati…

For most of human history, studying Earth meant looking backward. Scientists collected observations,...

AI agents combine language models with external data and tools that can modify files, call APIs, or execute code. Security failures can arise when adversarial content changes an agent's tool use or when the surrounding software contains vu…

Microsoft 宣布 Autopilot 预览版向首批客户推出，定位为在你不在时也能持续工作的持久、主动的个人智能体，由原 First Scout 团队负责构建。

用 Astra 作为编排器在电脑上玩游戏。效果非常好。

Unobserved common causes are pervasive in real-world time series and can induce spurious associations that causal discovery methods mistake for direct edges. We propose LUCID (Learning Under Confounding for Inference and Discovery, a regim…

Vision-Language-Action models (VLAs), while showing strong potential for robot control, require massive amounts of high-quality imitation learning data. Moreover, the absence of an explicit world model casts further doubt on their control …

德国柏林警方9月24日在犯罪高发的科特布斯门地区启动AI监控摄像头计划，前四周安装调试后正式投入使用，后续计划推广至市内其他高发区域。该摄像头通过空间人工智能技术分析人员动作识别潜在犯罪与暴力行为，识别到异常时显示屏亮起并向警员推送信息。今年2月德国多部门已宣布引入AI技术加强对有组织犯罪的打击。

美团 LongCat 发布 LongCat-2.5-Preview，总参数 1.6T、激活约 48B，提供 1M token 上下文窗口，原生多模态，面向长程任务，覆盖终端、浏览器、GUI、表格和设计工具。现已开放 API（https://longcat.ai/platform/）和聊天入口（https://longcat.ai/chat/）。

TechCrunch Disrupt 2026 为被裁员工提供 75 美元的 Expo+ 通行证，限前 100 名符合条件者，或截至 10 月 13 日 8 a.m. PT 大会开幕。

Tabular in-context learners such as TabPFN, Mitra, or ConTextTab rely on alternating row and column attention over 2D sequences of latent embeddings. These attention patterns differ markedly from the one-dimensional case in language models…

We study geometric moment contraction (GMC) of the constant-parameter stochastic Nesterov recursion \[ Y_k=Θ_k+β(Θ_k-Θ_{k-1}),\qquad Θ_{k+1}=Y_k-γG(Y_k,X_{k+1}). \] Under mean strong monotonicity and stochastic $L^p$ Lipschitz continuity, …

Large language model agents increasingly act on software systems, no longer merely generating text but also changing databases and online services. However, an approved database update may succeed yet leave an unapproved notification becau…

Ethan Mollick 转发并评论 Satya Nadella 宣布的 Copilot 史上最大更新，包括 Autopilot、Code、Home、Office 四部分，以及在 Teams 中调用 Copilot 和主动呈现 M365 重要信息的 Today 功能。

By Tanbir Hossain Ramim. Project page: VoiceMax. Source: github.com/TanbirRamim/VoiceMax. VoiceMax...

微软论文《Coding Agents are Strong Prompt Optimizers》提出 Coding-Agent Skill Distillation（CASD），让现成编码代理读取完整智能体轨迹日志，自行编写分析代码统计失败模式并将发现写成规则提示词，无需环境访问和验证数据。

微软等机构提出 Taste-Bench，衡量前沿智能体在长任务决策点能否选对方向，最佳模型正确率仅 59.7%。该基准从工程与研究运行中的并行尝试和绕路自动挖掘分叉点，越依赖轨迹后段证据的分叉越难，加大推理预算也无法提升准确率。作者将教师模型对结果的判断蒸馏进学生模型，在留出的 SWE-bench Pro 任务上提升了端到端成功率。

If you read my last post about how we keep Mautic's docs up to date with Promptless, you already know...

One base, many sources Raku documentation does not live in one place. The language docs...

新 Copilot 背后的动机：智能正在加速发展。现在我们需要把它普及到各处。

n8n and NVIDIA will demonstrate a practical approach to governing AI agent actions at NVIDIA GTC...

AI 开启了一个新的学习时代。 Jamie 在经营 Crimson Education 的同时攻读了 12 个高级学位。 但在创业中学习本身就是一种技能。传统学校很少提供这种机会，而有了如今这么强大的 AI，没有什么能阻挡你。

听说 AI 监管的风声渐起，尤其是带着一种“开源危险、闭源安全”的错误论调，所以觉得有必要重发这篇。有一些方法可以让 AI 行业更安全，同时不削弱透明度、教育和竞争。https://www.interconnects.ai/p/banning-open-source-ai-would-be-a

How to build an app where two models get the same CSV, write matplotlib code, get sandboxed and...

阶跃 Step-5-Preview 实测显示输出稳定性突出，后训练扎实，写工程代码很少反复；在“硅基交警”Agent 测试中全程未指挥出事故，后端 Agentic Coding 与“硅基骑手”多轮测试得分 Δ 很小。其 Agent Loop 迭代能力可卡着极限数值优化，例如把送餐超时理解为送餐时间+5分钟以多赚积分。短板是前端、3D 场景与美学，较上一代有提升但距本代 SOTA 仍有差距。