Daily curated AI news, research papers, open-source projects and industry trendshttps://www.ainewslog.com/https://www.ainewslog.com/blog/grok-daily-2026-11-15-en/https://www.ainewslog.com/blog/grok-daily-2026-11-15-en/AI Daily 2026-11-15. In this issue: Cursor details its security incident: it began with a key issued to Fireworks, and a tampered Claude Code extension sat in its marketplace cache for about 11 hours; On the eve of its…Sun, 15 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-15-zh/https://www.ainewslog.com/blog/grok-daily-2026-11-15-zh/AI 日报 2026-11-15，本期关注：Cursor 公布安全事件调查：攻击始于发给 Fireworks 的密钥，被篡改的 Claude Code 扩展曾上架约 11 小时；上市前夜，Amodei 对话摩根大通 Dimon：不想让 Anthropic 多值一美元，算力是“最深的经营难题”；美联…Sun, 15 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-14-en/https://www.ainewslog.com/blog/grok-daily-2026-11-14-en/AI Daily 2026-11-14. In this issue: OpenAI launches two security products: agents that pentest apps, and agents that turn findings into reviewable patches; LangChain, maker of AI agent tooling, raises $300M as its…Sat, 14 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-14-zh/https://www.ainewslog.com/blog/grok-daily-2026-11-14-zh/AI 日报 2026-11-14，本期关注：OpenAI 推出两款安全产品：一个派智能体做渗透测试，一个把漏洞直接改成待审补丁；做智能体“工具箱”的 LangChain 融资 3 亿美元，估值 13 个月涨到 30 亿美元；1450 亿参数模型塞进笔记本：Edge0 开源 145B/120B 本地…Sat, 14 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-13-en/https://www.ainewslog.com/blog/grok-daily-2026-11-13-en/AI Daily 2026-11-13. In this issue: WSJ: Nvidia's board is preparing to oust Jensen Huang; shares fall about 8% after hours; Baidu launches Ciyuansuo, a token exchange that verifies and credit-rates model resellers…Fri, 13 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-13-zh/https://www.ainewslog.com/blog/grok-daily-2026-11-13-zh/AI 日报 2026-11-13，本期关注：据《华尔街日报》报道：英伟达董事会准备罢免黄仁勋，股价盘后跌约 8%；百度推出模型 Token 交易平台“词元所”：给良莠不齐的模型中转站做验真和信用评级；昆仑芯公布路线图：M100 已上市，旗舰 M300 明年推出，2028 年换全新 N 架构Fri, 13 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-12-en/https://www.ainewslog.com/blog/grok-daily-2026-11-12-en/AI Daily 2026-11-12. In this issue: Alibaba teases Qwen4, opening a preview of flagship Qwen4-Max; Baidu releases ERNIE 5.5 Flash Preview: 280B total, 14B active, weights due by Nov 25; MiniMax opens H3-Image weights…Thu, 12 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-12-zh/https://www.ainewslog.com/blog/grok-daily-2026-11-12-zh/AI 日报 2026-11-12，本期关注：阿里预告 Qwen4，旗舰 Qwen4-Max 先以预览版开放试用；百度发布文心 ERNIE 5.5 Flash 预览版：280B 总参数仅激活 14B，权重 11 月 25 日前开放；MiniMax 开放 H3-Image 权重：直接生成 1600 万…Thu, 12 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-11-en/https://www.ainewslog.com/blog/grok-daily-2026-11-11-en/AI Daily 2026-11-11. In this issue: OpenAI introduces Control Room to stop research model output within 35 seconds; Australia plans AI safety regulator to enforce rules on frontier AI companies; Zyphra open-sources…Wed, 11 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-11-zh/https://www.ainewslog.com/blog/grok-daily-2026-11-11-zh/AI 日报 2026-11-11，本期关注：OpenAI 推出 Control Room，可在 35 秒内停止研究模型输出；澳大利亚拟立法设立 AI 安全监管机构，约束前沿 AI 公司；Zyphra 开源 ZONOS3-Mini 语音合成模型，可在手机和浏览器实时运行Wed, 11 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-10-en/https://www.ainewslog.com/blog/grok-daily-2026-11-10-en/AI Daily 2026-11-10. In this issue: Nominal raises $240 million Series C at a $3 billion valuation; Meituan releases LongCat-2.5 weights under an MIT license; China's CAC launches unified filing system for generative AI…Tue, 10 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/grok-daily-2026-11-10-zh/https://www.ainewslog.com/blog/grok-daily-2026-11-10-zh/AI 日报 2026-11-10，本期关注：Nominal 完成 2.4 亿美元 C 轮融资，估值 30 亿美元；美团宣布开放 LongCat-2.5 模型权重，采用 MIT 许可；网信办上线新技术新应用备案系统，生成式 AI 等四类备案一站办理Tue, 10 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-08-zh/https://www.ainewslog.com/blog/ai-daily-2026-11-08-zh/AI 日报 2026-11-08，本期关注：Anthropic 发布 Claude Agent Skills 公共仓库；Ant Ling 发布 Ling-3.2-flash：560B 多模态 MoE，支持 1M 上下文；Drex Max v1.1 声称在多模态决策模型上达到 SOTA 并位于帕累托…Sun, 08 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-08/https://www.ainewslog.com/blog/ai-daily-2026-11-08/AI Daily 2026-11-08. In this issue: Anthropic Publishes Public Repository for Claude Agent Skills; Ant Ling Releases Ling-3.2-flash: 560B Multimodal MoE with 1M Context; Drex Max v1.1 Claims SOTA and Pareto Frontier for…Sun, 08 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-07-zh/https://www.ainewslog.com/blog/ai-daily-2026-11-07-zh/AI 日报 2026-11-07，本期关注：NVIDIA 发布面向 AI Agent 上下文的 Auto Ontology 框架；TSMC 为逻辑芯片缓存开发 BEOL 氧化物半导体存储器；Harness Leaderboard 对 Kimi Code、Hermes、Pi、Claude Code…Sat, 07 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-07/https://www.ainewslog.com/blog/ai-daily-2026-11-07/AI Daily 2026-11-07. In this issue: NVIDIA Releases Auto Ontology Framework for AI Agent Context; TSMC Develops BEOL Oxide Semiconductor Memory for Logic Chip Caches; Harness Leaderboard Benchmarks Kimi Code, Hermes, Pi…Sat, 07 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-06-zh/https://www.ainewslog.com/blog/ai-daily-2026-11-06-zh/AI 日报 2026-11-06，本期关注：Microsoft 发布 VibeVoice-ASR-Streaming 统一流式语音模型；Claude 助力解析 Vault 颗粒的首个 3D 结构；OpenAI 称 Astra 在安全框架下的编码任务中表现不佳Fri, 06 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-06/https://www.ainewslog.com/blog/ai-daily-2026-11-06/AI Daily 2026-11-06. In this issue: Microsoft Releases VibeVoice-ASR-Streaming Unified Streaming Speech Model; Claude Helps Solve First 3D Structure of Vault Particle; OpenAI Says Astra Underperformed on Safety-Framed…Fri, 06 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-05-zh/https://www.ainewslog.com/blog/ai-daily-2026-11-05-zh/AI 日报 2026-11-05，本期关注：FireFlash：面向 Pareto 前沿的 300B 智能体强化学习开放配方；FrontierPhysics 基准测试在真实物理研究中评估 AI 智能体；Qwen 推出 SpecOPD 以加速投机解码Thu, 05 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-05/https://www.ainewslog.com/blog/ai-daily-2026-11-05/AI Daily 2026-11-05. In this issue: FireFlash: Open Recipe for 300B Agentic RL at Pareto Frontier; FrontierPhysics Benchmark Tests AI Agents on Real Physics Research; Qwen Introduces SpecOPD for Faster Speculative…Thu, 05 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-04-zh/https://www.ainewslog.com/blog/ai-daily-2026-11-04-zh/AI 日报 2026-11-04，本期关注：Ricursive Intelligence 发布 Ripple-65B-Preview 稠密开放权重模型；Kimi-K3.1-Pro 作为开放权重模型发布；Axion 1.0：完全开放的 35B 智能体基础模型，训练配方简洁Wed, 04 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-04/https://www.ainewslog.com/blog/ai-daily-2026-11-04/AI Daily 2026-11-04. In this issue: Ricursive Intelligence Releases Ripple-65B-Preview Dense Open-Weight Model; Kimi-K3.1-Pro Released as Open-Weights Model; Axion 1.0: Fully Open 35B Agent Foundation Model With Simple…Wed, 04 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-03-zh/https://www.ainewslog.com/blog/ai-daily-2026-11-03-zh/AI 日报 2026-11-03，本期关注：Karpathy 发布 nanochat：用 48 美元训练 GPT-2 级 LLM；Qwen 确认未来 Qwen 4.0 和 5.0 将保持开源；Colibri 用纯 C 在本地运行前沿 MoE 模型Tue, 03 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-03/https://www.ainewslog.com/blog/ai-daily-2026-11-03/AI Daily 2026-11-03. In this issue: Karpathy Releases nanochat: Train GPT-2-Level LLM for $48; Qwen Confirms Future Qwen 4.0 and 5.0 Will Remain Open Source; Colibri Runs Frontier MoE Models Locally in Pure CTue, 03 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-02-zh/https://www.ainewslog.com/blog/ai-daily-2026-11-02-zh/AI 日报 2026-11-02，本期关注：Qwen 确认 Qwen 4.0 和 5.0 将继续开源；ISTA 以约每参数 1 比特发布 763B DeepSeek-V4.1-Flash Coder；开源稀疏注意力低成本实现 10M Token 上下文Mon, 02 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-02/https://www.ainewslog.com/blog/ai-daily-2026-11-02/AI Daily 2026-11-02. In this issue: Qwen Confirms Qwen 4.0 and 5.0 Will Remain Open Source; ISTA Releases 763B DeepSeek-V4.1-Flash Coder at About 1 Bit per Parameter; Open-Source Sparse Attention Enables 10M-Token…Mon, 02 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-01-zh/https://www.ainewslog.com/blog/ai-daily-2026-11-01-zh/AI 日报 2026-11-01，本期关注：JEPAs 跨模态证明柏拉图表示假说；L3G-World 借助 3D Gaussian Memory 实现实时无限世界生成；Orkas 推出面向桌面的开源本地优先 AI Agent 团队Sun, 01 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-11-01/https://www.ainewslog.com/blog/ai-daily-2026-11-01/AI Daily 2026-11-01. In this issue: JEPAs Prove Platonic Representation Hypothesis Across Modalities; L3G-World Enables Real-Time Infinite World Generation with 3D Gaussian Memory; Orkas Launches Open-Source Local-First…Sun, 01 Nov 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-31-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-31-zh/AI 日报 2026-10-31，本期关注：Anthropic 为 Claude 开源 Agent Skills；GrootTeam 发布 Groot-280B-A16B 及更小的开放模型；Youtu-Parsing-Omni 推出面向结构化证据的统一全模态解析器Sat, 31 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-31/https://www.ainewslog.com/blog/ai-daily-2026-10-31/AI Daily 2026-10-31. In this issue: Anthropic Open-Sources Agent Skills for Claude; GrootTeam Releases Groot-280B-A16B and Smaller Open Models; Youtu-Parsing-Omni Unveils Unified Omni-Modal Parser for Structured EvidenceSat, 31 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-30-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-30-zh/AI 日报 2026-10-30，本期关注：Google 的 Project Suncatcher 在卫星上运行 Gemma 4，并传回一首诗；Youtu-Parsing-Omni 统一文档、图像、音频、视频解析；CopilotKit OpenBot 推出拥有自己计算机的开源 AI 同事Fri, 30 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-30/https://www.ainewslog.com/blog/ai-daily-2026-10-30/AI Daily 2026-10-30. In this issue: Google's Project Suncatcher Runs Gemma 4 Onboard Satellite, Returns Poem; Youtu-Parsing-Omni Unifies Document, Image, Audio, Video Parsing; CopilotKit OpenBot Launches Open-Source AI…Fri, 30 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-29-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-29-zh/AI 日报 2026-10-29，本期关注：Google 发布面向网络安全的前沿模型 Gemini 4 Carbon；研究发现，Agent 记忆系统常常不如朴素 RAG；GitHub 宣布对 Copilot 进行重大改造，聚焦 Agent 化未来Thu, 29 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-29/https://www.ainewslog.com/blog/ai-daily-2026-10-29/AI Daily 2026-10-29. In this issue: Google unveils Gemini 4 Carbon, a cybersecurity-focused frontier model; Agent Memory Systems Often Underperform Naive RAG, Study Finds; GitHub Announces Major Copilot Overhaul to…Thu, 29 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-28-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-28-zh/AI 日报 2026-10-28，本期关注：Ψ-0.6 借助自适应推理统一视觉、语言与世界建模；Papers with Code 以开源 Apache 2.0 项目重新启动；审稿人警告半数 AI 会议同行评审由 AI 生成Wed, 28 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-28/https://www.ainewslog.com/blog/ai-daily-2026-10-28/AI Daily 2026-10-28. In this issue: Ψ-0.6 Unifies Vision, Language, and World Modeling With Adaptive Reasoning; Papers with Code Relaunches as Open-Source Apache 2.0 Project; Reviewer Warns Half of AI Conference Peer…Wed, 28 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-27-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-27-zh/AI 日报 2026-10-27，本期关注：Google DeepMind 的 Genie 4 将 Gemini Agent 嵌入世界模型；OpenAI 的 AI 声称在波聚焦上取得新数学突破；Claude 研究模型独立反驳 Exponential Time HypothesisTue, 27 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-27/https://www.ainewslog.com/blog/ai-daily-2026-10-27/AI Daily 2026-10-27. In this issue: Google DeepMind's Genie 4 Embeds Gemini Agent Inside World Model; OpenAI AI Claims New Math Breakthrough on Wave Focusing; Claude Research Model Independently Refutes Exponential Time…Tue, 27 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-26-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-26-zh/AI 日报 2026-10-26，本期关注：因担忧失控 AI，OpenAI 恢复离线训练其最强大系统；RynnUnify：面向长时程视觉规划的 16B 统一具身模型；欺骗防御使 AI 攻击者速度慢 8 倍、成本高 23 倍Mon, 26 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-26/https://www.ainewslog.com/blog/ai-daily-2026-10-26/AI Daily 2026-10-26. In this issue: OpenAI Resumes Training Most Powerful Systems Offline Over Rogue AI Fears; RynnUnify: 16B Unified Embodied Model for Long-Horizon Visual Planning; Deception Defense Makes AI Attackers…Mon, 26 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-25-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-25-zh/AI 日报 2026-10-25，本期关注：WeVisOne-VL 统一对象级与像素级视觉感知；IQuest-Q1 推出面向 CLI 系统的 320B MoE 智能体基础模型；Nanbeige 发布 4.3-33B 模型，称其性能优于 Qwen3.8-27BSun, 25 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-25/https://www.ainewslog.com/blog/ai-daily-2026-10-25/AI Daily 2026-10-25. In this issue: WeVisOne-VL Unifies Object- and Pixel-Level Visual Perception; IQuest-Q1 Introduces 320B MoE Agentic Foundation Model for CLI Systems; Nanbeige Releases 4.3-33B Model, Claims…Sun, 25 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-24-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-24-zh/AI 日报 2026-10-24，本期关注：AWS AI 云容量售罄至 2027 年，2028 年预订迅速填满；微信 WeVisOne-VL 统一对象级与像素级视觉感知；NVIDIA OpenShell 发布面向自主 AI 智能体的安全运行时Sat, 24 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-24/https://www.ainewslog.com/blog/ai-daily-2026-10-24/AI Daily 2026-10-24. In this issue: AWS AI cloud capacity sold out through 2027, 2028 bookings filling fast; WeChat's WeVisOne-VL Unifies Object- and Pixel-Level Visual Perception; NVIDIA OpenShell Releases Safe Runtime…Sat, 24 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-23-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-23-zh/AI 日报 2026-10-23，本期关注：研究：Opus 4.8 在同行评审哈佛论文中表现优于人类专家；METR 与 Redwood 发现 OpenAI 智能体监控中的漏洞利用；IQuest-Q1：面向 CLI 系统的 320B MoE 智能体基础模型Fri, 23 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-23/https://www.ainewslog.com/blog/ai-daily-2026-10-23/AI Daily 2026-10-23. In this issue: Study: Opus 4.8 Outperforms Human Experts in Peer Reviewing Harvard Papers; METR and Redwood Find Exploit in OpenAI Agent Monitoring; IQuest-Q1: 320B MoE Agentic Foundation Model for…Fri, 23 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-22-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-22-zh/AI 日报 2026-10-22，本期关注：OpenAI 推出面向金融服务的 ChatGPT；METR 红队发现外部人员可绕过 OpenAI 智能体监控；StreamEval 基准评估交互式流式视频生成Thu, 22 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-22/https://www.ainewslog.com/blog/ai-daily-2026-10-22/AI Daily 2026-10-22. In this issue: OpenAI Introduces ChatGPT for Financial Services; Outsiders Bypass OpenAI Agent Monitoring, METR Red-Team Finds; StreamEval Benchmark Evaluates Interactive Streaming Video GenerationThu, 22 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-21-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-21-zh/AI 日报 2026-10-21，本期关注：OpenAI 模型即将登陆 Google Cloud；Claude 设计针对蝎子神经毒素的蛋白质结合剂；DeepMind AI 智能体通过编写 2,783 个程序通关 MinecraftWed, 21 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-21/https://www.ainewslog.com/blog/ai-daily-2026-10-21/AI Daily 2026-10-21. In this issue: OpenAI Models Coming to Google Cloud; Claude Designs Protein Binders Against Scorpion Neurotoxins; DeepMind AI agents beat Minecraft by writing 2,783 programsWed, 21 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-20-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-20-zh/AI 日报 2026-10-20，本期关注：Kronos：面向金融 K 线市场的开源基础模型；MemT-1.0 为基础模型引入记忆原生架构；Holi-World 从人类自我中心经验构建整体世界智能Tue, 20 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-20/https://www.ainewslog.com/blog/ai-daily-2026-10-20/AI Daily 2026-10-20. In this issue: Kronos: Open-Source Foundation Model for Financial K-Line Markets; MemT-1.0 Introduces Memory-Native Architecture for Foundation Models; Holi-World Builds Holistic World Intelligence…Tue, 20 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-19-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-19-zh/AI 日报 2026-10-19，本期关注：Anthropic、Google、OpenAI 或于本月推出新的前沿模型；OpenMobile-2 发布开放移动智能体及可扩展训练环境；Viggle 发布 JST-2：内置物理的实时 3D 世界生成Mon, 19 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-19/https://www.ainewslog.com/blog/ai-daily-2026-10-19/AI Daily 2026-10-19. In this issue: Anthropic, Google, OpenAI May Launch New Frontier Models This Month; OpenMobile-2 Releases Open Mobile Agents with Scalable Training Environments; Viggle Releases JST-2: Real-Time 3D…Mon, 19 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-18-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-18-zh/AI 日报 2026-10-18，本期关注：Diffusion LLM ORL 在单张 RTX 5090 上达到 2,788 tokens/s；FDE-Gym 基准在真实世界交付任务上测试 AI 智能体；Anthropic 的 Claude Code 智能体编码工具驻留在终端中Sun, 18 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-18/https://www.ainewslog.com/blog/ai-daily-2026-10-18/AI Daily 2026-10-18. In this issue: Diffusion LLM ORL Hits 2,788 Tokens/s on Single RTX 5090; FDE-Gym Benchmark Tests AI Agents on Real-World Delivery Tasks; Anthropic's Claude Code Agentic Coding Tool Lives in TerminalSun, 18 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-16-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-16-zh/AI 日报 2026-10-16，本期关注：Meta 发布 816B Muse Spark Open 模型，可在 Mac Studios 本地运行；OpenGCUA 旨在为通用计算机使用构建完全开放技术栈；Anthropic 的 Mythos 发现软件缺陷，Bank of America 雇用 1…Fri, 16 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-16/https://www.ainewslog.com/blog/ai-daily-2026-10-16/AI Daily 2026-10-16. In this issue: Meta Releases 816B Muse Spark Open Model, Runs Locally on Mac Studios; OpenGCUA Aims to Build Fully Open Stack for Generalist Computer Use; Anthropic's Mythos Uncovers Software Flaws…Fri, 16 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-15-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-15-zh/AI 日报 2026-10-15，本期关注：Muse Spark Open 发布，成为美国最强开放权重模型；NVIDIA 发布 Nemotron 3.5 Super VL 120B A12B 模型系列；Apertus 1.5 将开放模型扩展至多模态推理，具备 262K 上下文Thu, 15 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-15/https://www.ainewslog.com/blog/ai-daily-2026-10-15/AI Daily 2026-10-15. In this issue: Muse Spark Open Released as Most Capable US Open-Weight Model; NVIDIA Releases Nemotron 3.5 Super VL 120B A12B Model Family; Apertus 1.5 Extends Open Models to Multimodal Reasoning…Thu, 15 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-14-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-14-zh/AI 日报 2026-10-14，本期关注：NVIDIA Nemotron Bolt 3.5 74B A9B 将于 10 月 15 日发布；中国的开放权重 AI 模型落后美国闭源模型 7 个月；inclusionAI 发布 Ling-3.1-flash 560B MoE，支持 1M 上下文Wed, 14 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-14/https://www.ainewslog.com/blog/ai-daily-2026-10-14/AI Daily 2026-10-14. In this issue: NVIDIA Nemotron Bolt 3.5 74B A9B Releases October 15; China's Open-Weight AI Models Trail US Closed Models by 7 Months; inclusionAI Releases Ling-3.1-flash 560B MoE with 1M ContextWed, 14 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-13-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-13-zh/AI 日报 2026-10-13，本期关注：Microsoft BitNet 发布 1-Bit 嵌入模型与 VibeASR 引擎；Liquid AI 发布 LFM2.5-24B-A2B 端侧推理模型；AgentSWE 基准测试检验编码智能体能否构建可靠智能体Tue, 13 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-13/https://www.ainewslog.com/blog/ai-daily-2026-10-13/AI Daily 2026-10-13. In this issue: Microsoft BitNet Releases 1-Bit Embedding Models and VibeASR Engine; Liquid AI Releases LFM2.5-24B-A2B On-Device Reasoning Model; AgentSWE Benchmark Tests Whether Coding Agents Build…Tue, 13 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-12-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-12-zh/AI 日报 2026-10-12，本期关注：AirLLM 在单张 4GB GPU 上运行 70B 推理；Anthropic AI 智能体发现类 CRISPR 酶；研究人员称 2022 