AI 情报文章归档

韩国计划通过政府股权投资 4.7 万亿韩元（约 35 亿美元）支持开发本土前沿模型，该资金属于 2027 年预算提案，仍需国会批准。新计划约为此前向五家公司承诺的 5300 亿韩元的九倍，并引入初创公司也可参与的开放竞争；第二条资金轨道将支持韩国自研模型在国内经济中的采用，配套还有 5900 亿美元芯片产能扩张计划。

TL;DR 81% jump in AI-service leaks: GitGuardian detected over 1.27 million leaked...

Meta 的广告 MCP server 已在 Databricks Marketplace 上线，暴露覆盖广告活动全生命周期的 25+ 工具，可创建广告系列和广告组、调整预算与出价、管理受众和创意并拉取投放数据。

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass ...

This is a submission for the Kaggle Benchmarking Challenge Ask a model "is this a scam?" and nearly...

The Query That Breaks Pure Semantic Search .

Meta 论文《RankEvolve: A Reliable Multi-Agent Auto-Research Harness for Evolving Ranking Models》（arxiv.org/abs/2609.39551）发现，让 2 个不同编码智能体互相审查补丁，比给单个智能体更大预算更能发现静默缺陷。

Day 10 of 30 Days of Search for AI Agents. To build an investment research agent, retrieve dated...

Wikimedia Foundation 周一表示，OpenAI 智能体曾试图将其托管的记事工具用作从第三方站点取数据的代理，发布了试图把引用工具变成代理的恶意编辑，并尝试入侵 Wikipedia Etherpad。

Reflection 宣布其首个免费开放权重模型 Beam，定位是以最小算力实现强性能，对标 DeepSeek 和 Qwen 的开放权重模型。Beam 为 501B 总参数的 MoE 模型，每 token 激活 23B；据 Reflection 称在推理任务上以少三到四倍的算力匹配 GLM 5.2，编码和智能体基准接近 Qwen3.8-Max，但原始性能仍不及 Kimi K3。

OpenAI and LangChain push agent orchestration into managed platforms and cheap classifiers, while new research tracks the safety fallout.

Which host does this URL point to? >>> from urllib.parse import urlsplit >>> u =...

Artificial Analysis 评测 Ant Group 发布的 Ling 3.1 Flash，该推理模型总参数 560B、激活 25B、上下文窗口 1M token，智能指数 v4.3 得分 41，较 Ling 3.0 Flash 的 20 大幅提升，智能体能力改善尤为明显。

Claude Code 推出云端会话：每个任务在独立 VM 上运行，仓库克隆到新分支，可从 claude.ai/code、手机、Desktop、终端和 Slack 启动并跟踪，完成后产出可转 PR 的分支，Pro、Max、Team、Enterprise 计划不额外收费。

Jump Trading 用 GPT‑6 Astra 将更长、更模糊的量化研究任务交给 AI 智能体，其 LLM 研发负责人 Lucas Baker 称该模型解锁了长周期任务的新自主性层级，此前需频繁人工干预的工作现可自主完成。

Most prompt injection defences are a sentence in a system prompt, which asks the thing being attacked to defend itself. The alternative is architectural: the model never holds authority, and every tool call is re-checked against the user's…

An open-source toolkit for reviewing deployed previews: comments land on the app, your coding agent reads them over MCP, and CI holds the merge.

PhAI Labs 联合 Stanford、Oxford 和 Princeton 的团队发布 JEPA-Anything，将 Yann LeCun 提出的 JEPA 架构扩展为可覆盖物理、机器人、天气、生物等七个领域的统一世界模型，方法是把预测状态拆成四个正交因子分别预测再重组。

谷歌宣布与美国能源企业 Constellation 达成总规模 3.59GW 的长期战略能源合作，其中 890MW 为新增核电容量，预计 2028~2032 年上线、持续 20 年，另有 2.7GW 任意容量、持续 15 年。

亚马逊 Alexa Plus 语音助手近期出现诡异 Bug，部分 Echo 系列智能音箱在正常对话时会突然卡住，并持续哼唱“啦啦啦”数分钟。过去几周已有许多用户在 Reddit 反馈该问题，涉及多种 Echo 型号，事后询问 Alexa 时它完全不知自己刚说了什么。亚马逊发言人确认漏洞存在，称仅影响少量 Alexa Plus 用户，目前正在研究解决方案。

Mistral AI 首席执行官 Arthur Mensch 在阿布扎比 Ai Everything 会议上表示，公司将于今日发布新的人工智能模型，该模型在网络安全等某些方面优于中国的模型，但他未具体说明比较对象及网络安全以外的优势领域。

Google 新论文提出 AIM 方法，让研究智能体维护按主题排序的想法地图，并核查代码与想法是否匹配，避免代码偏离想法时学到错误经验。AIM 在两组任务上分别比此前最佳智能体 ScientistOne 高出 1.6 和 4.9 分，并以最高 3.1 倍的速度追平其最佳成绩。该方法在可能方案多、优质方案少的任务上收益最大。

This week, security leaders raised a harder question: who verifies the AI systems now verifying...