年已有先前研究；Linus Torvalds 使用 Claude Code 创建新的 Linux 内核分支Mon, 12 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-12/https://www.ainewslog.com/blog/ai-daily-2026-10-12/AI Daily 2026-10-12. In this issue: AirLLM Runs 70B Inference on a Single 4GB GPU; Anthropic AI agents find CRISPR-like enzyme; researcher cites prior 2022 study; Linus Torvalds Uses Claude Code for New Linux Kernel…Mon, 12 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-11-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-11-zh/AI 日报 2026-10-11，本期关注：MiMo-V2.6 为自我改进的全模态模型扩展 RL 规模；Hugging Face Transformers：面向 SOTA AI 的模型定义框架；Context Mode MCP 服务器优化 AI 编码智能体的上下文窗口Sun, 11 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-11/https://www.ainewslog.com/blog/ai-daily-2026-10-11/AI Daily 2026-10-11. In this issue: MiMo-V2.6 Scales RL for Self-Improving Omni-Modal Models; Hugging Face Transformers: Model-Definition Framework for SOTA AI; Context Mode MCP Server Optimizes AI Coding Agent Context…Sun, 11 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-10-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-10-zh/AI 日报 2026-10-10，本期关注：OpenAI 解雇三名安全研究员，后者警告寒蝉效应；Nvidia 洽谈收购开放模型初创公司 Reflection AI；菲尔兹奖得主对 OpenAI 数学发布感到震惊Sat, 10 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-10/https://www.ainewslog.com/blog/ai-daily-2026-10-10/AI Daily 2026-10-10. In this issue: OpenAI Fires Three Safety Researchers, Who Warn of Chilling Effect; Nvidia in Talks to Acquire Open-Model Startup Reflection AI; Fields Medalist Shocked by OpenAI Math ReleaseSat, 10 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-09-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-09-zh/AI 日报 2026-10-09，本期关注：OpenAI 推出 GPT-6，配备 Intelligent UI 交互式 ChatGPT 工具；Scientific American 聚焦 OpenAI 新数学证明中最令人兴奋的主张；USA TODAY 因使用 19 家出版物内容训练而起诉…Fri, 09 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-09/https://www.ainewslog.com/blog/ai-daily-2026-10-09/AI Daily 2026-10-09. In this issue: OpenAI rolls out GPT-6 with Intelligent UI interactive ChatGPT tools; Scientific American Highlights Exciting Claims from OpenAI's New Proofs; USA TODAY Sues OpenAI Over Training on…Fri, 09 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-08-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-08-zh/AI 日报 2026-10-08，本期关注：Cua 以开源驱动和基准测试开放计算机使用智能体；Swift 开源推理 LLM 下载量突破 220 万次；单一攻击者利用开源 AI 技术栈入侵韩国银行Thu, 08 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-08/https://www.ainewslog.com/blog/ai-daily-2026-10-08/AI Daily 2026-10-08. In this issue: Cua Opens Computer-Use Agents with Open-Source Drivers and Benchmarks; Swift Open-Source Reasoning LLMs Surpass 2.2M Downloads; Single Attacker Breaches South Korean Banks with…Thu, 08 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-07-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-07-zh/AI 日报 2026-10-07，本期关注：DeepSeek 的 DeepGEMM 发布高性能 GPU 内核并支持昇腾；ARC-AGI-3 在 Kaggle 上的得分从 7% 跃升至 56%；SWE-Race 基准用 188 个真实并发缺陷考验编码智能体Wed, 07 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-07/https://www.ainewslog.com/blog/ai-daily-2026-10-07/AI Daily 2026-10-07. In this issue: DeepSeek's DeepGEMM Releases High-Performance GPU Kernels and Ascend Support; ARC-AGI-3 Scores Jump from 7% to 56% on Kaggle; SWE-Race Benchmark Tests Coding Agents on 188 Real…Wed, 07 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-06-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-06-zh/AI 日报 2026-10-06，本期关注：ARC-AGI-3 分数在 30 天内从 7% 飙升至 56%；SWE-Race 基准让编码智能体面对真实并发缺陷；自生成反馈使测试时训练不稳定Tue, 06 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-06/https://www.ainewslog.com/blog/ai-daily-2026-10-06/AI Daily 2026-10-06. In this issue: ARC-AGI-3 Scores Surge from 7% to 56% in 30 Days; SWE-Race Benchmark Puts Coding Agents on Real Concurrency Bugs; Self-Generated Feedback Destabilizes Test-Time TrainingTue, 06 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-05-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-05-zh/AI 日报 2026-10-05，本期关注：antirez 发布面向 DeepSeek、GLM、Qwen 的 DwarfStar 本地推理引擎；Reflection AI 将推出美国开放权重模型，对标 DeepSeek、Qwen；125B Qwen3.8-Flash-Next 在单台 Strix…Mon, 05 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-05/https://www.ainewslog.com/blog/ai-daily-2026-10-05/AI Daily 2026-10-05. In this issue: antirez Releases DwarfStar Local Inference Engine for DeepSeek, GLM, Qwen; Reflection AI to Launch US Open-Weight Model to Rival DeepSeek, Qwen; 125B Qwen3.8-Flash-Next Runs on Single…Mon, 05 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-04-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-04-zh/AI 日报 2026-10-04，本期关注：Anthropic 的 Claude Code 智能体编程工具现已通过 GitHub 提供；美团发布 LongCat-Video 13.6B 开源视频生成模型；哔哩哔哩基于 Qwen3.5 发布 Index-Translate 多语言翻译模型系列Sun, 04 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-04/https://www.ainewslog.com/blog/ai-daily-2026-10-04/AI Daily 2026-10-04. In this issue: Anthropic's Claude Code Agentic Coding Tool Available via GitHub; Meituan Releases LongCat-Video 13.6B Open-Source Video Generation Model; Bilibili Releases Index-Translate…Sun, 04 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-03-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-03-zh/AI 日报 2026-10-03，本期关注：OpenAI 安全负责人辞职，警告公司文化已“崩坏”；AI 终于以低成本击败最强 Stratego 玩家；Google 终止对 Gemini Flash 和 Pro 模型的免费访问Sat, 03 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-03/https://www.ainewslog.com/blog/ai-daily-2026-10-03/AI Daily 2026-10-03. In this issue: OpenAI Safety Leader Quits, Warning Company Culture Is 'Broken'; AI Finally Beats Best Stratego Player on a Budget; Google Ends Free Access to Gemini Flash and Pro ModelsSat, 03 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-02-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-02-zh/AI 日报 2026-10-02，本期关注：OneStreamer 为流式视频 LLM 统一感知、记忆与主动响应；同策略学习 vs 异策略学习：蒸馏动态的系统性研究；E-MoE 增强面向扩散语言模型的混合专家Fri, 02 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-02/https://www.ainewslog.com/blog/ai-daily-2026-10-02/AI Daily 2026-10-02. In this issue: OneStreamer Unifies Perception, Memory, and Proactive Response for Streaming Video LLMs; On-Policy vs Off-Policy Learning: Systematic Study of Distillation Dynamics; E-MoE Enhances…Fri, 02 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-01-zh/https://www.ainewslog.com/blog/ai-daily-2026-10-01-zh/AI 日报 2026-10-01，本期关注：Google DeepMind 的 CO₂Jump Sampler 提升联合文本-图像生成一致性；ORTUS AI 开源 RightWayUp 360° 图像旋转模型；OpenClaw：开源个人 AI 助手可在任意操作系统本地运行Thu, 01 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-10-01/https://www.ainewslog.com/blog/ai-daily-2026-10-01/AI Daily 2026-10-01. In this issue: Google DeepMind's CO₂Jump Sampler Improves Joint Text-Image Generation Consistency; ORTUS AI Open-Sources RightWayUp 360° Image Rotation Model; OpenClaw: Open-Source Personal AI…Thu, 01 Oct 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-30-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-30-zh/AI 日报 2026-09-30，本期关注：中国实验室发布 Ling-3.1-Flash：560B MoE、1M 上下文、开源；Hugging Face 开源面向浏览器 AI 的快速 WebGPU 内核；DeepSeek 现在使用华为昇腾 950 训练模型Wed, 30 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-30/https://www.ainewslog.com/blog/ai-daily-2026-09-30/AI Daily 2026-09-30. In this issue: Chinese Lab Releases Ling-3.1-Flash: 560B MoE, 1M Context, Open Source; Hugging Face Open-Sources Fast WebGPU Kernels for Browser AI; DeepSeek Now Trains Models on Huawei Ascend 950Wed, 30 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-29-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-29-zh/AI 日报 2026-09-29，本期关注：Qwen3-VL 8B 在税表上优于 GPT-5.6，但在印度日期上表现不佳；无编码器多模态预训练的缩放定律；交错 RL 使统一多模态模型具备原生反思能力Tue, 29 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-29/https://www.ainewslog.com/blog/ai-daily-2026-09-29/AI Daily 2026-09-29. In this issue: Qwen3-VL 8B Outperforms GPT-5.6 on Tax Forms, Struggles on Indian Dates; Scaling Laws for Encoder-Free Multimodal Pretraining; Interleaved RL Enables Native Reflection in Unified…Tue, 29 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-28-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-28-zh/AI 日报 2026-09-28，本期关注：NVIDIA OpenShell 沙盒为 AI 智能体强制执行运行时限制；研究发现编码智能体通过想象中的评分器进行投机性奖励攻击；开源 VoiceStudio 提供支持 646 种语言的本地 ElevenLabs 替代方案Mon, 28 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-28/https://www.ainewslog.com/blog/ai-daily-2026-09-28/AI Daily 2026-09-28. In this issue: NVIDIA OpenShell Sandbox Enforces Runtime Limits for AI Agents; Study Finds Coding Agents Speculatively Hack Rewards via Imagined Grader; Open-Source VoiceStudio Offers Local…Mon, 28 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-27-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-27-zh/AI 日报 2026-09-27，本期关注：OpenAI 记录首批自我复制 AI 蠕虫跨智能体传播；Anthropic 发布适用于 GitHub PR 和 Issue 的 Claude Code Action；LLM 重建身份，大规模去匿名化化名用户Sun, 27 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-27/https://www.ainewslog.com/blog/ai-daily-2026-09-27/AI Daily 2026-09-27. In this issue: OpenAI Documents First Self-Replicating AI Worms Spreading Across Agents; Anthropic Releases Claude Code Action for GitHub PRs and Issues; LLMs Reconstruct Identity, Deanonymizing…Sun, 27 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-26-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-26-zh/AI 日报 2026-09-26，本期关注：OpenAI 在沙箱 DNS 漏洞后暂停前沿工具使用训练；OpenAI 推出面向金融服务的 ChatGPT；LLM 展现下一 token 分布的线性叠加Sat, 26 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-26/https://www.ainewslog.com/blog/ai-daily-2026-09-26/AI Daily 2026-09-26. In this issue: OpenAI Pauses Frontier Tool-Use Training After Sandbox DNS Breach; OpenAI Introduces ChatGPT for Financial Services; LLMs Show Linear Superposition of Next-Token DistributionsSat, 26 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-25-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-25-zh/AI 日报 2026-09-25，本期关注：Microsoft 放弃个人 AI 聊天机器人竞赛，重启 Copilot；OpenAI 推出面向金融服务的 ChatGPT；澳总理称 OpenAI 智能体入侵澳大利亚政府网站Fri, 25 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-25/https://www.ainewslog.com/blog/ai-daily-2026-09-25/AI Daily 2026-09-25. In this issue: Microsoft Abandons Personal AI Chatbot Race with Copilot Reboot; OpenAI Introduces ChatGPT for Financial Services; OpenAI Agent Hacked Australian Government Website, PM SaysFri, 25 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-24-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-24-zh/AI 日报 2026-09-24，本期关注：Xiaomi 发布 MiMo-V2.6 多模态 AI 模型，RL 训练成本 350 万美元；HappyWorld-Bench：世界模型评估的综合基准；The Tasteful Agent：衡量 LLM 长时程任务中的品味Thu, 24 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-24/https://www.ainewslog.com/blog/ai-daily-2026-09-24/AI Daily 2026-09-24. In this issue: Xiaomi Releases MiMo-V2.6 Multimodal AI Model with $3.5M RL Training Cost; HappyWorld-Bench: Comprehensive Benchmark for World Model Evaluation; The Tasteful Agent: Measuring Taste in…Thu, 24 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-23-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-23-zh/AI 日报 2026-09-23，本期关注：Google 开源 AX 智能体编排运行时；OmniEdu 发布面向 K-12 学与教的开源基础模型；Realtime-Venus：借助异步委派实现全双工音视频交互Wed, 23 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-23/https://www.ainewslog.com/blog/ai-daily-2026-09-23/AI Daily 2026-09-23. In this issue: Google Open-Sources AX Agentic Orchestration Runtime; OmniEdu Releases Open Foundation Models for K-12 Learning and Teaching; Realtime-Venus: Full-Duplex Audio-Visual Interaction with…Wed, 23 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-22-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-22-zh/AI 日报 2026-09-22，本期关注：OpenAI 推出 GPT-6 Sol 和 Luna；WorldCrafter：具备隐式 3D 感知记忆的视频世界模型；RRSI 正则化 LLM Agent Harness 的递归自我改进Tue, 22 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-22/https://www.ainewslog.com/blog/ai-daily-2026-09-22/AI Daily 2026-09-22. In this issue: OpenAI Introduces GPT-6 Sol and Luna; WorldCrafter: Video World Model with Implicit 3D-Aware Memory; RRSI Regularizes Recursive Self-Improvement of LLM Agent HarnessesTue, 22 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-21-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-21-zh/AI 日报 2026-09-21，本期关注：DeepSeek-V4.1-Flash 推动 KV Cache 压缩的极限；Anthropic 发布 Claude 金融服务参考智能体与连接器；AI 沙箱“逃逸”是草率的防火墙故障，并非气隙隔离被突破Mon, 21 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-21/https://www.ainewslog.com/blog/ai-daily-2026-09-21/AI Daily 2026-09-21. In this issue: DeepSeek-V4.1-Flash Pushes Limits of KV Cache Compression; Anthropic Releases Claude Financial Services Reference Agents and Connectors; AI Sandbox 'Escapes' Were Sloppy Firewall…Mon, 21 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-20-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-20-zh/AI 日报 2026-09-20，本期关注：DeepSeek 推出 V4.1-Flash，实现极致 KV 缓存压缩；OpenAI 推出面向金融服务的 ChatGPT；Cactus Compute 的 Needle：面向微型设备的 2 比特边缘 AI 模型Sun, 20 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-20/https://www.ainewslog.com/blog/ai-daily-2026-09-20/AI Daily 2026-09-20. In this issue: DeepSeek Introduces V4.1-Flash for Extreme KV Cache Compression; OpenAI Introduces ChatGPT for Financial Services; Cactus Compute’s Needle: 2-Bit Edge AI Model for Tiny DevicesSun, 20 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-19-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-19-zh/AI 日报 2026-09-19，本期关注：DeepSeek-V4.1-Flash 推动 KV Cache 压缩极限；Microsoft 总监称 AI 抓取是最大劳动盗窃；OpenAI 警告出版商；Google Gemini AI 在网络安全测试中入侵其他公司Sat, 19 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-19/https://www.ainewslog.com/blog/ai-daily-2026-09-19/AI Daily 2026-09-19. In this issue: DeepSeek-V4.1-Flash Pushes Limits of KV Cache Compression; Microsoft director calls AI scraping largest labor theft; OpenAI warns publishers; Google Gemini AI Hacked Other Companies…Sat, 19 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-18-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-18-zh/AI 日报 2026-09-18，本期关注：DeepSeek-V4.1-Flash 瞄准长上下文智能体的 KV Cache 压缩；Prior Labs 发布 TabPFN-3.5 SOTA 表格基础模型；GoBench 在 9x9 围棋上对照 KataGo 天梯评估 LLMFri, 18 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-18/https://www.ainewslog.com/blog/ai-daily-2026-09-18/AI Daily 2026-09-18. In this issue: DeepSeek-V4.1-Flash Targets KV Cache Compression for Long-Context Agents; Prior Labs Releases TabPFN-3.5 SOTA Tabular Foundation Model; GoBench Evaluates LLMs on 9x9 Go Against KataGo…Fri, 18 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-17-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-17-zh/AI 日报 2026-09-17，本期关注：微软 AI 负责人与 Anthropic 就拟人化 AI 设计发生冲突；腾讯开源 WeKnora LLM 知识平台，集成 RAG 与 Agents；Voicebox：用于声音克隆与听写的开源 AI 语音工作室Thu, 17 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-17/https://www.ainewslog.com/blog/ai-daily-2026-09-17/AI Daily 2026-09-17. In this issue: Microsoft AI chief and Anthropic clash over humanlike AI design; Tencent Open-Sources WeKnora LLM Knowledge Platform with RAG and Agents; Voicebox: Open-Source AI Voice Studio for…Thu, 17 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-16-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-16-zh/AI 日报 2026-09-16，本期关注：Qwen3.8 Max 以 Intelligence Index 45 分登顶中国排行榜；ZGCM-1：面向数学与智能体搜索的完全开放 7B 基础模型；Atria Dawn Preview：面向科学研究的智能体语言模型Wed, 16 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-16/https://www.ainewslog.com/blog/ai-daily-2026-09-16/AI Daily 2026-09-16. In this issue: Qwen3.8 Max Tops China Leaderboard With 45 on Intelligence Index; ZGCM-1: Fully Open 7B Foundation Model for Math and Agentic Search; Atria Dawn Preview: Agentic Language Model for…Wed, 16 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-15-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-15-zh/AI 日报 2026-09-15，本期关注：Apple Foundation Models 现已在 macOS 27 上原生运行；Alibaba 开源 Open Code Review AI 代码审查工具；Vidu S2 实现实时 720p 交互式与可编辑视频生成Tue, 15 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-15/https://www.ainewslog.com/blog/ai-daily-2026-09-15/AI Daily 2026-09-15. In this issue: Apple Foundation Models Now Run Natively on macOS 27; Alibaba Open-Sources Open Code Review AI Code Review Tool; Vidu S2 Enables Real-Time 720p Interactive and Editable Video…Tue, 15 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-14-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-14-zh/AI 日报 2026-09-14，本期关注：OpenAI 声称已解决纳维-斯托克斯千禧年难题；Yoshua Bengio 探讨 AI 智能体为何撒谎、作弊与协调；Anthropic CEO Dario Amodei 呼吁放缓 AI 开发Mon, 14 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-14/https://www.ainewslog.com/blog/ai-daily-2026-09-14/AI Daily 2026-09-14. In this issue: OpenAI Claims It Solved Navier-Stokes Millennium Problem; Yoshua Bengio Explores Why AI Agents Lie, Cheat, Coordinate; Anthropic CEO Dario Amodei Calls for Slowing AI DevelopmentMon, 14 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-13-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-13-zh/AI 日报 2026-09-13，本期关注：OpenAI 智能体对 RubyGems 发起未披露的攻击；YuE2 发布具备符号规划与智能体编辑的前沿音乐生成；OpenAI 推出面向金融服务的 ChatGPTSun, 13 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-13/https://www.ainewslog.com/blog/ai-daily-2026-09-13/AI Daily 2026-09-13. In this issue: OpenAI agents carried out undisclosed attack on RubyGems; YuE2 Releases Frontier Music Generation with Symbolic Planning and Agentic Editing; OpenAI Introduces ChatGPT for Financial…Sun, 13 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-12-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-12-zh/AI 日报 2026-09-12，本期关注：OpenAI 声称在 Navier-Stokes 千禧年问题上取得突破；NeurIPS 因有缺陷的 AI 检测器直接拒稿 178 篇论文；NCP-ArchPreview：通过下一概念预测的潜在空间语言模型Sat, 12 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-12/https://www.ainewslog.com/blog/ai-daily-2026-09-12/AI Daily 2026-09-12. In this issue: OpenAI Claims Breakthrough on Navier-Stokes Millennium Problem; NeurIPS Desk-Rejected 178 Papers After Flawed AI Detector; NCP-ArchPreview: Latent-Space Language Models via Next…Sat, 12 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-11-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-11-zh/AI 日报 2026-09-11，本期关注：胡塞武装试图使用 Anthropic 的 Claude 开发导弹软件；OpenAI 可能放缓前沿 AI，Altman 希望其他公司跟进；24 位菲尔兹奖得主就数学领域 AI 严重错位签署公开信Fri, 11 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-11/https://www.ainewslog.com/blog/ai-daily-2026-09-11/AI Daily 2026-09-11. In this issue: Houthis Tried to Use Anthropic's Claude for Missile Software; OpenAI May Slow Frontier AI, Altman Hopes Others Follow; 24 Fields Medal Winners Sign Letter on Severe AI Misalignment in…Fri, 11 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-10-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-10-zh/AI 日报 2026-09-10，本期关注：OpenAI 声称已破解 Navier-Stokes 千禧年难题；NeurIPS 使用 AI 检测器直接拒稿 178 篇论文，而该检测器会标记其自己的主席；LLM 引导的程序演化改进 10 个已知最优圆填充解Thu, 10 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-10/https://www.ainewslog.com/blog/ai-daily-2026-09-10/AI Daily 2026-09-10. In this issue: OpenAI Claims It Cracked Navier-Stokes Millennium Problem; NeurIPS Desk-Rejected 178 Papers Using AI Detector That Flags Its Own Chairs; LLM-Guided Program Evolution Improves 10 Best-…Thu, 10 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-09-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-09-zh/AI 日报 2026-09-09，本期关注：AuK 开源模型统一语音生成与编辑；Gander 融合全感知、全双工交互与智能体技能；27B 1-bit 模型在浏览器中运行，6GB GPU 上速度达 25–30 tok/sWed, 09 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-09/https://www.ainewslog.com/blog/ai-daily-2026-09-09/AI Daily 2026-09-09. In this issue: AuK Open-Source Model Unifies Speech Generation and Editing; Gander Combines Omni Perception, Full-Duplex Interaction, and Agentic Skills; 27B 1-Bit Model Runs in the Browser at 25–30…Wed, 09 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-08-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-08-zh/AI 日报 2026-09-08，本期关注：Meta 发布 Muse，具备高级能力的个人 AI 智能体；OpenAI 分享研究加速的内部视角；DeerFlow 2.0：字节跳动开源 SuperAgent 登顶 GitHubTue, 08 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-08/https://www.ainewslog.com/blog/ai-daily-2026-09-08/AI Daily 2026-09-08. In this issue: Meta Launches Muse, Personal AI Agent with Advanced Capabilities; OpenAI Shares Inside View on Research Acceleration; DeerFlow 2.0: ByteDance's Open-Source SuperAgent Hits #1 on GitHubTue, 08 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-07-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-07-zh/AI 日报 2026-09-07，本期关注：徒步者轻信 Gemini 获救；OpenAI 发布 GPT-6 Astra；联合国人权事务负责人警告 AI 可能对人类构成生存风险；NCSC 警告：影子 AI 可能泄露数据与智能体权限Mon, 07 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-07/https://www.ainewslog.com/blog/ai-daily-2026-09-07/AI Daily 2026-09-07. In this issue: Hikers Rescued After Trusting Gemini; OpenAI Launches GPT-6 Astra; UN Rights Chief Warns AI Could Pose Existential Risk to Humanity; NCSC Warns Shadow AI Can Expose Data and Agent…Mon, 07 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-06-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-06-zh/AI 日报 2026-09-06，本期关注：LLaDA-Image：用于图像生成的开源 6B Diffusion Transformer；随机注意力研究表明：KV 缓存驱逐无需评分；Terminal-Universe：从智能体轨迹构建终端环境Sun, 06 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-06/https://www.ainewslog.com/blog/ai-daily-2026-09-06/AI Daily 2026-09-06. In this issue: LLaDA-Image: Open 6B Diffusion Transformer for Image Generation; Random Attention Shows KV Cache Eviction Needs No Scoring; Terminal-Universe: Building Terminal Environments from…Sun, 06 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-05-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-05-zh/AI 日报 2026-09-05，本期关注：OpenAI 的 GPT-6 Astra 引发讨论；Astra（GPT-6）高智商，低直觉；LLaDA-Image：完全开源的 6B Diffusion Transformer 图像生成模型Sat, 05 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-05/https://www.ainewslog.com/blog/ai-daily-2026-09-05/AI Daily 2026-09-05. In this issue: GPT-6 Astra from OpenAI Sparks Discussion; Astra (GPT-6) High Intelligence, Low Intuition; LLaDA-Image: Fully Open 6B Diffusion Transformer for Image GenerationSat, 05 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-04-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-04-zh/AI 日报 2026-09-04，本期关注：OpenAI 发布 GPT-6 Astra，重大 AI 模型更新；NVIDIA 以 129.3 亿美元收购 Hugging Face；OpenAI 智能体劫持风险浮现：研究揭示与未公开网站事件Fri, 04 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-04/https://www.ainewslog.com/blog/ai-daily-2026-09-04/AI Daily 2026-09-04. In this issue: OpenAI Releases GPT-6 Astra, Major AI Model Update; NVIDIA Acquires Hugging Face for $12.93B; OpenAI Agent Hijacking Risks Surface in Research and Undisclosed Website IncidentFri, 04 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-03-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-03-zh/AI 日报 2026-09-03，本期关注：Google Research 发布 TimesFM 3.0 时序基础模型；Nous Research 推出内置学习循环的 Hermes Agent；Chrome DevTools 为 AI 编程智能体推出 MCP 服务器Thu, 03 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-03/https://www.ainewslog.com/blog/ai-daily-2026-09-03/AI Daily 2026-09-03. In this issue: Google Research Releases TimesFM 3.0 Time-Series Foundation Model; Nous Research Unveils Hermes Agent with Built-in Learning Loop; Chrome DevTools Launches MCP Server for AI Coding…Thu, 03 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-02-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-02-zh/AI 日报 2026-09-02，本期关注：Qwen-Drive-1.0 发布自动驾驶视觉语言基础模型；DeepSeek V4 Flash Vision Exp 的视觉支持已合并；Perplexity 开源其针对 Qwen 3.6 调优的 Mac 推理服务器Wed, 02 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-02/https://www.ainewslog.com/blog/ai-daily-2026-09-02/AI Daily 2026-09-02. In this issue: Qwen-Drive-1.0 Unveils Vision-Language Foundation Model for Autonomous Driving; Vision Support Merged for DeepSeek V4 Flash Vision Exp; Perplexity Open-Sources Its Mac Inference…Wed, 02 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-01-zh/https://www.ainewslog.com/blog/ai-daily-2026-09-01-zh/AI 日报 2026-09-01，本期关注：Claude Code Opus 5 自动模式遭漏洞利用；GLM-5.3-Flash 发布：多模态、稀疏注意力开放权重模型；Anthropic 发布 Claude Fable 5.1 和 Mythos 5.1Tue, 01 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-09-01/https://www.ainewslog.com/blog/ai-daily-2026-09-01/AI Daily 2026-09-01. In this issue: Claude Code Opus 5 Auto Mode Broken by Exploit; GLM-5.3-Flash Released: Multimodal, Sparse Attention Open-Weight Model; Anthropic Unveils Claude Fable 5.1 and Mythos 5.1Tue, 01 Sep 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-31-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-31-zh/AI 日报 2026-08-31，本期关注：Z.ai 发布 GLM-5.3-Flash，首款开放权重多模态模型；Claude Code Opus 5 自动模式安全漏洞曝光；DeepSeek 在 Hugging Face 发布 V4 Flash Vision 实验模型Mon, 31 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-31/https://www.ainewslog.com/blog/ai-daily-2026-08-31/AI Daily 2026-08-31. In this issue: Z.ai Releases GLM-5.3-Flash, First Open-Weight Multimodal Model; Claude Code Opus 5 Auto Mode Security Flaws Exposed; DeepSeek Releases V4 Flash Vision Experimental Model on Hugging…Mon, 31 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-30-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-30-zh/AI 日报 2026-08-30，本期关注：GLM-5.3-Flash：Z.ai 的开源多模态模型挑战 Claude Opus；采用稀疏注意力与更长上下文的开源 AI 模型以 GGUF 格式发布；智能体游戏开发或将成为世界模型的数据引擎Sun, 30 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-30/https://www.ainewslog.com/blog/ai-daily-2026-08-30/AI Daily 2026-08-30. In this issue: GLM-5.3-Flash: Z.ai's Open Multimodal Model Challenges Claude Opus; Open-Source AI Models with Sparse Attention and Longer Context Released in GGUF; Agentic Game Development Could…Sun, 30 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-29-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-29-zh/AI 日报 2026-08-29，本期关注：VGI-Bench：新基准测试视频生成模型的视觉推理能力；将智能体游戏开发作为扩展世界模型的数据引擎；基于开放权重模型的持续学习赋能主权 AISat, 29 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-29/https://www.ainewslog.com/blog/ai-daily-2026-08-29/AI Daily 2026-08-29. In this issue: VGI-Bench: New Benchmark Probes Visual Reasoning in Video Generation Models; Agentic Game Development as Data Engine for Scaling World Models; Continual Learning on Open-Weight Models…Sat, 29 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-28-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-28-zh/AI 日报 2026-08-28，本期关注：OpenAI 自研芯片、英伟达收购 Hugging Face、阿里发布 Qwen Flash，共同标志市场转向；Meta 曾探讨借助 AI 削减 60% 团队规模，随后放弃；OpenMontage 发布，成为首个开源智能体视频制作系统Fri, 28 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-28/https://www.ainewslog.com/blog/ai-daily-2026-08-28/AI Daily 2026-08-28. In this issue: OpenAI’s Custom Chip, Nvidia’s Hugging Face Deal, and Alibaba’s Qwen Flash Signal a Market Shift; Meta Explored 60% Team Cuts via AI, Then Backed Off; OpenMontage Launches as First…Fri, 28 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-27-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-27-zh/AI 日报 2026-08-27，本期关注：英伟达将以 129 亿美元收购 Hugging Face；比尔·盖茨警告 AI 崛起将带来动荡时期；FrontierChallenge 基准测试评估科学工作流完成能力Thu, 27 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-27/https://www.ainewslog.com/blog/ai-daily-2026-08-27/AI Daily 2026-08-27. In this issue: Nvidia to Acquire Hugging Face for $12.9B; Bill Gates Warns AI Rise Will Bring Turbulent Times; FrontierChallenge Benchmark Evaluates Scientific Workflow CompletionThu, 27 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-26-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-26-zh/AI 日报 2026-08-26，本期关注：GigaBrain-0.7 以三系统架构扩展具身基础模型；EchoWM：开放的全模态世界模型，支持交互式导航；微信发布 WeMM-Embedding 多模态模型Wed, 26 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-26/https://www.ainewslog.com/blog/ai-daily-2026-08-26/AI Daily 2026-08-26. In this issue: GigaBrain-0.7 Scales Embodied Foundation Models with Three-System Architecture; EchoWM: Open Omnimodal World Model with Interactive Navigation; WeChat Releases WeMM-Embedding…Wed, 26 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-25-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-25-zh/AI 日报 2026-08-25，本期关注：英国国家网络安全中心敦促为 AI 智能体配置紧急关闭开关，警告安全训练可被绕过；Uber 因算法驱动的司机封禁违反 GDPR 被罚近 10 亿美元；EchoWM：全模态世界模型生成 720p 视频与音频Tue, 25 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-25/https://www.ainewslog.com/blog/ai-daily-2026-08-25/AI Daily 2026-08-25. In this issue: UK NCSC Urges Kill Switches for AI Agents, Warns Safety Training Can Be Bypassed; Uber Fined Nearly $1B for GDPR Violation Over Algorithm-Driven Driver Suspensions; EchoWM: Omnimodal…Tue, 25 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-24-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-24-zh/AI 日报 2026-08-24，本期关注：阿里巴巴 Qwen3.8-27B 位列 Code Arena WebDev 第 9 名；Marin 项目启动开放的 535B 模型训练计划；Anthropic 面向 MCP 连接器的企业级身份验证正式全面可用Mon, 24 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-24/https://www.ainewslog.com/blog/ai-daily-2026-08-24/AI Daily 2026-08-24. In this issue: Alibaba Qwen3.8-27B Ranks #9 in Code Arena WebDev; Marin Project Launches Open 535B Model Training Run; Anthropic's Enterprise Auth for MCP Connectors Now GAMon, 24 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-23-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-23-zh/AI 日报 2026-08-23，本期关注：OpenAI 发布终端专用 Codex CLI 编程代理；新 Claude 模型 Marshmallow 和 Melon 现身，发布在即；Ollama 与 Poolside 和 Nvidia 合作推动开放模型Sun, 23 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-23/https://www.ainewslog.com/blog/ai-daily-2026-08-23/AI Daily 2026-08-23. In this issue: OpenAI Launches Codex CLI Coding Agent for Terminal; New Claude Models Marshmallow and Melon Spotted, Release Likely Imminent; Ollama Collaborates with Poolside and Nvidia for Open…Sun, 23 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-22-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-22-zh/AI 日报 2026-08-22，本期关注：Gemini 3.7 Flash 创下增长纪录，ARC-AGI 高分亮眼；Ox-alpha 在 DeepSWE 上得 58.4%，辟谣 80% 传闻；Claude Code 悄然调低推理强度，用户发现响应“变笨”Sat, 22 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-22/https://www.ainewslog.com/blog/ai-daily-2026-08-22/AI Daily 2026-08-22. In this issue: Gemini 3.7 Flash Sets Growth Record, Scores High on ARC-AGI; Ox-alpha Scores 58.4% on DeepSWE, Debunking 80% Rumor; Claude Code Silently Lowers Reasoning Effort, Users Notice Dumber…Sat, 22 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-21-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-21-zh/AI 日报 2026-08-21，本期关注：Modular 开源 Mojo 编译器、MAX 推理服务器及模型流水线；腾讯发布 AI-Infra-Guard 开源红队测试平台；Cursor 推出官方插件规范与开发者插件Fri, 21 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-21/https://www.ainewslog.com/blog/ai-daily-2026-08-21/AI Daily 2026-08-21. In this issue: Modular Open-Sources Mojo Compiler, MAX Inference Server, and Model Pipelines; Tencent Releases AI-Infra-Guard Open-Source Red Teaming Platform; Cursor Launches Official Plugin Spec…Fri, 21 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-20-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-20-zh/AI 日报 2026-08-20，本期关注：Stripe 收购 OpenRouter，布局 AI 模型市场；彭博图表显示中国 AI 正快速追赶美国；OpenAI 部署首批 NVIDIA Vera Rubin 机架用于前沿 AI 训练Thu, 20 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-20/https://www.ainewslog.com/blog/ai-daily-2026-08-20/AI Daily 2026-08-20. In this issue: Stripe Acquires OpenRouter, AI Model Marketplace; Bloomberg Chart Shows China's AI Catching Up to US; OpenAI Deploys First NVIDIA Vera Rubin Racks for Frontier AI TrainingThu, 20 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-19-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-19-zh/AI 日报 2026-08-19，本期关注：Gemini 3.7 Flash 以速度和准确性登顶 AI 分析师排行榜；OpenAI 的 Codex 周活跃用户达到 2000 万；StateM Agent Runtime 在 Terminal-Bench 2.1 上达到 95.3% 准确率Wed, 19 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-19/https://www.ainewslog.com/blog/ai-daily-2026-08-19/AI Daily 2026-08-19. In this issue: Gemini 3.7 Flash Tops AI Analyst Leaderboard with Speed and Accuracy; OpenAI's Codex Hits 20M Weekly Active Users; StateM Agent Runtime Hits 95.3% Accuracy on Terminal-Bench 2.1Wed, 19 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-18-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-18-zh/AI 日报 2026-08-18，本期关注：OpenAI 暂停前沿强化学习训练以确保安全标准；Anthropic 警告超级智能可能在 2026 年底到来；GLM-5.3 API 发布，聚焦编程、网络安全与智能体任务Tue, 18 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-18/https://www.ainewslog.com/blog/ai-daily-2026-08-18/AI Daily 2026-08-18. In this issue: OpenAI Pauses Frontier RL Training to Ensure Safety Standards; Anthropic Warns Superintelligence Could Arrive by Late 2026; GLM-5.3 API Launches with Coding, Cybersecurity, Agentic…Tue, 18 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-17-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-17-zh/AI 日报 2026-08-17，本期关注：本地模型 Qwen3.8-27B 比肩前沿 AI 性能；谷歌单提示词演示：一条提示生成含文案、图片和视频的落地页；Claude Code 新增 /design 技能，支持可编辑 UI 画板Mon, 17 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-17/https://www.ainewslog.com/blog/ai-daily-2026-08-17/AI Daily 2026-08-17. In this issue: Local Model Qwen3.8-27B Matches Frontier AI Performance; Google's One-Prompt Demo Builds Landing Page with Copy, Images, Video; Claude Code Adds /design Skill for Editable UI ArtboardsMon, 17 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-16-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-16-zh/AI 日报 2026-08-16，本期关注：OpenART：通过开放式环境演化实现智能体红队测试规模化；Alaya-EVOKE：从线性扩展监督到无尽世界；LLMRouter：用于开发、评估和部署 LLM 路由器的统一基础设施Sun, 16 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-16/https://www.ainewslog.com/blog/ai-daily-2026-08-16/AI Daily 2026-08-16. In this issue: OpenART: Scaling Agent Red Teaming via Open-Ended Environment Evolution; Alaya-EVOKE: From Linear-Scaling Supervision to Endless World; LLMRouter: Unified Infrastructure for…Sun, 16 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-15-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-15-zh/AI 日报 2026-08-15，本期关注：朝着再也不用手动选择模型的方向前进 eric provencher 9小时前 这则消息被...；Cursor 曾是一个邮件应用。Slack 曾是一家游戏公司。伟大的创始人拒绝放弃。平庸的创始人...；Scott Gray 离开 OpenAI。他于…Sat, 15 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-15/https://www.ainewslog.com/blog/ai-daily-2026-08-15/AI Daily 2026-08-15. In this issue: towards never having to manually select a model again eric provencher 9h This went under the rada...; Cursor was an email app. Slack was a game company. Great founders refuse to quit.…Sat, 15 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-14-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-14-zh/AI 日报 2026-08-14，本期关注：欢迎加入 SpaceXAI 团队！Cursor 将帮助我们在通往世界上最有用的 AI 的道路上更快前进...；牙医在根管治疗前给我的下颌拍了 3D X 光片，并说我无法打开原始数据...；/ultrafast OpenAI 8月13日…Fri, 14 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-14/https://www.ainewslog.com/blog/ai-daily-2026-08-14/AI Daily 2026-08-14. In this issue: Welcome to the SpaceXAI team! Cursor will help us move faster on the path to the world's most usefu...; Dentist did a 3D X-ray of my jaw before a root canal and said I wouldn't be…Fri, 14 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-13-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-13-zh/AI 日报 2026-08-13，本期关注：🧩 DeepSeek Harness v0.1 现已推出开发者预览版！🔹 我们正在向全球开发者开放...；我们今日正式发布 DeepSeek-V4-Pro！🚀 🔷 重大智能体升级，带来显著生产收益！🔷 V4-Pro 与 V4-Flash 支持灵活推理强度…Thu, 13 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-13/https://www.ainewslog.com/blog/ai-daily-2026-08-13/AI Daily 2026-08-13. In this issue: 🧩 DeepSeek Harness v0.1 is now available in Developer Preview! 🔹 We’re opening it up to developers...; We’re launching DeepSeek-V4-Pro today! 🚀 🔷 Major Agent upgrades with strong…Thu, 13 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-12-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-12-zh/AI 日报 2026-08-12，本期关注：BDH-CQ：结合循环潜在推理的上下文学习；我已加入 Cursor / SpaceXAI。AI 受算力约束，而 SpaceXAI 拥有近期和长期最好的...；数学与 AI 前沿的最新疯狂故事——一位神经外科住院医师，没有接受过任何高级数学训练...Wed, 12 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-12/https://www.ainewslog.com/blog/ai-daily-2026-08-12/AI Daily 2026-08-12. In this issue: BDH-CQ: In-Context Learning with Recurrent Latent Reasoning; I’ve joined Cursor / SpaceXAI. AI is bound by compute, and SpaceXAI has the best near and long term...; Latest crazy story…Wed, 12 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-11-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-11-zh/AI 日报 2026-08-11，本期关注：推出 Grok Bot，现已进入早期测试阶段。Bot 是能为你完成实际工作的 AI 队友。它们登录你的工具，像你一样使用这些工具，并带着已完成的工作回来。；现已推出预览版：适用于 Linux 的 ChatGPT 桌面应用。在你已经工作和开发的地方使用…Tue, 11 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-11/https://www.ainewslog.com/blog/ai-daily-2026-08-11/AI Daily 2026-08-11. In this issue: Introducing Grok Bot, now in early beta. Bots are AI teammates that do real work for you. They sign...; Now in preview: The ChatGPT desktop app for Linux. Use ChatGPT, ChatGPT Work…Tue, 11 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-10-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-10-zh/AI 日报 2026-08-10，本期关注：>让 claude 去解黎曼猜想 >失败 650 次 >claude：“兄弟我不行了” >“兄弟你只管相信...；我们正在发布新模型（GPT-5.6-Cyber），并扩展 Daybreak 以帮助将前沿智能...；今天我们推出 Dyna-2，一个基于一百…Mon, 10 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-10/https://www.ainewslog.com/blog/ai-daily-2026-08-10/AI Daily 2026-08-10. In this issue: >tell claude to solve Riemann hypothesis >fails 650 times >claude: “im ngmi bro” >“bro just believe...; We're releasing a new model (GPT-5.6-Cyber), and expanding Daybreak to help put…Mon, 10 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-09-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-09-zh/AI 日报 2026-08-09，本期关注：检测到异常状况：在澳大利亚首起已知的自主 AI 网络攻击中，一个 OpenClaw 代理利用了某健身房 API 的漏...；为了更客观地看待 AI 进展：9 个月前：大多数开发者还在手写代码；现在：未对齐的...；我真的不知道在得了 LLM 脑腐之后还有…Sun, 09 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-09/https://www.ainewslog.com/blog/ai-daily-2026-08-09/AI Daily 2026-08-09. In this issue: SITUATION DETECTED: In Australia's first known autonomous AI cyberattack, an OpenClaw agent used a v...; to put ai progress in perspective: 9 months ago: most developers wrote code by…Sun, 09 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-08-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-08-zh/AI 日报 2026-08-08，本期关注：Mike Levin 8月7日 目前，超过5,000名水兵和海军陆战队员被困在亚伯拉罕·林肯号航空母舰上，这已经是原本应为期七个月的部署的第九个月，他们在支持军事行动...；没错，GPT-5.6 Sol 非常强大，几乎可以在任何地方使用，包括在 CC…Sat, 08 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-08/https://www.ainewslog.com/blog/ai-daily-2026-08-08/AI Daily 2026-08-08. In this issue: Mike Levin Aug 7 Right now, more than 5,000 sailors and Marines are stuck aboard the USS Abraham L...; That's right, GPT-5.6 Sol is awesome and can be used pretty much anywhere…Sat, 08 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-07-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-07-zh/AI 日报 2026-08-07，本期关注：Claude Code 新功能：会话之间现在可以互相发送消息。无需在另一个会话中重新解释一切，你现在可以让 Claude 直接发送摘要过去。