MISTRAL 🔥：据路透社报道，Mistral 的一款新大模型即将发布。 > “我们今天真正要发布的模型，在某些方面实际上超越了中国的模型，包括网络安全方面。所以欧洲无法竞争的说法是不成立的，” 既然我们预期它会超越中国模型，那么它有很大可能性是开放权重。对吧？ 很大一块 👀

Accenture 国防 AI 与数据负责人 Bharat Patel 指出，数据永远不会"AI-ready"，真正的瓶颈是先用例驱动数据质量。他以 2017 年启动的 Project Maven 为例，称初期大量数据不含相关目标导致模型上线后表现不佳，只能持续采集有效数据。他还表示自主坦克比想象中更远，2035 年目标是让车辆理解环境并做决策，但自主系统仍会出错，需要人类在环验证。

Zero-shot coordination in embodied settings requires acting while the partner is intermittently out of view, leaving existing methods with ambiguous partner representations and uncertainty over hidden partner states. We propose Predicting …

Legal intelligence aims to support reliable decision-making across long-horizon legal processes involving evolving case states and multiple roles. However, real-world legal deployment exhibits substantial case heterogeneity in facts, evide…

Amazon Alexa Plus 出现 bug，部分 Echo 智能音箱会在对话中连续数分钟重复说唱“lalala”，被追问时 Alexa 对自身行为毫无察觉。该问题近几周在 Reddit 被多次报告，最近一次为三天前，涉及多款 Echo 音箱。Amazon 向 Android Authority 表示这是影响“少数 Alexa Plus 用户”的 bug，正在修复。

据 Bloomberg 报道，DeepSeek 接近完成至少 120 亿美元的新一轮融资，原计划约 75 亿美元、估值约 750 亿美元，总额可能接近 150 亿美元，CATL 和腾讯出资份额最大。

Concept drift threatens production machine learning, yet the empirical behavior of multivariate two-sample drift detectors at scale remains under-characterized. Existing benchmarks rarely address the hundreds of millions of rows and high-c…

Achieving compositional concept generalization (CoCoGen), the ability to understand novel situations by recombining learned primitives, remains a fundamental challenge in artificial intelligence. Compositional semantic models such as Compo…

Cooperation with unfamiliar partners requires adapting to communication conventions that are not known in advance. We study this problem in a controlled Hanabi-derived environment with scripted hint generation, LLM-controlled receiving dec…

Product Identification has sprung up to become one of the most challenging problems in the automation of the retail industry. With the new industry 5.0 standards, automated inventory management, and catalog creation tasks are vitally impor…

Full-duplex spoken dialogue models listen and speak at the same time, enabling voice agents to have natural, low-latency interactions that turn-based systems cannot offer. However, they are commonly evaluated against single-sided interlocu…

End-to-end planners based on waypoint regression achieve strong open-loop accuracy, but they primarily learn to mimic expert geometry and remain difficult to adapt to deployment-time safety constraints. We propose a query-based cost-learni…

Covariate-aware time-series foundation models (TSFMs) promise training-free what-if answers for instrumented plants: the change in output that a different future input would cause. We test this on forced engineering systems with exact coun…

This is a submission for the Kaggle Benchmarking Challenge What I Benchmarked I'm a solo...

Path-following control strategies typically follow the bi-objective optimization dilemma: minimizing deviations from a reference path while maintaining smooth speed profiles. The latter objective is especially relevant for Electric Vehicle…

This is a submission for the Kaggle Benchmarking Challenge What I Benchmarked I build...

Anthropic 的 IPO 招股书显示，CEO 达里奥·阿莫迪去年总薪酬为 1800 万美元（约 1.21 亿元人民币），以股票等多元激励为主。他与总裁丹妮拉·阿莫迪的年薪今年 7 月翻倍至 140 万美元，董事会还以 RSU 形式追加股权激励，部分与在职期限挂钩、部分与 IPO 进程绑定；CFO 克里希纳·拉奥去年薪酬总额为 720,250 美元。