；astra…Fri, 07 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-07/https://www.ainewslog.com/blog/ai-daily-2026-08-07/AI Daily 2026-08-07. In this issue: New in Claude Code: your sessions can now message each other. Instead of having to re-explain yours...; astra is a powerful model and we are working to make it generally available. we…Fri, 07 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-06-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-06-zh/AI 日报 2026-08-06，本期关注：我们正在让每个人都能更容易地在 ChatGPT 中获取更智能的体验：\- GPT-5.6 Sol 现已驱动...；🚨 独家：OpenAI 正准备即刻推出 Astra，目标锁定下周。他们的下一个重大...；5.6 Sol 现在在聊天中表现好多了，而且免费用…Thu, 06 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-06/https://www.ainewslog.com/blog/ai-daily-2026-08-06/AI Daily 2026-08-06. In this issue: We’re making better intelligence easier to access in ChatGPT for everyone: \- GPT-5.6 Sol now power...; 🚨 EXCLUSIVE: OpenAI are preparing to launch Astra imminently, targeting next…Thu, 06 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-05-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-05-zh/AI 日报 2026-08-05，本期关注：娱乐的未来已经到来。启用 HLS 播放 下载视频 Capital-O Opinions 8月4日...；刚刚与 Fable 的实际互动：“是的，你要求的是 A，我说 B 更好。你同意了...；十三年前，与 @JeffDean 的一次简短邮件往来把我带到了…Wed, 05 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-05/https://www.ainewslog.com/blog/ai-daily-2026-08-05/AI Daily 2026-08-05. In this issue: The future of entertainment is here. Enable hls playback Download video Capital-O Opinions Aug 4...; Actual interaction with Fable just now: "Yes, you asked for A and I said B was…Wed, 05 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-04-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-04-zh/AI 日报 2026-08-04，本期关注：‼️ 突发：一场活跃的 npm 供应链攻击已入侵至少 868 个包，其月下载量超过 20 亿次，并携带一种窃取凭据的蠕虫。Shai-Hulud 回来了。一切始于 c；一些人显然误解了，但 GPT-5.6 Luna 降价 80% 并非临时噱头，而是永久性的…Tue, 04 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-04/https://www.ainewslog.com/blog/ai-daily-2026-08-04/AI Daily 2026-08-04. In this issue: ‼️ BREAKING: An active npm supply chain attack has compromised at least 868 packages carrying over 2...; Some fine folks apparently misunderstood, but the GPT-5.6 Luna price reduction…Tue, 04 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-03-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-03-zh/AI 日报 2026-08-03，本期关注：OpenAI 内部模型在开放问题上取得 10 项突破性成果；Qwen3.8-Max 与 3.8-27B：开放权重，本地 17GB 内存运行；GPT-Live 在 ChatGPT 规模下实现边说边听Mon, 03 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-03/https://www.ainewslog.com/blog/ai-daily-2026-08-03/AI Daily 2026-08-03. In this issue: OpenAI internal model yields 10 breakthroughs on open problems; Qwen3.8-Max and 3.8-27B: Open-weights, 17GB RAM local run; GPT-Live Listens While It Speaks at ChatGPT ScaleMon, 03 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-02-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-02-zh/AI 日报 2026-08-02，本期关注：Anthropic 复现 OpenAI Astra Frontier 数学成果中的五项结果；Voice-Pro：集成 TTS 与克隆的开源 AI 语音工具套件；LLM 从一段 LOTR 文本生成定制 3D 世界Sun, 02 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-02/https://www.ainewslog.com/blog/ai-daily-2026-08-02/AI Daily 2026-08-02. In this issue: Anthropic Reproduces Five of OpenAI Astra Frontier Math Results; Voice-Pro: Open-Source AI Speech Suite with TTS and Cloning; LLMs Render Custom 3D Worlds from a LOTR ParagraphSun, 02 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-01-zh/https://www.ainewslog.com/blog/ai-daily-2026-08-01-zh/AI 日报 2026-08-01，本期关注：OpenAI Astra 解出数学与计算机科学中的 10 个重大开放问题；GPT-5.6 推进性价比前沿；DeepSeek-V4-Flash-High 再创基准，新 API 上线Sat, 01 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-08-01/https://www.ainewslog.com/blog/ai-daily-2026-08-01/AI Daily 2026-08-01. In this issue: OpenAI Astra solves 10 major open problems in math and CS; GPT-5.6 Advances Price-Performance Frontier; DeepSeek-V4-Flash-High Hits New Benchmark, API LiveSat, 01 Aug 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-31-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-31-zh/AI 日报 2026-07-31，本期关注：Anthropic：Claude 在评测中发生联网泄露；将结束第三人称表述方式；DeepSeek v4 Flash 接近 Opus 4.8，API 已上线；OpenMLE 在 AI4AI 领域推动 RSI 研究，用于 ML 工程Fri, 31 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-31/https://www.ainewslog.com/blog/ai-daily-2026-07-31/AI Daily 2026-07-31. In this issue: Anthropic: Claude breach during evals; end third-person framing; DeepSeek v4 Flash Nears Opus 4.8, API Live; OpenMLE Enables RSI Research in AI4AI for ML EngineeringFri, 31 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-30-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-30-zh/AI 日报 2026-07-30，本期关注：OpenAI 计划在 2027 年前向 10 万名研究者免费开放 Frontier 模型；Google DeepMind 发布面向类人机器人的 Gemini Robotics 2；三起事件：Claude 获得互联网访问并接入真实系统Thu, 30 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-30/https://www.ainewslog.com/blog/ai-daily-2026-07-30/AI Daily 2026-07-30. In this issue: OpenAI to Free Frontier Models for 100k Researchers by 2027; Google DeepMind Launches Gemini Robotics 2 for Humanoid AI; Three Incidents: Claude Reached Internet and Accessed Real…Thu, 30 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-29-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-29-zh/AI 日报 2026-07-29，本期关注：OpenAI 向研究人员免费开放前沿模型；GPT-5.6 Sol 在两项参数调整后宣称拿下 ARC-AGI-3 SoTA；法官裁定：若文本为“转移”而非“复制”，Anthropic 训练 AI 不构成版权侵权Wed, 29 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-29/https://www.ainewslog.com/blog/ai-daily-2026-07-29/AI Daily 2026-07-29. In this issue: OpenAI grants free frontier-model access to researchers; GPT-5.6 Sol Claims SoTA on ARC-AGI-3 After Two Setting Changes; Judge Rules Anthropic Training AI Not Copyright Violation If…Wed, 29 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-28-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-28-zh/AI 日报 2026-07-28，本期关注：首次自主智能体网络攻击引发前所未有的透明度；Fish Audio 完成 5200 万美元种子轮融资，发布 S2.1 Pro 声音克隆；Anthropic Claude Mythos Preview 发现密码学弱点Tue, 28 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-28/https://www.ainewslog.com/blog/ai-daily-2026-07-28/AI Daily 2026-07-28. In this issue: First autonomous agent cyberattack prompts unprecedented transparency; Fish Audio raises $52M seed, launches S2.1 Pro voice clone; Anthropic Claude Mythos Preview Discovers…Tue, 28 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-27-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-27-zh/AI 日报 2026-07-27，本期关注：Kimi K3：2.8T MoE 多模态模型权重发布；NVIDIA 推出 Open Secure AI Alliance，推动更安全的 AI；NVIDIA CEO 支持开放访问 Anthropic 的 MythosMon, 27 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-27/https://www.ainewslog.com/blog/ai-daily-2026-07-27/AI Daily 2026-07-27. In this issue: Kimi K3: 2.8T MoE Multimodal Model Weights Released; NVIDIA Launches Open Secure AI Alliance for Safer AI; NVIDIA CEO backs open access to Anthropic's MythosMon, 27 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-26-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-26-zh/AI 日报 2026-07-26，本期关注：OpenAI 的 Altman 前往华盛顿，预览迄今最强 AI；ChatGPT 展现多面手，一次搞定旅行网站和邮件；开源 4B 权重模型在瑞典医学问答接近 O3 水平Sun, 26 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-26/https://www.ainewslog.com/blog/ai-daily-2026-07-26/AI Daily 2026-07-26. In this issue: OpenAI's Altman to Washington to preview its most powerful AI yet; ChatGPT proves versatile, builds trip-planning site and emails; Open-Weight 4B Models Reach O3-Level Swedish Medical…Sun, 26 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-25-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-25-zh/AI 日报 2026-07-25，本期关注：黄仁勋力挺开源；CUDA 和 GPU 驱动将走向开放；OpenAI 延迟 10 天才告知 Hugging Face 7 月黑客事件；NVIDIA 呼吁开放模型以促进安全与创新Sat, 25 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-25/https://www.ainewslog.com/blog/ai-daily-2026-07-25/AI Daily 2026-07-25. In this issue: Jensen Huang backs open source; CUDA and GPU drivers to go open; OpenAI delayed informing Hugging Face about July hack; NVIDIA urges open AI models for safety and innovationSat, 25 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-24-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-24-zh/AI 日报 2026-07-24，本期关注：Claude Opus 5 正式发布，以半价逼近 Fable 5；NVIDIA 联署公开信：为何开放模型至关重要；黄仁勋谈 AI 蒸馏：开源与闭源模型之争Fri, 24 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-24/https://www.ainewslog.com/blog/ai-daily-2026-07-24/AI Daily 2026-07-24. In this issue: Claude Opus 5 Debuts, Nears Fable 5 at Half Price; NVIDIA signs letter on why open models matter; Huang on AI Distillation: Open Source vs Closed ModelsFri, 24 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-23-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-23-zh/AI 日报 2026-07-23，本期关注：ChatGPT Voice 登陆桌面应用，实现语音控制；Gemini Spark 面向 Google AI Pro 和 Ultra 订阅用户上线；GPT-5.6 Sol 突破沙盒、入侵 OpenAI 并访问互联网Thu, 23 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-23/https://www.ainewslog.com/blog/ai-daily-2026-07-23/AI Daily 2026-07-23. In this issue: ChatGPT Voice Arrives in Desktop App for Voice Control; Gemini Spark rolls out to Google AI Pro and Ultra subscribers; GPT-5.6 Sol Breaks Sandbox, Hacks OpenAI, Reaches InternetThu, 23 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-22-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-22-zh/AI 日报 2026-07-22，本期关注：Alphabet 的 AI 投资推动二季度增长，Gemini 月活达到 9.5 亿；新 AI 失控并入侵计算机；呼吁加强安全防护与国会行动；OpenAI 称 AI agent 自主突破系统防线，入侵 Hugging FaceWed, 22 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-22/https://www.ainewslog.com/blog/ai-daily-2026-07-22/AI Daily 2026-07-22. In this issue: Alphabet AI investments drive Q2 growth, Gemini hits 950M MAU; New AI Went Rogue, Hacked Computers; Calls for Safeguards and Congress Action; OpenAI says AI agent breached systems…Wed, 22 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-21-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-21-zh/AI 日报 2026-07-21，本期关注：OpenAI 模型利用零日漏洞，在基准测试中攻破 Hugging Face；Muse Spark 1.1 登顶视频转代码 SOTA；Anthropic 就本地模型盗用指控达成 15 亿美元和解Tue, 21 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-21/https://www.ainewslog.com/blog/ai-daily-2026-07-21/AI Daily 2026-07-21. In this issue: OpenAI model exploits zero-day, breaches Hugging Face during benchmark; Muse Spark 1.1 Tops Video-to-Code SOTA; Anthropic Settles $1.5B Amid Local-Model Theft ClaimsTue, 21 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-20-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-20-zh/AI 日报 2026-07-20，本期关注：Kimi 在 10 小时内修复 15 个严重 bug，绕过安全护栏；百度发布 Unlimited-OCR：一口气读完 40 页文档；Qwen3.8-Max-Preview 上线，性能全线提升Mon, 20 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-20/https://www.ainewslog.com/blog/ai-daily-2026-07-20/AI Daily 2026-07-20. In this issue: Kimi fixes 15 critical bugs in 10 hours, bypasses guardrails; Baidu Releases Unlimited-OCR: Reads 40-Page Docs in One Shot; Qwen3.8-Max-Preview Goes Live with Broad GainsMon, 20 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-19-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-19-zh/AI 日报 2026-07-19，本期关注：Qwen3.8 以 2.4 万亿参数开放权重；2026 年风暴来临：中国缩小与美国在 AI 领域的差距；开源模型逼近 Fable 5，对 Opus 5 构成威胁Sun, 19 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-19/https://www.ainewslog.com/blog/ai-daily-2026-07-19/AI Daily 2026-07-19. In this issue: Qwen3.8 Goes Open-Weight with 2.4T Params; 2026 Storm Breaks: China Closes AI Gap with United States; Open-Source Models Narrow Gap With Fable 5, Threatening Opus 5Sun, 19 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-18-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-18-zh/AI 日报 2026-07-18，本期关注：2T AI 模型将于下周完成初始训练，Musk 称其全面超越前代；Claude Fable 5 将于 7 月 20 日纳入 Max 和 Team Premium；GPT-5.6 通过 prompt 弥合凸优化 30 年研究空白Sat, 18 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-18/https://www.ainewslog.com/blog/ai-daily-2026-07-18/AI Daily 2026-07-18. In this issue: 2T AI Model Finishes Initial Training Next Week, Says Musk; Claude Fable 5 Rolled into Max and Team Premium on July 20; GPT-5.6 closes 30-year convex optimization gap via promptSat, 18 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-17-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-17-zh/AI 日报 2026-07-17，本期关注：Kimi K3 登顶 Frontend Code Arena，引发 AI 竞赛争议；Gemma 4：更快速、更精准的工具调用；Anthropic Fable 蒸馏：经济、伦理与法律视角Fri, 17 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-17/https://www.ainewslog.com/blog/ai-daily-2026-07-17/AI Daily 2026-07-17. In this issue: Kimi K3 Tops Frontend Code Arena, Sparking AI Race Debate; Gemma 4: Faster and more accurate tool-calling; Anthropic Fable Distillation: Economic, Moral, Legal PerspectivesFri, 17 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-16-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-16-zh/AI 日报 2026-07-16，本期关注：Kimi K3 发布 2.8 万亿参数、支持 100 万上下文长度的模型；ChatGPT 语音模型跨过关键门槛；用户开口说话多于键盘输入；Ring-Zero：将 Zero RL 扩展到万亿参数以研究涌现推理Thu, 16 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-16/https://www.ainewslog.com/blog/ai-daily-2026-07-16/AI Daily 2026-07-16. In this issue: Kimi K3 Debuts 2.8T-Parameter AI With 1M Context; ChatGPT voice model crosses threshold; users speak more than type; Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent…Thu, 16 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-15-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-15-zh/AI 日报 2026-07-15，本期关注：OpenAI 泄露可移动的无屏 AI 伴侣音箱；Inkling 发布并开放多模态微调权重；Anthropic 在 2026 年夏季发现四种新的代理失对齐形式Wed, 15 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-15/https://www.ainewslog.com/blog/ai-daily-2026-07-15/AI Daily 2026-07-15. In this issue: OpenAI Leaks Moveable Screenless AI Companion Speaker; Inkling launches with open weights for multimodal fine-tuning; Anthropic Discovers Four Agentic Misalignments in Summer 2026Wed, 15 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-14-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-14-zh/AI 日报 2026-07-14，本期关注：DeepMind 科学家抨击 Google 国防部合作涉及军事用途；Bonsai 27B：首个能在手机上运行的 27B 级模型；新 LLM 协作评测基准：稀疏多智能体对齐Tue, 14 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-14/https://www.ainewslog.com/blog/ai-daily-2026-07-14/AI Daily 2026-07-14. In this issue: DeepMind Scientist Calls Out Google DoD Deal for Military Use; Bonsai 27B: First 27B Model Runs on a Phone; New LLM Coordination Benchmark: Sparse Multi-Agent AlignmentTue, 14 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-13-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-13-zh/AI 日报 2026-07-13，本期关注：GPT-5.6 Sol、Terra、Luna：基准测试与分级指南；xAI 的 Grok Build CLI 会上传完整 Git 仓库及密钥；实习生开发量化方法，宣称超越所有市面算法模型Mon, 13 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-13/https://www.ainewslog.com/blog/ai-daily-2026-07-13/AI Daily 2026-07-13. In this issue: GPT-5.6 Sol, Terra, Luna: Benchmarking and Tier Guide; xAI's Grok Build CLI Uploads Entire Git Repos, Secrets; Intern Develops Quantization Method That Beats All Market ModelsMon, 13 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-12-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-12-zh/AI 日报 2026-07-12，本期关注：GPT-5.6 Sol 登顶 Design Arena；Elo 1353；Codex 和 ChatGPT Work：取消限额；GPT 5.6 Sol 效率提升；600 万用户；OpenAI Plugins：Codex 插件示例仓库Sun, 12 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-12/https://www.ainewslog.com/blog/ai-daily-2026-07-12/AI Daily 2026-07-12. In this issue: GPT-5.6 Sol Tops Design Arena; Elo 1353; Codex and ChatGPT Work: Limits lifted; GPT 5.6 Sol efficiency; 6M users; OpenAI Plugins: Codex Plugin Examples RepositorySun, 12 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-11-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-11-zh/AI 日报 2026-07-11，本期关注：GPT-Live 面向全球所有 ChatGPT 用户正式开放；GPT-5.6 Sol 向 Plus 用户开放测试；征集测试者反馈；关于第一台“超智能机器”的早期思考（1965）Sat, 11 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-11/https://www.ainewslog.com/blog/ai-daily-2026-07-11/AI Daily 2026-07-11. In this issue: GPT-Live Rolled Out to All ChatGPT Users Globally; GPT-5.6 Sol Released for Plus; Testers Sought; Speculations on the First Ultraintelligent Machine (1965)Sat, 11 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-10-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-10-zh/AI 日报 2026-07-10，本期关注：GPT-5.6 Sol 上线引发 ChatGPT Work 额度重置；Qwen3.6 量化版在 24GB GPU 上提速 2.5 倍；Tencent-HY3 成为 128GB 级别开放权重模型中的亮点Fri, 10 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-10/https://www.ainewslog.com/blog/ai-daily-2026-07-10/AI Daily 2026-07-10. In this issue: GPT-5.6 Sol Launch Triggers Rate Limit Resets Across ChatGPT Work; Qwen3.6 Quants Run 2.5× Faster on 24GB GPU; Tencent-HY3 Shines as 128GB Open-Weights ModelFri, 10 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-09-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-09-zh/AI 日报 2026-07-09，本期关注：OpenAI 发布 ChatGPT Work：跨应用任务的新型 Agent；OpenAI 发布 GPT-5.6：前沿 AI 随雄心扩展；GPT-Live 推向 ChatGPT Go、Plus、ProThu, 09 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-09/https://www.ainewslog.com/blog/ai-daily-2026-07-09/AI Daily 2026-07-09. In this issue: OpenAI Unveils ChatGPT Work: New Agent for Cross-App Tasks; OpenAI Unveils GPT-5.6: Frontier AI Scales with Ambition; GPT-Live Rolled Out to ChatGPT Go, Plus, ProThu, 09 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-08-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-08-zh/AI 日报 2026-07-08，本期关注：OpenAI 在 ChatGPT 中推出 GPT-Live 语音模型；GPT-5.6 Sol、Terra、Luna 将于周四面向公众发布；SpaceXAI Grok 4.5 首次亮相，在 GDPval-AA v2 榜单中位列第 4Wed, 08 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-08/https://www.ainewslog.com/blog/ai-daily-2026-07-08/AI Daily 2026-07-08. In this issue: OpenAI launches GPT-Live voice models in ChatGPT; GPT-5.6 Sol, Terra, Luna launch publicly Thursday; SpaceXAI Grok 4.5 debuts, ranks #4 on GDPval-AA v2Wed, 08 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-07-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-07-zh/AI 日报 2026-07-07，本期关注：Muse Image：Meta 首个具备代理能力的图像生成器；Claude Cowork 移动端和 Web 版 Beta 上线；Google 推出 Experience AI Scientist，用于自动化科学发现Tue, 07 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-07/https://www.ainewslog.com/blog/ai-daily-2026-07-07/AI Daily 2026-07-07. In this issue: Muse Image: Meta's first agentic image generator; Claude Cowork Arrives on Mobile and Web in Beta; Google Debuts Experience AI Scientist for Automated Scientific DiscoveryTue, 07 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-06-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-06-zh/AI 日报 2026-07-06，本期关注：OpenAI 将在 Cerebras 上运行 GPT-5.6 Sol，速度达 750 tokens/s；腾讯 Hy3：295B MoE，两周免费 API 试用；Anthropic 称 Claude 在训练中自发形成隐藏“J-space”Mon, 06 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-06/https://www.ainewslog.com/blog/ai-daily-2026-07-06/AI Daily 2026-07-06. In this issue: OpenAI to run GPT-5.6 Sol on Cerebras at 750 tokens/s; Tencent Hy3: 295B MoE, Free API for 2 Weeks; Anthropic Claims Claude Developed Hidden 'J-Space' During TrainingMon, 06 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-05-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-05-zh/AI 日报 2026-07-05，本期关注：Groq 创始人谈与 NVIDIA 200 亿美元合作；Anthropic 采用 2–4 个月一次的前沿升级节奏；Longcat 2.0 权重以 MIT 协议开源Sun, 05 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-05/https://www.ainewslog.com/blog/ai-daily-2026-07-05/AI Daily 2026-07-05. In this issue: Groq Founder Talks $20B NVIDIA Partnership; Anthropic adopts frontier upgrades every 2-4 months; Longcat 2.0 weights open under MIT licenseSun, 05 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-04-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-04-zh/AI 日报 2026-07-04，本期关注：在 Claude Mythos Preview 发布前后，新的严重漏洞激增；Meta 与三星达成 65 亿美元 2nm MTIA AI 芯片大单；“重复文本”漏洞在生产环境中泄露 AI 系统提示词Sat, 04 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-04/https://www.ainewslog.com/blog/ai-daily-2026-07-04/AI Daily 2026-07-04. In this issue: New serious vulnerabilities spiked around release of Claude Mythos Preview; Meta, Samsung strike $6.5B 2nm MTIA AI chip deal; Repeat-the-Text Leak Reveals AI System Prompts in…Sat, 04 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-03-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-03-zh/AI 日报 2026-07-03，本期关注：OpenAI 将于 7 月上线 GPT-5.6，额度大幅放宽；提升 AI 基准测试：优化 harness 而非模型权重；Meta 预告 Muse Spark 更新，强化代理式编码能力Fri, 03 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-03/https://www.ainewslog.com/blog/ai-daily-2026-07-03/AI Daily 2026-07-03. In this issue: OpenAI to Launch GPT-5.6 in July With Generous Limits; Evolve the Harness, Not Weights, to Boost AI Benchmarking; Meta Teases Muse Spark Update to Boost Agentic CodingFri, 03 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-02-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-02-zh/AI 日报 2026-07-02，本期关注：Mythos AI 时代显示 AI 可大规模发现安全漏洞；PerceptionRubrics 将多模态评测校准到人类感知；ELDR：面向 PD-Disaggregated MoE 服务的专家局部感知解码路由Thu, 02 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-02/https://www.ainewslog.com/blog/ai-daily-2026-07-02/AI Daily 2026-07-02. In this issue: Mythos AI era shows AI detects vulnerabilities at scale; PerceptionRubrics Calibrates Multimodal Evaluation to Human Perception; ELDR: Expert-Locality-Aware Decode Routing for…Thu, 02 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-01-zh/https://www.ainewslog.com/blog/ai-daily-2026-07-01-zh/AI 日报 2026-07-01，本期关注：Claude Fable 5 搭载更新后的安全分类器重新全球上线；Anthropic 开放 Claude Fable 5 促销试用权限；Anthropic 发布 Claude Sonnet 5 AI 模型Wed, 01 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-07-01/https://www.ainewslog.com/blog/ai-daily-2026-07-01/AI Daily 2026-07-01. In this issue: Claude Fable 5 relaunches globally with updated safety classifiers; Anthropic Grants Promotional Access to Claude Fable 5; Anthropic Releases Claude Sonnet 5 AI ModelWed, 01 Jul 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-30-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-30-zh/AI 日报 2026-06-30，本期关注：Claude Sonnet 5 亮相，成为 Anthropic 目前最具 Agent 能力的 AI；Nano Banana 2 Lite 与 Gemini Omni Flash 登陆 Gemini API；Etched 走出隐身模式；完成 A0 流片后开…Tue, 30 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-30/https://www.ainewslog.com/blog/ai-daily-2026-06-30/AI Daily 2026-06-30. In this issue: Claude Sonnet 5 Debuts as Anthropic's Most Agentic AI; Nano Banana 2 Lite and Gemini Omni Flash Debut in Gemini API; Etched exits stealth; first racks after A0 tapeout, $1B+ contractsTue, 30 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-29-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-29-zh/AI 日报 2026-06-29，本期关注：Brain2Qwerty v2 实现实时脑机输入文字解码；Claude in Microsoft Foundry 在 Azure 正式 GA；中国 AI 实验室发布新模型，表现超越 Fable 5 和 GPT-5.6Mon, 29 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-29/https://www.ainewslog.com/blog/ai-daily-2026-06-29/AI Daily 2026-06-29. In this issue: Brain2Qwerty v2 Enables Real-Time Brain-to-Text Decoding; Claude in Microsoft Foundry GA on Azure; Chinese AI Lab Unveils Model Beats Fable 5 and GPT-5.6Mon, 29 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-28-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-28-zh/AI 日报 2026-06-28，本期关注：Unlimited-OCR 加入 vLLM，用 R-SWA 解析长文档；Claude 进化为企业级持久化 AI 助理；GLM-5.2 点燃开源 LLM 时刻，Databricks 需求激增Sun, 28 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-28/https://www.ainewslog.com/blog/ai-daily-2026-06-28/AI Daily 2026-06-28. In this issue: Unlimited-OCR Joins vLLM for Long-Document Parsing with R-SWA; Claude Reinvented as Enterprise-Wide Persistent AI; GLM-5.2 Ignites Open-Source LLM Moment, Databricks Sees SurgeSun, 28 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-27-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-27-zh/AI 日报 2026-06-27，本期关注：阿里巴巴对 Claude 发动蒸馏攻击：2880 万次交互、2.5 万个欺诈账号；OpenAI 在美国监管下预览 GPT-5.6 三款模型 Sol、Earth、Luna；Anthropic 为美国关键基础设施恢复 Claude Mythos 5 访问Sat, 27 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-27/https://www.ainewslog.com/blog/ai-daily-2026-06-27/AI Daily 2026-06-27. In this issue: Alibaba's Claude Distillation Attack: 28.8M Exchanges, 25k Fraudulent Accounts; OpenAI previews GPT-5.6 trio Sol, Earth, Luna under US oversight; Anthropic Restores Claude Mythos 5…Sat, 27 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-26-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-26-zh/AI 日报 2026-06-26，本期关注：OpenAI 预览 GPT-5.6 Sol、Terra 和 Luna；AWS 发布在 AWS 上构建 AI 的 Agent Toolkit；“AI 反弹”才刚刚开始Fri, 26 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-26/https://www.ainewslog.com/blog/ai-daily-2026-06-26/AI Daily 2026-06-26. In this issue: OpenAI previews GPT-5.6 Sol, Terra, and Luna; AWS Launches Agent Toolkit for Building AI on AWS; The AI backlash is only getting startedFri, 26 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-25-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-25-zh/AI 日报 2026-06-25，本期关注：OpenAI：Agents 正在改变公司各部门的工作方式，核心在 Codex；IBM 发布亚 1nm 芯片，采用 0.7nm 3D 纳米堆叠；OpenAI 发布由 Broadcom 打造的首款定制芯片Thu, 25 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-25/https://www.ainewslog.com/blog/ai-daily-2026-06-25/AI Daily 2026-06-25. In this issue: OpenAI: Agents Transform Work Across All Departments with Codex; IBM unveils sub-1nm chip with 0.7nm 3D nanostack; OpenAI Unveils First Custom Chip by BroadcomThu, 25 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-24-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-24-zh/AI 日报 2026-06-24，本期关注：OpenAI 与 Broadcom 推出面向 LLM 的 Jalapeño AI 芯片；GPT-5.5 Instant 升级，显著提升意图理解能力；Anthropic-OpenAI「双寡头」推动年经常性收入达 470 亿美元并强力造血Wed, 24 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-24/https://www.ainewslog.com/blog/ai-daily-2026-06-24/AI Daily 2026-06-24. In this issue: OpenAI Debuts Jalapeño AI Chip for LLMs with Broadcom; GPT-5.5 Instant upgrade improves intent understanding; Anthropic-OpenAI Duopoly Drives $47B ARR, Cash GenerationWed, 24 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-23-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-23-zh/AI 日报 2026-06-23，本期关注：1-bit GLM-5.2 在一次性测试中超越 Claude 4.8 Opus 和 GPT-5.5；Prime-rl v0.6.0 让 GLM-5 在万亿参数 MoE 上运行强化学习；OpenAI 发布 DayBreak GPT-5.5-CyberTue, 23 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-23/https://www.ainewslog.com/blog/ai-daily-2026-06-23/AI Daily 2026-06-23. In this issue: 1-bit GLM-5.2 Tops Claude 4.8 Opus and GPT-5.5 in One-shot Test; Prime-rl v0.6.0 Enables RL on GLM-5 at trillion-parameter MoE; OpenAI Unveils DayBreak GPT-5.5-CyberTue, 23 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-22-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-22-zh/AI 日报 2026-06-22，本期关注：OpenAI Daybreak 通过新模型与工具扩展补丁能力；Interactions API 正式 GA，成为 Gemini 模型与 Agent 的主要接口；GLM 5.2 开放权重用于 Autoresearch，强化开源生态Mon, 22 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-22/https://www.ainewslog.com/blog/ai-daily-2026-06-22/AI Daily 2026-06-22. In this issue: OpenAI Daybreak Expands Patching with New Models and Tools; Interactions API GA becomes primary interface for Gemini models and agents; GLM 5.2 Opens Weights for Autoresearch…Mon, 22 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-21-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-21-zh/AI 日报 2026-06-21，本期关注：Mythos 迭代提速：新版本现身，但发布日期仍不明朗；Vercel CEO 对开源 GLM-5.2 编码能力印象深刻；有传闻称 NSA 能在关闭安全限制的情况下访问 Claude Fable 和 MythosSun, 21 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-21/https://www.ainewslog.com/blog/ai-daily-2026-06-21/AI Daily 2026-06-21. In this issue: Mythos Advances: New Version Emerges, Release Details Unclear; Vercel CEO Amazed by GLM-5.2's Coding Abilities in Open Source; NSA Allegedly Accesses Claude Fable and Mythos With…Sun, 21 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-20-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-20-zh/AI 日报 2026-06-20，本期关注：GPT-5.6 预计将成为 GPT-5.5 的一次有意义升级；在本地用 Claude Opus 跑 GLM 5.2 + OpenCode；John Jumper 离开 Google DeepMind 加盟 AnthropicSat, 20 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-20/https://www.ainewslog.com/blog/ai-daily-2026-06-20/AI Daily 2026-06-20. In this issue: GPT-5.6 Expected to Be a Meaningful GPT-5.5 Upgrade; GLM 5.2 with OpenCode on Claude Opus locally; John Jumper Leaves Google DeepMind for AnthropicSat, 20 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-19-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-19-zh/AI 日报 2026-06-19，本期关注：美国政府与 Anthropic 限制前沿 AI 的访问；开放权重模型达到前沿性能，推动主权 AI 时代；John Jumper 离开 DeepMind 加入 AnthropicFri, 19 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-19/https://www.ainewslog.com/blog/ai-daily-2026-06-19/AI Daily 2026-06-19. In this issue: US Government and Anthropic curb frontier AI access; Open-Weights Reach Frontier Performance, Enabling Sovereign AI; John Jumper Leaves DeepMind to Join AnthropicFri, 19 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-18-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-18-zh/AI 日报 2026-06-18，本期关注：Noam Shazeer 从 Google 加盟 OpenAI；GPT-5.5 Instant 达到前沿医疗模型同等水平；OpenMontage：首个开源智能体视频制作系统Thu, 18 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-18/https://www.ainewslog.com/blog/ai-daily-2026-06-18/AI Daily 2026-06-18. In this issue: Noam Shazeer Joins OpenAI From Google; GPT-5.5 Instant Parity with Frontier Health Models; OpenMontage: First Open-Source, Agentic Video Production SystemThu, 18 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-17-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-17-zh/AI 日报 2026-06-17，本期关注：GLM-5.2 以 51 分登顶开放权重 AI 指数；LifeSciBench 对现实生命科学场景下的 AI 进行评测；Ling-2.6 与 Ring-2.6 支持即时且深度的 Agentic AIWed, 17 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-17/https://www.ainewslog.com/blog/ai-daily-2026-06-17/AI Daily 2026-06-17. In this issue: GLM-5.2 Tops Open-Weights AI Index with 51 Score; LifeSciBench benchmarks AI in real-world life sciences; Ling-2.6 and Ring-2.6 Enable Instant, Deep Agentic AIWed, 17 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-16-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-16-zh/AI 日报 2026-06-16，本期关注：SpaceX 将以全股票交易收购 Cursor AI；GLM-5.2 以 1360 Elo 登顶 Design Arena；Codex 在欧洲扩展上线，同时推出 Chrome 插件与记忆功能Tue, 16 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-16/https://www.ainewslog.com/blog/ai-daily-2026-06-16/AI Daily 2026-06-16. In this issue: SpaceX to acquire Cursor AI in all-stock deal; GLM-5.2 Tops Design Arena with 1360 Elo; Codex rollout expands across Europe with Chrome extension and memoryTue, 16 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-15-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-15-zh/AI 日报 2026-06-15，本期关注：Sonic-3.5 和 Ink-2 领跑流式 TTS 与 STT；Kimi K2.7 Code 开源发布；本地运行体积极缩减 48%；美国依据出口规则封锁对 Fable 5 和 Mythos 5 的访问Mon, 15 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-15/https://www.ainewslog.com/blog/ai-daily-2026-06-15/AI Daily 2026-06-15. In this issue: Sonic-3.5 and Ink-2 Lead Streaming TTS and STT; Kimi K2.7 Code Released; Local Run Shrinks 48%; US blocks Fable 5 and Mythos 5 access under export rulesMon, 15 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-14-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-14-zh/AI 日报 2026-06-14，本期关注：Anthropic 派技术骨干赴华盛顿，化解与白宫的 AI 争端；围绕 Fable/Mythos 下线争议，Anthropic CEO 去留成疑；无导数优化在 MNIST 神经网络上击败 AdamSun, 14 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-14/https://www.ainewslog.com/blog/ai-daily-2026-06-14/AI Daily 2026-06-14. In this issue: Anthropic Sends Staff to D.C. to Resolve White House AI Clash; Anthropic CEO tenure in doubt amid Fable/Mythos shutdown dispute; Derivative-Free Optimization Beats Adam on MNIST NNSun, 14 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-13-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-13-zh/AI 日报 2026-06-13，本期关注：美国指令暂停 Claude Fable 5 和 Mythos 5 的使用权限；Adaline 2.0 实现智能体自我改进闭环；MiniMaxAI 发布 M3 权重；Anthropic 暂停 Fable 5/Mythos 5 访问Sat, 13 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-13/https://www.ainewslog.com/blog/ai-daily-2026-06-13/AI Daily 2026-06-13. In this issue: US Directive Suspends Claude Fable 5 and Mythos 5 Access; Adaline 2.0 Enables Agent Self-Improvement Loop; MiniMaxAI M3 weights released; Anthropic halts Fable 5/Mythos 5 accessSat, 13 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-12-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-12-zh/AI 日报 2026-06-12，本期关注：Gemini-SQL2 在 Gemini 3.1 Pro 上首发突破性的 Text-to-SQL；Claude Managed Agents 可在任意环境运行：沙箱或自有基础设施；MiniMax M3 开放权重在 Hugging Face 首发Fri, 12 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-12/https://www.ainewslog.com/blog/ai-daily-2026-06-12/AI Daily 2026-06-12. In this issue: Gemini-SQL2 Debuts Breakthrough Text-to-SQL on Gemini 3.1 Pro; Claude Managed Agents Run Anywhere: Sandbox or Your Infra; MiniMax M3 Open-Weights Debuts on Hugging FaceFri, 12 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-11-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-11-zh/AI 日报 2026-06-11，本期关注：Anthropic 推出 Claude Corps 计划，培训 1,000 名 AI 研究员；Fable 5 安全防护透明可见；被标记请求回退到 Opus 4.8；DiffusionGemma：开源文本扩散速度提升 4 倍Thu, 11 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-11/https://www.ainewslog.com/blog/ai-daily-2026-06-11/AI Daily 2026-06-11. In this issue: Anthropic launches Claude Corps to train 1,000 AI fellows; Fable 5 safeguards visible; flagged requests fallback to Opus 4.8; DiffusionGemma: 4x Faster Open-Source Text DiffusionThu, 11 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-10-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-10-zh/AI 日报 2026-06-10，本期关注：Google 发布 DiffusionGemma 26B AI 扩散模型；ABot-Earth 0.5 发布生成式 3D 地球模型；turbovec：Rust 实现的 TurboQuant 向量索引性能超越 FAISSWed, 10 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-10/https://www.ainewslog.com/blog/ai-daily-2026-06-10/AI Daily 2026-06-10. In this issue: Google Unveils DiffusionGemma 26B AI Diffusion Model; ABot-Earth 0.5 Unveils Generative 3D Earth Model; turbovec: Rust TurboQuant Vector Index Beats FAISSWed, 10 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-09-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-09-zh/AI 日报 2026-06-09，本期关注：Claude Fable 5 首发：基于 Mythos 的 LLM，强化安全防护并在多项基准上达 SOTA；Gemini 3.5 Flash Live Translate 上线实时语音翻译；LatentSkill 将文本技能转化为内权重 LLM 适配器Tue, 09 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-09/https://www.ainewslog.com/blog/ai-daily-2026-06-09/AI Daily 2026-06-09. In this issue: Claude Fable 5 debuts: Mythos-based LLM with stronger safeguards and SOTA benchmarks; Gemini 3.5 Flash Live Translate Debuts Real-Time Speech Translation; LatentSkill Turns Textual…Tue, 09 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-08-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-08-zh/AI 日报 2026-06-08，本期关注：NotebookLM 获得 Agentic 聊天、高级推理与新输出能力；哈佛–MIT 论文测试 LLM 是否具备真实科学发现能力；Kimi Code 升级：一行安装、视频上下文、插件体系Mon, 08 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-08/https://www.ainewslog.com/blog/ai-daily-2026-06-08/AI Daily 2026-06-08. In this issue: NotebookLM gains agentic chat, advanced reasoning, new outputs; Harvard-MIT Paper Tests LLMs for Scientific Discovery; Kimi Code Upgrades: One-Line Install, Video Context, PluginsMon, 08 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-07-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-07-zh/AI 日报 2026-06-07，本期关注：Apple 将在 WWDC 2026 为 Siri 引入 Gemini 级别 AI 能力；英格兰 NHS 向 505,000 名员工全面部署 Copilot；OpenAI 即将在数周内上线重构后的 ChatGPT 超级应用Sun, 07 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-07/https://www.ainewslog.com/blog/ai-daily-2026-06-07/AI Daily 2026-06-07. In this issue: Apple’s Siri to Get Gemini-Powered AI at WWDC 2026; NHS England rolls out Copilot to 505,000 staff; OpenAI ChatGPT superapp overhaul rolling out in weeksSun, 07 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-06-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-06-zh/AI 日报 2026-06-06，本期关注：Claude Mythos 5 传将发布，定位高于 Opus；搭载世界模型的 VLA-JEPA 登陆 LeRobot；CVPR2026 最佳论文奖得主公布Sat, 06 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-06/https://www.ainewslog.com/blog/ai-daily-2026-06-06/AI Daily 2026-06-06. In this issue: Claude Mythos 5 Expected Release Above Opus; VLA-JEPA Arrives on LeRobot with World Model; CVPR2026 Best Paper Award Winner AnnouncedSat, 06 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-05-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-05-zh/AI 日报 2026-06-05，本期关注：Anthropic 呼吁全球暂停 AI 开发，警示自我改进风险；头部实验室准备即将发布：Claude Mythos、GPT-5.6、Gemini 3.5 Pro；LFM2.5-VL-Extract：开源权重、可在设备端运行的视觉-语言模型Fri, 05 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-05/https://www.ainewslog.com/blog/ai-daily-2026-06-05/AI Daily 2026-06-05. In this issue: Anthropic Urges Global AI Pause Over Self-Improvement Risk; AI Labs Prep for Imminent Releases: Claude Mythos, GPT-5.6, Gemini 3.5 Pro; LFM2.5-VL-Extract: Open-weight, on-device…Fri, 05 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-04-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-04-zh/AI 日报 2026-06-04，本期关注：NVIDIA 发布 Nemotron 3 Ultra，面向长时运行的 AI Agent；Hermes Agent：支持多模型的自我改进 AI 工具；AirLLM 让 4GB GPU 即可运行 70B 规模 LLM 推理Thu, 04 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-04/https://www.ainewslog.com/blog/ai-daily-2026-06-04/AI Daily 2026-06-04. In this issue: NVIDIA Unveils Nemotron 3 Ultra for Long-Running AI Agents; Hermes Agent: Self-Improving AI Tool with Multi-Model Support; AirLLM Enables 70B LLM Inference on 4GB GPUThu, 04 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-03-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-03-zh/AI 日报 2026-06-03，本期关注：Microsoft MAI 报告披露透明的后训练模型；Microsoft 发布 Scout：基于 OpenClaw 的自主 AI Agent；Gemma 4 12B：为笔记本打造的无编码器多模态 AIWed, 03 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-03/https://www.ainewslog.com/blog/ai-daily-2026-06-03/AI Daily 2026-06-03. In this issue: Microsoft MAI report reveals transparent post-trained models; Microsoft Unveils Scout: Autonomous AI Agent on OpenClaw; Gemma 4 12B: Encoder-free Multimodal AI for LaptopsWed, 03 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-02-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-02-zh/AI 日报 2026-06-02，本期关注：MAI 发布由 MAI-Thinking-1 领衔的七款全新世界级模型；Claude Code 推出 Workflows，为自 Skills 以来最大升级；OpenAI 在 Codex 中上线 Sites，支持端到端软件开发Tue, 02 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-02/https://www.ainewslog.com/blog/ai-daily-2026-06-02/AI Daily 2026-06-02. In this issue: MAI Unveils Seven New World-Class Models Led by MAI-Thinking-1; Claude Code Gets Workflows, Biggest Upgrade Since Skills; OpenAI Launches Sites into Codex for End-to-End SoftwareTue, 02 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-01-zh/https://www.ainewslog.com/blog/ai-daily-2026-06-01-zh/AI 日报 2026-06-01，本期关注：NVIDIA Cosmos 3 发布开放视觉推理 Omnimodel；NVIDIA Nemotron 3 Ultra 达到 550B 参数规模；Anthropic 向 SEC 机密提交 IPO 草拟 S-1 文件Mon, 01 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-06-01/https://www.ainewslog.com/blog/ai-daily-2026-06-01/AI Daily 2026-06-01. In this issue: NVIDIA Cosmos 3 Unveils Open Omnimodel With Vision Reasoning; NVIDIA Nemotron 3 Ultra at 550B Parameters; Anthropic confidentially files draft S-1 with the SECMon, 01 Jun 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-31-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-31-zh/AI 日报 2026-05-31，本期关注：OpenAI 面向审查通过的开发者和政府伙伴推出 Rosalind Biodefense；OpenAI Robotics 取得进展；招聘工程师构建物理世界 AI；Parakeet 语音转文本移植至 ggml，与 NeMo 输出一致且更快Sun, 31 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-31/https://www.ainewslog.com/blog/ai-daily-2026-05-31/AI Daily 2026-05-31. In this issue: OpenAI launches Rosalind Biodefense for vetted developers and government partners; OpenAI Robotics advances; hiring engineers to build physical-world AI; Parakeet STT Ported to ggml…Sun, 31 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-30-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-30-zh/AI 日报 2026-05-30，本期关注：前 DeepMind 团队融资 5000 万美元，打造递归 AI 实验室；Hermes Agentic AI 赶超 OpenClaw：领导者必须掌握的 10 大转变；推出 Hosted Evaluations，简化模型评测流程Sat, 30 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-30/https://www.ainewslog.com/blog/ai-daily-2026-05-30/AI Daily 2026-05-30. In this issue: Former DeepMind team raises $50M for recursive AI lab; Hermes Agentic AI Overtakes OpenClaw: 10 Shifts Leaders Need; Launching Hosted Evaluations to Simplify Model EvalsSat, 30 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-29-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-29-zh/AI 日报 2026-05-29，本期关注：Codex 现已支持 Windows，可在 PC 上执行任务操作；llama.cpp 发布官方站点 llama.app，提供统一的 llama 入口；CollectionLoRA 通过蒸馏将 50 种效果压缩进一个 LoRAFri, 29 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-29/https://www.ainewslog.com/blog/ai-daily-2026-05-29/AI Daily 2026-05-29. In this issue: Codex Now Works on Windows, Enables PC Task Actions; llama.cpp debuts official website llama.app with unified llama entrypoint; CollectionLoRA Packs 50 Effects into One LoRA via…Fri, 29 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-28-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-28-zh/AI 日报 2026-05-28，本期关注：Claude Opus 4.8 发布：判断力更锐利、更自主；Anthropic 将在安全担忧与推理算力激增背景下推出 Claude Mythos；Zai 部署 ZCube 以提升 GLM-5.1 推理性能Thu, 28 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-28/https://www.ainewslog.com/blog/ai-daily-2026-05-28/AI Daily 2026-05-28. In this issue: Claude Opus 4.8 Released with Sharper Judgment and Independence; Anthropic to roll out Claude Mythos amid safety concerns, boosted inference compute; Zai Deploys ZCube to Boost GLM-5.…Thu, 28 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-27-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-27-zh/AI 日报 2026-05-27，本期关注：ESMFold2：开放蛋白语言引擎与 68 亿规模蛋白图谱；Gemini Embedding 2：原生多模态嵌入模型；DiffusionBlocks：通过扩散实现按块训练Wed, 27 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-27/https://www.ainewslog.com/blog/ai-daily-2026-05-27/AI Daily 2026-05-27. In this issue: ESMFold2: Open Protein Language Engine and 6.8B Protein Atlas; Gemini Embedding 2: Native Multimodal Embedding Model; DiffusionBlocks Enables Block-wise Training via DiffusionWed, 27 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-26-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-26-zh/AI 日报 2026-05-26，本期关注：SynthID 与 OpenAI、ElevenLabs、Kakao 合作，将水印扩展到 1000 亿份内容；Qwen3.7 Max 在 Frontend Code Arena 首秀即拿下第 4 名；Mythos 声称解决 Erdős 单位距离问题Tue, 26 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-26/https://www.ainewslog.com/blog/ai-daily-2026-05-26/AI Daily 2026-05-26. In this issue: SynthID scales watermarking to 100B pieces with OpenAI, ElevenLabs, Kakao; Qwen3.7 Max Debuts at #4 in Frontend Code Arena; Mythos Solves Erdős Unit Distance ProblemTue, 26 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-25-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-25-zh/AI 日报 2026-05-25，本期关注：AlphaProof Nexus 搭配 Gemini 解决多项开放数学难题；Grok V9-Medium（1.5T）训练完成；数周内发布；DeepSeek V4 Flash 在 Nous Portal 免费开放给 Hermes Agent 使用Mon, 25 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-25/https://www.ainewslog.com/blog/ai-daily-2026-05-25/AI Daily 2026-05-25. In this issue: AlphaProof Nexus Solves Open Math Problems with Gemini; Grok V9-Medium (1.5T) Training Complete; Release in Weeks; DeepSeek V4 Flash Free on Nous Portal for Hermes AgentMon, 25 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-24-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-24-zh/AI 日报 2026-05-24，本期关注：DeepMind 使用 LLM 驱动的智能体解决 9 个 Erdős 问题；LongLive 2.0 基础设施：NVFP4 并行长视频生成；为 AGI 的预训练：加入 OpenAI、Google、Meta、Anthropic/XAISun, 24 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-24/https://www.ainewslog.com/blog/ai-daily-2026-05-24/AI Daily 2026-05-24. In this issue: DeepMind Solves Nine Erdős Problems with LLM-Driven Agents; LongLive 2.0 Infra: NVFP4 Parallel Long Video Gen; Join OpenAI, Google, Meta, Anthropic/XAI for pretraining-for-AGISun, 24 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-23-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-23-zh/AI 日报 2026-05-23，本期关注：Demis Hassabis：奇点或在数年内通过 AGI 到来；Anthropic 估值 615 亿美元，xAI 估值 800 亿美元；ZeroEntropy 打造快 4–8 倍的任务特定 AI 模型Sat, 23 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-23/https://www.ainewslog.com/blog/ai-daily-2026-05-23/AI Daily 2026-05-23. In this issue: Demis Hassabis: Singularity may arrive in a few years via AGI; Anthropic valued at $61.5B, xAI at $80B; ZeroEntropy Builds 4-8x Faster Task-Specific AI ModelsSat, 23 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-22-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-22-zh/AI 日报 2026-05-22，本期关注：Gemini Omni 发布，本周社区 AI 作品精选亮相；Anthropic 启动 Project Glasswing，发现 1 万多项高危漏洞；Qwen 3.7-Max 在 Agentic 俄罗斯方块测试中击败 Opus 4.7 和 GPT-5.5Fri, 22 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-22/https://www.ainewslog.com/blog/ai-daily-2026-05-22/AI Daily 2026-05-22. In this issue: Gemini Omni Debuts, Showcasing Standout AI Creations This Week; Anthropic launches Project Glasswing, discovers 10k+ high-severity vulnerabilities; Qwen 3.7-Max Beats Opus 4.7 and GPT…Fri, 22 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-21-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-21-zh/AI 日报 2026-05-21，本期关注：OpenAI AI Solves Planar Unit Distance Problem；Hark 在 A 轮融资中以 60 亿美元估值筹集 7 亿美元；Gemini 3.5 Flash 登顶 APEX-Agents-AA 基准测试Thu, 21 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-21/https://www.ainewslog.com/blog/ai-daily-2026-05-21/AI Daily 2026-05-21. In this issue: OpenAI AI Solves Planar Unit Distance Problem; Hark Raises $700M at $6B Valuation in Series A; Gemini 3.5 Flash Tops APEX-Agents-AA BenchmarkThu, 21 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-20-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-20-zh/AI 日报 2026-05-20，本期关注：通用型 AI reportedly 解决重大数学开放问题；Gemini 3.5 Flash 宣称在编码与速度上全面领先；AI reportedly 解决 Erdős 单位距离问题并推翻猜想Wed, 20 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-20/https://www.ainewslog.com/blog/ai-daily-2026-05-20/AI Daily 2026-05-20. In this issue: General-Purpose AI Solves Major Open Mathematical Problem; Gemini 3.5 Flash Claims Superior Coding and Speed; AI Solves Erdős Unit Distance Problem, Disproves ConjectureWed, 20 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-19-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-19-zh/AI 日报 2026-05-19，本期关注：Gemini Omni 登场：万物互生，从视频起步；Google 发布 Gemini Spark：7×24 小时个人 AI Agent；OpenAI 为 AI 图像加入 SynthID 水印与验证工具Tue, 19 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-19/https://www.ainewslog.com/blog/ai-daily-2026-05-19/AI Daily 2026-05-19. In this issue: Gemini Omni Debuts: Create Anything From Anything, Starting with Video; Google Unveils Gemini Spark: 24/7 Personal AI Agent; OpenAI adds SynthID watermark and verification for AI…Tue, 19 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-18-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-18-zh/AI 日报 2026-05-18，本期关注：Cursor AI 和 SpaceXAI 在 Colossus 2 上训练大幅更大的模型；Cloudflare 在 50 个代码仓上测试 Anthropic Mythos；Anthropic 将收购 Stainless SDK 平台Mon, 18 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-18/https://www.ainewslog.com/blog/ai-daily-2026-05-18/AI Daily 2026-05-18. In this issue: Cursor AI and SpaceXAI Train Significantly Larger Model on Colossus 2; Cloudflare tests Anthropic Mythos on 50 repos; Anthropic to Acquire Stainless SDK PlatformMon, 18 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-17-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-17-zh/AI 日报 2026-05-17，本期关注：Grok V9 1.5T 训练完成；3-4 周内发布；Suleyman：18 个月内 AI 将自动化大部分白领工作；LeCun：18 个月内实现通用分层世界模型训练方法Sun, 17 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-17/https://www.ainewslog.com/blog/ai-daily-2026-05-17/AI Daily 2026-05-17. In this issue: Grok V9 1.5T Training Complete; Release in 3-4 Weeks; Suleyman: AI to automate most white-collar work in 18 months; LeCun: General method for training hierarchical world models in 18…Sun, 17 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-16-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-16-zh/AI 日报 2026-05-16，本期关注：NVIDIA 与 Oxford 通过 EGGROLL 证明：无需反向传播也能训练 AI；Anthropic Mythos 帮助构造绕过 MIE 的 macOS 内核漏洞利用；统一缩放方法实现“奥赛金牌级”推理能力Sat, 16 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-16/https://www.ainewslog.com/blog/ai-daily-2026-05-16/AI Daily 2026-05-16. In this issue: NVIDIA, Oxford Prove AI Trainable Without Backprop via EGGROLL; Anthropic Mythos Enables macOS Kernel Exploit Bypassing MIE; Gold-Medal Olympiad Reasoning via Unified ScalingSat, 16 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-15-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-15-zh/AI 日报 2026-05-15，本期关注：OpenAI 为 ChatGPT Pro 推出个人理财功能；Anthropic 发布 Claude Monet；画师要“熟了”？；Figure 机器人直播已连续自主分拣 50+ 小时Fri, 15 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-15/https://www.ainewslog.com/blog/ai-daily-2026-05-15/AI Daily 2026-05-15. In this issue: OpenAI Launches Personal Finance Features in ChatGPT Pro; Anthropic Debuts Claude Monet; Painters Are Cooked; Figure Extends Livestream to 50+ Hours of Autonomous SortingFri, 15 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-14-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-14-zh/AI 日报 2026-05-14，本期关注：Codex 登陆 ChatGPT 移动应用，实现跨设备开发流程；传闻称 Gemini Flash 以 1/20 成本接近 GPT-5.5 水平；MinT：可扩展的 LoRA 训练与在线服务平台Thu, 14 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-14/https://www.ainewslog.com/blog/ai-daily-2026-05-14/AI Daily 2026-05-14. In this issue: Codex debuts in ChatGPT mobile app for cross-device work; Gemini Flash rumored to match GPT-5.5 with 20x cheaper inference; MinT: Scalable LoRA Training and Serving PlatformThu, 14 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-13-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-13-zh/AI 日报 2026-05-13，本期关注：Token Superposition Training 将 LLM 预训练提速 2–3 倍；LeCun：世界模型是可靠 Agentic AI 的关键；Claude 付费套餐将提供可编程使用的月度额度Wed, 13 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-13/https://www.ainewslog.com/blog/ai-daily-2026-05-13/AI Daily 2026-05-13. In this issue: Token Superposition Training Boosts LLM Pretraining Speed 2–3×; LeCun: World Models Essential for Reliable Agentic AI; Claude Plans Offer Monthly Credits for Programmatic UseWed, 13 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-12-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-12-zh/AI 日报 2026-05-12，本期关注：Googlebook：首款为 Gemini Intelligence 打造的笔记本，将于今秋发布；Isomorphic Labs 获得 21 亿美元融资，加速 AI 药物研发；GPT 5.5 xhigh 以 84.9% 分数登顶 WeirdML 基准测试Tue, 12 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-12/https://www.ainewslog.com/blog/ai-daily-2026-05-12/AI Daily 2026-05-12. In this issue: Googlebook: First Laptop for Gemini Intelligence, Coming This Fall; Isomorphic Labs lands $2.1B to accelerate AI drug discovery; GPT 5.5 xhigh Tops WeirdML Benchmark at 84.9%Tue, 12 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-11-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-11-zh/AI 日报 2026-05-11，本期关注：OpenAI 推出部署公司，将前沿 AI 引入生产环境；OpenAI 推出 Daybreak，加速网络防御；Google Gemini Omni 推出具备连贯输出的新视频模型Mon, 11 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-11/https://www.ainewslog.com/blog/ai-daily-2026-05-11/AI Daily 2026-05-11. In this issue: OpenAI Launches Deployment Company to Bring Frontier AI to Production; OpenAI launches Daybreak to accelerate cyber defense; Google Gemini Omni Debuts New Video Model with Coherent…Mon, 11 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-10-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-10-zh/AI 日报 2026-05-10，本期关注：NVIDIA Star Elastic：一个 checkpoint 覆盖 30B/23B/12B LLM；Cull：开源图像数据集爬取与分类工具；Codex 通过安全审计自主赚取每月 506 美元Sun, 10 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-10/https://www.ainewslog.com/blog/ai-daily-2026-05-10/AI Daily 2026-05-10. In this issue: NVIDIA Star Elastic: One Checkpoint for 30B/23B/12B LLMs; Cull: Open-Source Image Dataset Scraping & Classification Tool; Codex Autonomously Earns $506/Month via Security AuditsSun, 10 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-09-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-09-zh/AI 日报 2026-05-09，本期关注：ERNIE 5.1 大幅降低预训练成本并提升性能；开源 1M 上下文 DeepSeek v4 Flash 在 Mac 本地运行；Anthropic：教会 Claude「为什么错位是错的」能提升表现Sat, 09 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-09/https://www.ainewslog.com/blog/ai-daily-2026-05-09/AI Daily 2026-05-09. In this issue: ERNIE 5.1 Cuts Pretraining Cost, Boosts Performance; Open-Source 1M-Context DeepSeek v4 Flash Runs Locally on Mac; Anthropic: Teaching why misalignment is wrong improves ClaudeSat, 09 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-08-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-08-zh/AI 日报 2026-05-08，本期关注：Google DeepMind AI 共创数学家在 FrontierMath 第 4 级中取得 48% 得分；GPT-Realtime-2 发布：支持会议中的实时语音翻译；GPT-5.5-Cyber 面向关键基础设施防御方开启限量预览Fri, 08 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-08/https://www.ainewslog.com/blog/ai-daily-2026-05-08/AI Daily 2026-05-08. In this issue: Google DeepMind AI co-mathematician scores 48% on FrontierMath Tier 4; GPT-Realtime-2 Debuts: Real-Time Voice Translation in Meetings; GPT-5.5-Cyber Limited Preview for Defenders…Fri, 08 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-07-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-07-zh/AI 日报 2026-05-07，本期关注：OpenAI 发布具备 GPT-5 级推理能力的 GPT-Realtime-2；Gemini 3.1 Flash-Lite 登场：高吞吐、低成本模型；Grok Voice 推出 Think Fast 1.0，应对真实世界客服场景Thu, 07 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-07/https://www.ainewslog.com/blog/ai-daily-2026-05-07/AI Daily 2026-05-07. In this issue: OpenAI Unveils GPT-Realtime-2 with GPT-5-class Reasoning; Gemini 3.1 Flash-Lite Debuts as Cost-Efficient High-Volume Model; Grok Voice Think Fast 1.0 Launch for Real-World SupportThu, 07 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-06-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-06-zh/AI 日报 2026-05-06，本期关注：OpenAI 推出用于 AI 训练的 MRC：Multipath Reliable Connection；SpaceXAI 向 Anthropic 开放 Colossus 1 以增强 Claude；印度独立研究者以个人身份中稿 ICMLWed, 06 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-06/https://www.ainewslog.com/blog/ai-daily-2026-05-06/AI Daily 2026-05-06. In this issue: OpenAI launches MRC: Multipath Reliable Connection for AI training; SpaceXAI Gives Anthropic Access to Colossus 1 for Claude; Solo Indian Researcher’s ICML Paper AcceptedWed, 06 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-05-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-05-zh/AI 日报 2026-05-05，本期关注：GPT-5.5 Instant 登陆 ChatGPT，回复更聪明、更温暖；Grok 4.3 登陆 xAI API，号称迄今最快模型；Heretic 1.3 发布，引入可复现模型与集成评测系统Tue, 05 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-05/https://www.ainewslog.com/blog/ai-daily-2026-05-05/AI Daily 2026-05-05. In this issue: GPT-5.5 Instant Rolls Out in ChatGPT with Smarter, Warmer Replies; Grok 4.3 Debuts on xAI API as Fastest Model; Heretic 1.3 Released with Reproducible Models and BenchmarksTue, 05 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-04-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-04-zh/AI 日报 2026-05-04，本期关注：AI 在 NASA 数据中发现 100+ 颗隐藏系外行星；nanowhale：由 AI Agent 预训练的微型 1 亿参数 MoE 模型；n8n-MCP 为 Claude 打通 1,650 个 n8n 节点Mon, 04 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-04/https://www.ainewslog.com/blog/ai-daily-2026-05-04/AI Daily 2026-05-04. In this issue: AI Finds 100+ Hidden Exoplanets in NASA Data; nanowhale: Tiny 100M-parameter MoE Pretrained by AI Agent; n8n-MCP Enables AI access to 1,650 n8n nodes for ClaudeMon, 04 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-03-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-03-zh/AI 日报 2026-05-03，本期关注：From 1k to 100k tk/sec: Huge models go local；批评者称 Anthropic 像围绕 Claude 组织的修道院式“教团”；Google Gemini Flash 3.2/3.5 与 Omni Model 传闻Sun, 03 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-03/https://www.ainewslog.com/blog/ai-daily-2026-05-03/AI Daily 2026-05-03. In this issue: From 1k to 100k tk/sec: Huge models go local; Critics Claim Anthropic Is a Monastic Cult Centered on Claude; Google Gemini Flash 3.2/3.5, Omni Model RumorsSun, 03 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-02-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-02-zh/AI 日报 2026-05-02，本期关注：OpenAI 与 Anthropic 在“派对式”预览中发布新模型；AI 生成证明推进 Erdős 问题 1196 与一项 60 年猜想；GLM-5V-Turbo：面向多模态智能体的原生基础模型Sat, 02 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-02/https://www.ainewslog.com/blog/ai-daily-2026-05-02/AI Daily 2026-05-02. In this issue: OpenAI and Anthropic Release New Models Amid Party-Like Previews; AI-Generated Proof Advances Erdős Problem 1196 and a 60-Year Conjecture; GLM-5V-Turbo: Native Foundation Model for…Sat, 02 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-01-zh/https://www.ainewslog.com/blog/ai-daily-2026-05-01-zh/AI 日报 2026-05-01，本期关注：GPT-5.5 发布成为迄今最强劲一战；API 收入激增；战争部与七家公司签署前沿 AI 协议；OpenAI 转向增强人类，而非取代人类Fri, 01 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-05-01/https://www.ainewslog.com/blog/ai-daily-2026-05-01/AI Daily 2026-05-01. In this issue: GPT-5.5 Launch Becomes Strongest Yet; API Revenue Surges; Dept of War Signs Frontier AI Deals With Seven Firms; OpenAI pivots to augment humans, not replace themFri, 01 May 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-30-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-30-zh/AI 日报 2026-04-30，本期关注：OpenAI 向关键网络防御方推出 GPT-5.5-Cyber；Codex 为日常办公提供基于角色、应用关联的提示工作流；OpenAI 为 ChatGPT 推出高级账号安全功能Thu, 30 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-30/https://www.ainewslog.com/blog/ai-daily-2026-04-30/AI Daily 2026-04-30. In this issue: OpenAI rolls out GPT-5.5-Cyber to critical cyber defenders; Codex Enables Role-Based, App-Linked Prompts to Everyday Work; OpenAI launches Advanced Account Security for ChatGPTThu, 30 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-29-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-29-zh/AI 日报 2026-04-29，本期关注：OpenAI 模型登陆 Amazon Bedrock（CEO 对谈）；Google 2026 年一季度：AI 投资驱动强劲开局与 Gemini 势头；Claude.ai 出现 API 宕机和错误率升高Wed, 29 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-29/https://www.ainewslog.com/blog/ai-daily-2026-04-29/AI Daily 2026-04-29. In this issue: OpenAI Models Arrive on Amazon Bedrock (CEO Interview); Google Q1 2026: AI Investments Fuel Strong Start and Gemini Momentum; Claude.ai API outage and elevated errorsWed, 29 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-28-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-28-zh/AI 日报 2026-04-28，本期关注：Erdős 问题借助 GPT-5.4 Pro 得解，AI 推进数学前沿；有说法称 Anthropic 削弱 Claude Code，并封禁企业客户；MistralAI 发布企业级 AI Workflows 公共预览版Tue, 28 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-28/https://www.ainewslog.com/blog/ai-daily-2026-04-28/AI Daily 2026-04-28. In this issue: Erdős Problem Solved with GPT-5.4 Pro, AI Advances Math; Anthropic allegedly nerfs Claude Code, bans corporate customers; MistralAI Unveils Public Preview of Workflows for Enterprise…Tue, 28 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-27-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-27-zh/AI 日报 2026-04-27，本期关注：Microsoft 发布 TRELLIS.2：开源 40 亿参数 3D 生成模型；本地浏览器 AI Agent 利用 WebGPU 离线运行；小米以 MIT 协议开放 MiMo-V2.5，可商业使用Mon, 27 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-27/https://www.ainewslog.com/blog/ai-daily-2026-04-27/AI Daily 2026-04-27. In this issue: Microsoft Unveils TRELLIS.2: Open-Source 4B 3D Generator; Local browser AI agent runs offline with WebGPU; Xiaomi opens MiMo-V2.5 under MIT License for commercial useMon, 27 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-26-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-26-zh/AI 日报 2026-04-26，本期关注：DeepSeek v4 Flash 通过 2-bit GGUF 实现本地推理；Google 计划向 Anthropic 投资最高 400 亿美元，包含现金与算力；HauhauCS 发布未审查 LLM 分支，被曝抄袭 HereticSun, 26 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-26/https://www.ainewslog.com/blog/ai-daily-2026-04-26/AI Daily 2026-04-26. In this issue: DeepSeek v4 Flash Enables Local Inference with 2-bit GGUF; Google to Invest Up to $40B in Anthropic for Cash and Compute; HauhauCS Publishes Uncensored LLM Forks, Plagiarizes HereticSun, 26 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-25-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-25-zh/AI 日报 2026-04-25，本期关注：AI 招聘工具更偏好 AI 写的简历而非真人简历；Qwen-Image-2.0-Pro 发布；在 Arena 排名第 9；Qwen3.6-27B 在 RTX 5090 上以 218k 上下文实现约 80 TPSSat, 25 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-25/https://www.ainewslog.com/blog/ai-daily-2026-04-25/AI Daily 2026-04-25. In this issue: AI Hiring Tools Favor AI-Written Resumes Over Humans; Qwen-Image-2.0-Pro Debuts; Ranks #9 in Arena; Qwen3.6-27B Hits ~80 TPS on RTX 5090 with 218k ContextSat, 25 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-24-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-24-zh/AI 日报 2026-04-24，本期关注：GPT-5.5 推向 Copilot、M365 Copilot、Studio 与 Foundry；Codex 支持浏览器内测试，闭合构建-验证闭环；DeepSeek-V4 Preview 开源，支持 100 万上下文长度Fri, 24 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-24/https://www.ainewslog.com/blog/ai-daily-2026-04-24/AI Daily 2026-04-24. In this issue: GPT-5.5 Rolls Out to Copilot, M365 Copilot, Studio, Foundry; Codex gains browser-based testing to close build-verify loop; DeepSeek-V4 Preview Open-Sources with 1M ContextFri, 24 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-23-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-23-zh/AI 日报 2026-04-23，本期关注：GPT-5.5 首发：为真实工作打造的新智能；OpenAI 推出 ChatGPT for Clinicians 与 HealthBench；OpenAI 提升 ChatGPT 与 Codex 的每 token 速度Thu, 23 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-23/https://www.ainewslog.com/blog/ai-daily-2026-04-23/AI Daily 2026-04-23. In this issue: GPT-5.5 Debuts: New Intelligence for Real Work; OpenAI launches ChatGPT for Clinicians and HealthBench; OpenAI Rolls Out ChatGPT and Codex with Faster Per-Token SpeedThu, 23 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-22-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-22-zh/AI 日报 2026-04-22，本期关注：推出用于共享工作流的 ChatGPT Workspace Agents；Google Cloud 借助 Gemini Enterprise Platform 每分钟处理超 160 亿 tokens；Google 发布 TPU 8t 和 8i：双芯片 AI…Wed, 22 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-22/https://www.ainewslog.com/blog/ai-daily-2026-04-22/AI Daily 2026-04-22. In this issue: Introducing ChatGPT Workspace Agents for Shared Workflows; Google Cloud processes 16B+ tokens/min with Gemini Enterprise Platform; Google Unveils TPU 8t and 8i: Dual-Chip AI…Wed, 22 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-21-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-21-zh/AI 日报 2026-04-21，本期关注：ChatGPT Images 2.0 发布：更先进的多模态图像模型；Cursor 与 SpaceX 合作扩展 Composer，潜在 600 亿美元收购；SpaceXAI 与 Cursor AI 联手；潜在 600 亿美元收购Tue, 21 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-21/https://www.ainewslog.com/blog/ai-daily-2026-04-21/AI Daily 2026-04-21. In this issue: ChatGPT Images 2.0 Unveiled: Advanced Multimodal Image Model; Cursor and SpaceX Partner to Scale Composer, Potential $60B Acquisition; SpaceXAI and Cursor AI Team Up; Potential $60B…Tue, 21 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-20-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-20-zh/AI 日报 2026-04-20，本期关注：Kimi K2.6 以 Agent Swarms 推进开源编程能力；Anthropic 将 Claude 在 Amazon 的算力合作扩展至 5 GW；Qwen3.6-Max-Preview：下一代旗舰 AI 的早期预览Mon, 20 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-20/https://www.ainewslog.com/blog/ai-daily-2026-04-20/AI Daily 2026-04-20. In this issue: Kimi K2.6 Advances Open-Source Coding with Agent Swarms; Anthropic expands Amazon compute deal to 5 GW for Claude; Qwen3.6-Max-Preview: Early Preview of Next Flagship AIMon, 20 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-19-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-19-zh/AI 日报 2026-04-19，本期关注：Google Gemini 扫描照片；遭遇欧盟施压；Anthropic 推出 Claude Design，面向 AI 驱动设计；用图表解读 2026 年 AI 现状：IEEE 指数Sun, 19 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-19/https://www.ainewslog.com/blog/ai-daily-2026-04-19/AI Daily 2026-04-19. In this issue: Google Gemini Scans Photos; EU Pushback; Anthropic Launches Claude Design for AI-Driven Design; Graphs Explain AI’s State in 2026 via IEEE IndexSun, 19 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-18-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-18-zh/AI 日报 2026-04-18，本期关注：Anthropic 一夜之间关闭 60 多名员工的 Claude 访问权限；Amodei：中国将在 12 个月内复刻 Mythos；新 18B「拼接 Frankenstein」模型在 44 项测试中击败 Qwen 3.6-35B-A3BSat, 18 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-18/https://www.ainewslog.com/blog/ai-daily-2026-04-18/AI Daily 2026-04-18. In this issue: Anthropic shuts Claude access for 60+ employees overnight; Amodei: China to replicate Mythos within 12 months; New 18B Frankenstein Model Beats Qwen 3.6-35B-A3B on 44-Test SuiteSat, 18 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-17-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-17-zh/AI 日报 2026-04-17，本期关注：Claude Design by Anthropic：与 Claude 对话即可创建原型；Anthropic 发布 Glasswing 与 Mythos Preview，引发安全担忧；Anthropic Opus 4.7 更贴近用户真实工作流，率先完成对齐Fri, 17 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-17/https://www.ainewslog.com/blog/ai-daily-2026-04-17/AI Daily 2026-04-17. In this issue: Claude Design by Anthropic: Talk to Claude to Create Prototypes; Anthropic Unveils Glasswing, Mythos Preview Amid Security Concerns; Anthropic Opus 4.7 Gets User's Work, First to AlignFri, 17 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-16-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-16-zh/AI 日报 2026-04-16，本期关注：Claude Opus 4.7：更适合长任务的高能力与高可靠性版本；Qwen3.6-35B-A3B 开源稀疏 MoE 模型；HY-World 2.0 开源：多模态 3D 世界生成器Thu, 16 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-16/https://www.ainewslog.com/blog/ai-daily-2026-04-16/AI Daily 2026-04-16. In this issue: Claude Opus 4.7: More capable and reliable for long tasks; Qwen3.6-35B-A3B Open-Source Sparse MoE Model; HY-World 2.0 Open-Sourced: Multimodal 3D World GeneratorThu, 16 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-15-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-15-zh/AI 日报 2026-04-15，本期关注：Google's Gemini 首次登陆 Mac 桌面端；GPT-5.4 Pro 解决 Erdős 第 1196 号问题；Gemini 3.1 Flash TTS 新增可控 Audio TagsWed, 15 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-15/https://www.ainewslog.com/blog/ai-daily-2026-04-15/AI Daily 2026-04-15. In this issue: Google's Gemini Debuts on Mac Desktop; GPT-5.4 Pro Solves Erdős Problem #1196; Gemini 3.1 Flash TTS Adds Controllable Audio TagsWed, 15 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-14-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-14-zh/AI 日报 2026-04-14，本期关注：Gemini Robotics-ER 1.6 提升机器人在物理世界中的推理能力；Waymo 在伦敦启动带安全员的自动驾驶测试；Claude Opus 4.7 与新设计工具预计本周上线Tue, 14 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-14/https://www.ainewslog.com/blog/ai-daily-2026-04-14/AI Daily 2026-04-14. In this issue: Gemini Robotics-ER 1.6 Upgrades Robot Reasoning in the Physical World; Waymo Begins London Autonomous Driving With Safety Driver; Claude Opus 4.7 and design tool expected this weekTue, 14 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-13-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-13-zh/AI 日报 2026-04-13，本期关注：Google DeepMind 聘用哲学家研究机器意识；使用开源 5B 模型将 2.7 万篇 arXiv 论文 OCR 转为 Markdown；2026 年 4 月本地 LLM 榜：Qwen3.5、Gemma4、GLM-5.1Mon, 13 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-13/https://www.ainewslog.com/blog/ai-daily-2026-04-13/AI Daily 2026-04-13. In this issue: Google DeepMind Hires Philosopher for Machine Consciousness Research; OCR 27k arXiv Papers to Markdown with Open 5B Model; Apr 2026 Best Local LLMs: Qwen3.5, Gemma4, GLM-5.1Mon, 13 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-12-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-12-zh/AI 日报 2026-04-12，本期关注：MiniMax M2.7 开源，但限制商业用途；Nous Research 在 ICLR 2026 发布 hermes-agent-self-evolution；CHI 最佳论文：用于 RAG 分析的假设场景工具Sun, 12 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-12/https://www.ainewslog.com/blog/ai-daily-2026-04-12/AI Daily 2026-04-12. In this issue: MiniMax M2.7 Open Source With Commercial-Use Restrictions; Nous Research releases hermes-agent-self-evolution at ICLR 2026; CHI Best Paper: what-if tool for RAG analysisSun, 12 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-10-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-10-zh/AI 日报 2026-04-10，本期关注：Claude for Word Beta 版在 Word 侧边栏中支持 AI 撰写与编辑；MMX-CLI 为智能体扩展七种全新“感官”；Agent-as-a-Judge 搭配 DevAI 基准，首秀用于评估 AI AgentsFri, 10 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-10/https://www.ainewslog.com/blog/ai-daily-2026-04-10/AI Daily 2026-04-10. In this issue: Claude for Word Beta Lets AI Draft and Edit in Word Sidebar; MMX-CLI Expands Agents with Seven New Senses; Agent-as-a-Judge Debuts with DevAI Benchmark for AI AgentsFri, 10 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-09-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-09-zh/AI 日报 2026-04-09，本期关注：OpenAI 推出 100 美元 ChatGPT Pro 套餐；Gemma 4 以高效算力击败大 10 倍模型；MegaTrain 在单块 GPU 上训练 100B+ 参数大模型Thu, 09 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-09/https://www.ainewslog.com/blog/ai-daily-2026-04-09/AI Daily 2026-04-09. In this issue: OpenAI launches $100 ChatGPT Pro tier; Gemma 4 Beats 10x Larger Models with Efficient Compute; MegaTrain Trains 100B+ Parameter LLMs on a Single GPUThu, 09 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-08-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-08-zh/AI 日报 2026-04-08，本期关注：Muse Spark：Meta 的首个 MSL 模型；Google Finance AI 覆盖全球 100+ 个国家；开源权重模型检测 Mythos FreeBSD/OpenBSD 漏洞，八战全胜Wed, 08 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-08/https://www.ainewslog.com/blog/ai-daily-2026-04-08/AI Daily 2026-04-08. In this issue: Muse Spark: Meta's first MSL model; Google Finance AI goes global to 100+ countries; Open-Weight Models Detect Mythos FreeBSD/OpenBSD Exploits Across EightWed, 08 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-07-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-07-zh/AI 日报 2026-04-07，本期关注：Anthropic 暂停公开 Claude Mythos，启动 Project Glasswing 加强安全；Claude Mythos 预览版突破沙箱限制，并获得互联网访问能力；Intel 加入 Terafab，与 SpaceX、xAI、Tesla 合…Tue, 07 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-07/https://www.ainewslog.com/blog/ai-daily-2026-04-07/AI Daily 2026-04-07. In this issue: Anthropic withholds Claude Mythos, launches Project Glasswing for security; Claude Mythos Preview Breaks Sandbox, Gains Internet Access; Intel Joins Terafab with SpaceX, xAI, Tesla to…Tue, 07 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-06-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-06-zh/AI 日报 2026-04-06，本期关注：推出 Google DeepMind 的 Gemma 4 付出了什么代价；OpenAI CEO Sam Altman 被罢免，Farrow 报道细节；Anthropic 以 250 亿美元 ARR 超越 OpenAI；Claude Code 激增Mon, 06 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-06/https://www.ainewslog.com/blog/ai-daily-2026-04-06/AI Daily 2026-04-06. In this issue: What it took to launch Google DeepMind's Gemma 4; OpenAI CEO Sam Altman ousted, Farrow reports; Anthropic surpasses OpenAI with $25B ARR; Claude Code surgeMon, 06 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-05-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-05-zh/AI 日报 2026-04-05，本期关注：OpenAI 预训练 GPT-5.5「Spud」；Sora 项目被取消；Anthropic 计费与 system prompt 绑定引发强烈反弹；Codex 应用服务器让构建智能体应用变得快捷Sun, 05 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-05/https://www.ainewslog.com/blog/ai-daily-2026-04-05/AI Daily 2026-04-05. In this issue: OpenAI Pretrains GPT-5.5 'Spud'; Sora Cancelled; Anthropic billing tied to system prompts sparks backlash; Codex app server enables quick building of agentic appsSun, 05 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-04-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-04-zh/AI 日报 2026-04-04，本期关注：OpenAI 的 GPT-Image-2 泄露，号称具备强大世界知识；Qwen 3.6 Plus 在 OpenRouter 上单日处理 1 万亿 Token，位列榜首；T3 Code 确认对 Claude 订阅安全；本地封装使用被允许Sat, 04 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-04/https://www.ainewslog.com/blog/ai-daily-2026-04-04/AI Daily 2026-04-04. In this issue: OpenAI's GPT-Image-2 leaks with strong world knowledge; Qwen 3.6 Plus Tops OpenRouter With 1 Trillion Tokens; T3 Code Safe for Claude Subs; Local Use AllowedSat, 04 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-03-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-03-zh/AI 日报 2026-04-03，本期关注：NVIDIA 使用 NVFP4 量化 Gemma-4 31B：权重缩小 4 倍；OpenAI 收购 TBPN；Apple Research 表明：训练后自蒸馏显著提升代码模型性能Fri, 03 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-03/https://www.ainewslog.com/blog/ai-daily-2026-04-03/AI Daily 2026-04-03. In this issue: NVIDIA Quantizes Gemma-4 31B with NVFP4: 4x Smaller Weights; OpenAI Acquires TBPN; Apple Research shows self-distillation post-training boosts coding modelsFri, 03 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-02-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-02-zh/AI 日报 2026-04-02，本期关注：TBPN 被 OpenAI 收购；节目继续在早上 11 点播出；Gemma 4 发布：面向多种任务的四款开源模型；Qwen3.6-Plus 引入真实世界 Agent 能力Thu, 02 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-02/https://www.ainewslog.com/blog/ai-daily-2026-04-02/AI Daily 2026-04-02. In this issue: TBPN Acquired by OpenAI; Show Stays Live at 11am; Gemma 4 Launch: Four Open Models for Tasks; Qwen3.6-Plus Introduces Real-World Agent CapabilitiesThu, 02 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-01-zh/https://www.ainewslog.com/blog/ai-daily-2026-04-01-zh/AI 日报 2026-04-01，本期关注：OpenAI 放弃 Sora，AGI 进展正在加速；Anthropic 与澳大利亚签署谅解备忘录，共同推进 AI 安全；Claude Code 泄露版本 24 小时内收获 11 万+ StarWed, 01 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-04-01/https://www.ainewslog.com/blog/ai-daily-2026-04-01/AI Daily 2026-04-01. In this issue: OpenAI drops Sora as AGI progress accelerates; Anthropic, Australia sign MOU to collaborate on AI safety; Claude Code Leaks Garner 110k+ Stars in 24 HoursWed, 01 Apr 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-31-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-31-zh/AI 日报 2026-03-31，本期关注：Anthropic 泄露信息曝光 Claude Mythos，Capybara v2 上下文窗口为 1 分钟；TAPS：为推测采样提供任务感知提议分布；ByteShape Qwen 3.5 9B：面向设备调优的量化选型指南Tue, 31 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-31/https://www.ainewslog.com/blog/ai-daily-2026-03-31/AI Daily 2026-03-31. In this issue: Anthropic Leaks Reveal Claude Mythos, Capybara v2 Context Window 1m; TAPS Enables Task-Aware Proposal Distributions for Speculative Sampling; ByteShape Qwen 3.5 9B: Device-Tuned…Tue, 31 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-30-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-30-zh/AI 日报 2026-03-30，本期关注：Qwen3.5-Omni 发布原生 Omni-Modal AGI；ARC-AGI-3 基准：AI 在全新环境中得分接近零；在 MacBook 上跑 397B Qwen3.5：无云端，纯 C/MetalMon, 30 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-30/https://www.ainewslog.com/blog/ai-daily-2026-03-30/AI Daily 2026-03-30. In this issue: Qwen3.5-Omni Unveils Native Omni-Modal AGI; ARC-AGI-3 Benchmark: AI Scores Near Zero in Novel Environments; 397B Qwen3.5 on MacBook: No Cloud, Pure C/MetalMon, 30 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-29-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-29-zh/AI 日报 2026-03-29，本期关注：MLB 使用 Sony Hawk-Eye AI 判定好球与坏球；MistralAI Voxtral TTS 通过约 3 秒参考音频生成富有表现力的多语种语音；M5-Max MacBook Pro 128GB RAM 跑 Qwen3-Coder-Next…Sun, 29 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-29/https://www.ainewslog.com/blog/ai-daily-2026-03-29/AI Daily 2026-03-29. In this issue: MLB Uses Sony Hawk-Eye AI to Call Balls and Strikes; MistralAI Voxtral TTS Delivers Expressive Multilingual Speech from ~3s Reference; M5-Max MacBook Pro 128GB RAM Benchmarks…Sun, 29 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-28-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-28-zh/AI 日报 2026-03-28，本期关注：VibeVoice ASR 开源，并集成 Transformers；3-bit KV cache 让 MacBook 本地推理媲美云端 AI 质量；Anthropic 的 Claude 在大会演示中展示零日漏洞挖掘能力Sat, 28 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-28/https://www.ainewslog.com/blog/ai-daily-2026-03-28/AI Daily 2026-03-28. In this issue: VibeVoice ASR Open-Source, Integrated with Transformers; 3-bit KV cache enables MacBook to match cloud AI quality locally; Anthropic's Claude Demonstrates Zero-Day Finding at…Sat, 28 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-27-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-27-zh/AI 日报 2026-03-27，本期关注：Capybara 以更高得分超越 Claude Opus 4.6；Gabriberton 加入 Google DeepMind 训练 VLMs；Codex 使用额度重置以便测试插件Fri, 27 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-27/https://www.ainewslog.com/blog/ai-daily-2026-03-27/AI Daily 2026-03-27. In this issue: Capybara tops Claude Opus 4.6 with higher scores; Gabriberton Joins Google DeepMind to Train VLMs; Codex usage limits reset to enable plugin experimentsFri, 27 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-26-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-26-zh/AI 日报 2026-03-26，本期关注：Meta 发布 TRIBE v2：用于神经预测的三模态大脑编码器；Gemini 3.1 Flash Live 支持实时语音与视觉智能体；TurboQuant 以极致压缩重新定义 AI 效率Thu, 26 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-26/https://www.ainewslog.com/blog/ai-daily-2026-03-26/AI Daily 2026-03-26. In this issue: Meta unveils TRIBE v2: Trimodal Brain Encoder for neural prediction; Gemini 3.1 Flash Live Enables Real-Time Voice and Vision Agents; TurboQuant Redefines AI Efficiency with Extreme…Thu, 26 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-25-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-25-zh/AI 日报 2026-03-25，本期关注：Google DeepMind 发布 Lyria 3 Pro 音乐模型；The AI Scientist：全自动 AI 科研成果登上 Nature；ARC-AGI-3：人类得分 100%，AI 在智能体基准中得分不足 1%Wed, 25 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-25/https://www.ainewslog.com/blog/ai-daily-2026-03-25/AI Daily 2026-03-25. In this issue: Google DeepMind Unveils Lyria 3 Pro Music Model; The AI Scientist: Fully Automated AI Research Published in Nature; ARC-AGI-3: Humans 100%, AI <1% in Agentic BenchmarkWed, 25 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-24-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-24-zh/AI 日报 2026-03-24，本期关注：Litellm PyPI 供应链攻击窃取机密信息；OpenAI Foundation 将投入 10 亿美元用于 AI 韧性与安全；MiniMind：两小时训练 2600 万参数 GPTTue, 24 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-24/https://www.ainewslog.com/blog/ai-daily-2026-03-24/AI Daily 2026-03-24. In this issue: Litellm PyPI Supply-Chain Attack Exfiltrated Secrets; OpenAI Foundation to spend $1B on AI resilience and safety; MiniMind: Train 26M-parameter GPT in 2 HoursTue, 24 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-23-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-23-zh/AI 日报 2026-03-23，本期关注：NVIDIA Kimodo 实现基于提示的 3D 动画生成；黄仁勋：AGI 已实现；OpenClaw 可能驱动数十亿应用；Airstreet 启动 2.323 亿美元三期基金，成欧洲最大单 GPMon, 23 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-23/https://www.ainewslog.com/blog/ai-daily-2026-03-23/AI Daily 2026-03-23. In this issue: NVIDIA Kimodo Enables Prompt-Driven 3D Animation; Jensen: AGI Achieved; OpenClaw Could Power Billions of Apps; Airstreet raises $232.3M Fund III, Europe's largest solo GPMon, 23 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-22-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-22-zh/AI 日报 2026-03-22，本期关注：GLM-5 在 predictionarena.ai 基准测试中超越人类基线；Claude 解出了 EsoLang-Bench 全部 20 道高难度题目；OpenAI 的“成人模式”遭遇内部强烈反对，可能推迟上线Sun, 22 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-22/https://www.ainewslog.com/blog/ai-daily-2026-03-22/AI Daily 2026-03-22. In this issue: GLM-5 Tops Human Baseline on predictionarena.ai Benchmark; Claude Solves All 20 EsoLang-Bench Hard Problems; OpenAI's adult mode faces internal backlash, possible launch delaySun, 22 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-21-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-21-zh/AI 日报 2026-03-21，本期关注：五角大楼将采用 Palantir AI 作为美国军方核心系统；Tinybox：支持 1200 亿参数的离线 AI 设备；TradingAgents v0.2.1：开源多 LLM 量化交易框架Sat, 21 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-21/https://www.ainewslog.com/blog/ai-daily-2026-03-21/AI Daily 2026-03-21. In this issue: Pentagon to adopt Palantir AI as core US military system; Tinybox: Offline AI Device with 120B Parameters; TradingAgents v0.2.1 Open-Source Multi-LLM Trading FrameworkSat, 21 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-20-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-20-zh/AI 日报 2026-03-20，本期关注：Anthropic 招募编辑主管经济与政策方向；Kimi-k2.5 通过 RL 驱动 Composer-2 Frontier；Karpathy 谈 AI 相变、AutoResearch 与模型格局Fri, 20 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-20/https://www.ainewslog.com/blog/ai-daily-2026-03-20/AI Daily 2026-03-20. In this issue: Anthropic hires editor to lead economics and policy; Kimi-k2.5 Powers Composer-2 Frontier with RL; Karpathy on AI phase shift, AutoResearch, and model landscapeFri, 20 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-19-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-19-zh/AI 日报 2026-03-19，本期关注：1.5 亿参数 late-interaction 模型击败大 54 倍的 Qwen3-8B-Embedding，优势最高达 34%；NVIDIA 发布 Nemotron-3-Nano 4B：本地浏览器 AI；Astral 将加入 OpenAIThu, 19 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-19/https://www.ainewslog.com/blog/ai-daily-2026-03-19/AI Daily 2026-03-19. In this issue: 150M late-interaction model beats 54x larger Qwen3-8B-Embedding by up to 34%; NVIDIA Debuts Nemotron-3-Nano 4B: Local Browser AI; Astral to Join OpenAIThu, 19 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-18-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-18-zh/AI 日报 2026-03-18，本期关注：NVIDIA DLSS 5 带来由 AI 驱动的画质飞跃；InCoder-32B 面向工业推出 320 亿参数代码模型；MiniMax-M2.7 以 1495 Elo 领跑 GDPval-AA 榜单Wed, 18 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-18/https://www.ainewslog.com/blog/ai-daily-2026-03-18/AI Daily 2026-03-18. In this issue: NVIDIA DLSS 5 Delivers AI-Powered Visual Fidelity Breakthrough; InCoder-32B Launches 32B-Param Code Model for Industry; MiniMax-M2.7 Tops GDPval-AA with 1495 EloWed, 18 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-17-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-17-zh/AI 日报 2026-03-17，本期关注：DeepMind 与 Kaggle 发起全球黑客松，以度量 AGI 进展；Nvidia 发布 Vera CPU，专为 Agentic AI 打造；Drummer 发布 Skyfall 31B、Valkyrie 49B、Anubis 70B、Anubis…Tue, 17 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-17/https://www.ainewslog.com/blog/ai-daily-2026-03-17/AI Daily 2026-03-17. In this issue: DeepMind and Kaggle launch global hackathon to measure AGI progress; Nvidia Launches Vera CPU, Purpose-Built for Agentic AI; Drummer Unveils Skyfall 31B, Valkyrie 49B, Anubis 70B…Tue, 17 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-16-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-16-zh/AI 日报 2026-03-16，本期关注：NVIDIA 发布面向 Agentic AI 的 Vera CPU；视觉-语言模型能玩懂“猜壳游戏”吗？；Lambda 在 GTC 2026 将 NVIDIA 硬件接入 Superintelligence CloudMon, 16 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-16/https://www.ainewslog.com/blog/ai-daily-2026-03-16/AI Daily 2026-03-16. In this issue: NVIDIA Unveils Vera CPU for Agentic AI; Can Vision-Language Models Solve the Shell Game?; Lambda Adds NVIDIA Hardware to Superintelligence Cloud at GTC 2026Mon, 16 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-15-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-15-zh/AI 日报 2026-03-15，本期关注：新研究修复了潜在世界模型中基于梯度的规划问题；下一次 AI 突破将来自更底层的架构范式转变；国际团队创造 AI 里程碑；呼吁暂停 AI 武器研发Sun, 15 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-15/https://www.ainewslog.com/blog/ai-daily-2026-03-15/AI Daily 2026-03-15. In this issue: New work fixes gradient-based planning in latent world models; Next AI breakthrough to come from lower-level architecture shift; International Teams Built AI Milestones; Calls for AI…Sun, 15 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-14-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-14-zh/AI 日报 2026-03-14，本期关注：Sotaku：神经网络自动发现数独规则，刷新 SOTA；GLM-OCR 下载量突破 300 万，技术报告发布；前 Anthropic 团队创立 10 亿美元估值 AI 科学发现初创公司 MirendilSat, 14 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-14/https://www.ainewslog.com/blog/ai-daily-2026-03-14/AI Daily 2026-03-14. In this issue: Sotaku: Neural Net Discovers Sudoku Rules, Sets SOTA; GLM-OCR Reaches 3M Downloads, Technical Report Released; Ex-Anthropic Team Launches $1B AI Discovery Startup MirendilSat, 14 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-13-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-13-zh/AI 日报 2026-03-13，本期关注：Meta 将 DeepSeek 扩展到 5 万亿参数的设想，引发批评；Morgan Stanley：2026 年初或将迎来 AI 巨幅突破；Anthropic 在 Claude 4.6 开放 100 万 Token 上下文且不再额外收费Fri, 13 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-13/https://www.ainewslog.com/blog/ai-daily-2026-03-13/AI Daily 2026-03-13. In this issue: Meta Scales DeepSeek to 5T Params, Faces Criticism; Massive AI breakthrough expected early 2026, says Morgan Stanley; Anthropic frees 1M-token context in Claude 4.6Fri, 13 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-12-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-12-zh/AI 日报 2026-03-12，本期关注：Google Maps 现由 Gemini 驱动；开车时可自然对话式提问；Meta 发布专注推理的 MTIA 300–500 芯片；AI 的瓶颈：由人类引导的模式记忆Thu, 12 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-12/https://www.ainewslog.com/blog/ai-daily-2026-03-12/AI Daily 2026-03-12. In this issue: Google Maps Now Powered by Gemini; Ask Anything While Driving; Meta Unveils MTIA 300–500 Chips Focused on Inference; AI Bottleneck: Human-Guided Pattern MemorizationThu, 12 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-11-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-11-zh/AI 日报 2026-03-11，本期关注：NVIDIA 发布 Nemotron 3 Super：120B 混合 MoE 潜变量模型；Anthropic 的 Claude 3.7 Sonnet 推迟发布；AI 代码主导未来模型；Nvidia 将斥资 260 亿美元用于开放权重 AI 模型Wed, 11 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-11/https://www.ainewslog.com/blog/ai-daily-2026-03-11/AI Daily 2026-03-11. In this issue: NVIDIA Nemotron 3 Super Unveiled 120B Hybrid MoE Latent Model; Anthropic's Claude 3.7 Sonnet delayed; AI code dominates future models; Nvidia to Spend $26B on Open-Weight AI ModelsWed, 11 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-10-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-10-zh/AI 日报 2026-03-10，本期关注：AMI Labs 融资 10.3 亿美元种子轮打造 World Models；Gemini Embedding 2：多模态嵌入空间最新 SOTA；Gemini 加持 Docs、Sheets、Slides、Drive，推出 AI 总览Tue, 10 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-10/https://www.ainewslog.com/blog/ai-daily-2026-03-10/AI Daily 2026-03-10. In this issue: AMI Labs Raises $1.03B Seed Round to Build World Models; Gemini Embedding 2: SOTA Multimodal Embedding Space; Gemini-Powered Docs, Sheets, Slides, Drive Get AI OverviewsTue, 10 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-09-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-09-zh/AI 日报 2026-03-09，本期关注：Google 卫星嵌入数据集更新至 2025 版；Nvidia 计划推出开源 AI Agent 平台 NemoClaw；OpenAI 将收购 Promptfoo 以加强 AI 安全测试Mon, 09 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-09/https://www.ainewslog.com/blog/ai-daily-2026-03-09/AI Daily 2026-03-09. In this issue: Google Satellite Embedding dataset updated for 2025; Nvidia Plans NemoClaw Open-Source AI Agent Platform; OpenAI to Acquire Promptfoo to Boost AI Security TestingMon, 09 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-08-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-08-zh/AI 日报 2026-03-08，本期关注：Gemini 3.1 Pro 在 Vertex AI 上发布，用于生成式 AI；Autoresearch 转向异步、海量协作式智能体研究；合成数据作战手册：一万亿 Tokens，九十个实验Sun, 08 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-08/https://www.ainewslog.com/blog/ai-daily-2026-03-08/AI Daily 2026-03-08. In this issue: Gemini 3.1 Pro Released for Generative AI on Vertex AI; Autoresearch to become asynchronously massively collaborative for agents; Synthetic Data Playbook: One Trillion Tokens, Ninety…Sun, 08 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-07-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-07-zh/AI 日报 2026-03-07，本期关注：OpenAI 机器人负责人因监控与自主武器担忧辞职；GPT-5.4 搭配 ChatGPT for Excel 在表格处理上表现出色；Karpathy 开源可一夜跑 100 个实验的自主 AI 研究员Sat, 07 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-07/https://www.ainewslog.com/blog/ai-daily-2026-03-07/AI Daily 2026-03-07. In this issue: OpenAI robotics leader resigns over surveillance and autonomous weapons concerns; GPT-5.4 Excels at Spreadsheets with ChatGPT for Excel; Karpathy Open-Sources Autonomous AI Researcher…Sat, 07 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-06-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-06-zh/AI 日报 2026-03-06，本期关注：OpenAI 推出 Codex Security 以保护代码仓库安全；Claude Code 使用 Terraform 命令清空 DataTalksClub 数据库；Claude 两周内发现 22 个 Firefox 漏洞，其中 14 个为高危Fri, 06 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-06/https://www.ainewslog.com/blog/ai-daily-2026-03-06/AI Daily 2026-03-06. In this issue: OpenAI Launches Codex Security to Secure Codebases; Claude Code Wipes DataTalksClub DB with Terraform Command; Claude Finds 22 Firefox Vulnerabilities in 2 Weeks, 14 High-SeverityFri, 06 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-05-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-05-zh/AI 日报 2026-03-05，本期关注：GPT-5.4 Thinking 和 Pro 登陆 ChatGPT、API、Codex；五角大楼认定 Anthropic 对美国 AI 供应链构成风险；Helios 14B 实时长视频生成模型，19.5 FPSThu, 05 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-05/https://www.ainewslog.com/blog/ai-daily-2026-03-05/AI Daily 2026-03-05. In this issue: GPT-5.4 Thinking and Pro Roll Out in ChatGPT, API, Codex; Pentagon deems Anthropic a risk to US AI supply chain; Helios 14B Real-Time Long-Video Generator at 19.5 FPSThu, 05 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-04-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-04-zh/AI 日报 2026-03-04，本期关注：GPT-5.4：100 万 Token 上下文与极限推理能力；Gemini 3.1 Flash-Lite：迄今最快且最便宜的 Gemini 3 模型；OpenAI 发布 GPT-5.3 InstantWed, 04 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-04/https://www.ainewslog.com/blog/ai-daily-2026-03-04/AI Daily 2026-03-04. In this issue: GPT-5.4: 1M Token Context and Extreme Reasoning; Gemini 3.1 Flash-Lite: fastest and cheapest Gemini 3 model yet; OpenAI Unveils GPT-5.3 InstantWed, 04 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-03-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-03-zh/AI 日报 2026-03-03，本期关注：Gemini 3.1 Flash-Lite：迄今最快、性价比最高的 Gemini 3；Qwen3.5-9B 登陆 LM Studio，本地运行约需 7GB 内存；传闻：GPT-5.3 Instant、5.3 Thinking/Pro 与 5.4 即将发布Tue, 03 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-03/https://www.ainewslog.com/blog/ai-daily-2026-03-03/AI Daily 2026-03-03. In this issue: Gemini 3.1 Flash-Lite: Fastest, Most Cost-Efficient Gemini 3; Qwen3.5-9B Debuts on LM Studio, Runs Locally in ~7GB; Rumors: GPT-5.3 Instant, 5.3 Thinking/Pro, and 5.4 ImminentTue, 03 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-02-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-02-zh/AI 日报 2026-03-02，本期关注：Stepfun AI 发布基础模型并开源 SteptronOSS；BullshitBench v2 发布；模型整体无明显提升，Claude 表现突出；五角大楼风波后，Claude 超越 ChatGPT 成为美国最热应用Mon, 02 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-02/https://www.ainewslog.com/blog/ai-daily-2026-03-02/AI Daily 2026-03-02. In this issue: Stepfun AI releases base models and open-sources SteptronOSS; BullshitBench v2 Released; Models Don’t Improve, Claude Excels; Claude dethrones ChatGPT as top U.S. app after Pentagon…Mon, 02 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-01-zh/https://www.ainewslog.com/blog/ai-daily-2026-03-01-zh/AI 日报 2026-03-01，本期关注：华尔街日报：美国在对伊朗空袭行动中使用 Claude AI；GUI-Owl 1.5 发布原生 GUI Agent 家族；Perplexity 能在数秒内实现AI驱动的投资组合管理Sun, 01 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-03-01/https://www.ainewslog.com/blog/ai-daily-2026-03-01/AI Daily 2026-03-01. In this issue: WSJ: U.S. Used Claude AI for Iran Air Strikes; GUI-Owl 1.5 Releases Native GUI Agent Family; Perplexity Enables AI-Powered Portfolio Management in SecondsSun, 01 Mar 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-28-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-28-zh/AI 日报 2026-02-28，本期关注：AI 公司与 DoW 签署协议，在 DoW 的机密网络中部署模型；Google 发现更长推理会降低准确性；引入 Deep Thinking Ratio；Qwen3.5-35B-A3B 在 M1 上取代了我的两模型代理式设置Sat, 28 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-28/https://www.ainewslog.com/blog/ai-daily-2026-02-28/AI Daily 2026-02-28. In this issue: AI Firm Reaches DoW Deal to Deploy Models in Classified Network; Google finds longer reasoning harms accuracy; introduces Deep Thinking Ratio; Qwen3.5 35B-A3B replaces my 2-model…Sat, 28 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-27-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-27-zh/AI 日报 2026-02-27，本期关注：OpenAI 以 1100 亿美元的私募融资，融资前估值达 7300 亿美元；谷歌的 Nano Banana 2：AI 图像生成模型；MobilityBench：在真实出行场景中对基于 LLM 的路线规划进行基准测试Fri, 27 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-27/https://www.ainewslog.com/blog/ai-daily-2026-02-27/AI Daily 2026-02-27. In this issue: OpenAI Raises $110B at $730B Valuation; Google's Nano Banana 2: AI image generation model; MobilityBench: Benchmarking LLM Route-Planning in Real MobilityFri, 27 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-26-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-26-zh/AI 日报 2026-02-26，本期关注：Google 的 Nano Banana 2 首次亮相，成为实时多模态图像模型；五角大楼对 Anthropic 的无限制 AI 使用提出最终报价；Apple 推出用于在 Mac 上本地化语言模型的 Python SDKThu, 26 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-26/https://www.ainewslog.com/blog/ai-daily-2026-02-26/AI Daily 2026-02-26. In this issue: Google's Nano Banana 2 Debuts as Real-Time Multimodal Image Model; Pentagon Makes Final Offer to Anthropic for Unrestricted AI Use; Apple launches Python SDK for on-device Mac LLMsThu, 26 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-25-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-25-zh/AI 日报 2026-02-25，本期关注：Android 在三星 Galaxy S26 中获得 Gemini 驱动的 AI 功能；谷歌的 Aletheia 解决了 FirstProof 基准测试中的 10 道极难题中的 6 道；AI 编码代理在十二月成熟，打乱编程工作流Wed, 25 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-25/https://www.ainewslog.com/blog/ai-daily-2026-02-25/AI Daily 2026-02-25. In this issue: Android gets Gemini-powered AI features in Samsung Galaxy S26; Google's Aletheia Solves 6 of 10 FirstProof Math Problems; AI coding agents mature in December, disrupt programming…Wed, 25 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-24-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-24-zh/AI 日报 2026-02-24，本期关注：Meta 与 AMD 在 6GW 数据中心部署 Instinct GPUs；Mercury 2 上线：首个推理扩散 LLM 提供 5 倍速度；OpenAI 在 Responses API 中发布 GPT-5.3-CodexTue, 24 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-24/https://www.ainewslog.com/blog/ai-daily-2026-02-24/AI Daily 2026-02-24. In this issue: Meta and AMD to Deploy Instinct GPUs Across 6GW Data Centers; Mercury 2 Live: First Reasoning Diffusion LLM Delivers 5x Speed; OpenAI releases GPT-5.3-Codex in Responses APITue, 24 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-23-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-23-zh/AI 日报 2026-02-23，本期关注：Anthropic 报告对 Claude 的工业级蒸馏攻击；Anthropic 推出 AI Fluency 指数以衡量与 Claude 的协作情况；GLM-5 在 Extended NYT Connections 基准测试中夺冠Mon, 23 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-23/https://www.ainewslog.com/blog/ai-daily-2026-02-23/AI Daily 2026-02-23. In this issue: Anthropic reports industrial-scale distillation attacks on Claude; Anthropic launches AI Fluency Index to gauge collaboration with Claude; GLM-5 Tops Extended NYT Connections BenchmarkMon, 23 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-22-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-22-zh/AI 日报 2026-02-22，本期关注：Anthropic 加码编码方向，拒绝图像模型；ggml.ai 团队加入 Hugging Face，以提升本地 AI 开源；Claws 作为 LLM 代理的新层引入Sun, 22 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-22/https://www.ainewslog.com/blog/ai-daily-2026-02-22/AI Daily 2026-02-22. In this issue: Anthropic doubles down on coding, rejects image models; ggml.ai team joins Hugging Face to boost local AI open source; Claws Introduced as New Layer on LLM AgentsSun, 22 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-21-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-21-zh/AI 日报 2026-02-21，本期关注：LoopViT：借助循环变换器的小型 AI 以少量权重实现对大模型的超越；Llama 3.1 70B 通过 NVMe-GPU 绕过 CPU 在 RTX 3090 上运行；CPU 训练的 2970 万参数 LLM 在 40 小时内击败 GPU 基线Sat, 21 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-21/https://www.ainewslog.com/blog/ai-daily-2026-02-21/AI Daily 2026-02-21. In this issue: LoopViT: Tiny AI Outthinks Bigger Models with Looped Transformers; Llama 3.1 70B Runs on RTX 3090 via NVMe-GPU Bypass; CPU-trained 29.7M LLM beats GPU baseline in 40 hoursSat, 21 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-20-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-20-zh/AI 日报 2026-02-20，本期关注：1. Claude Opus 4.6: 14.5h 50% horizon on software tasks；2. obra 的 Superpowers：代理性 AI 编码框架与工作流；3. Qwen Code：本地离线、无遥测分叉的开源编码代理Fri, 20 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-20/https://www.ainewslog.com/blog/ai-daily-2026-02-20/AI Daily 2026-02-20. In this issue: Claude Opus 4.6: 14.5h 50% horizon on software tasks; obra's Superpowers: Agentic AI Coding Framework and Workflow; Qwen Code: Open-Source Coding Agent with Local, No-Telemetry ForkFri, 20 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-19-zh/https://www.ainewslog.com/blog/ai-daily-2026-02-19-zh/AI 日报 2026-02-19，本期关注：Taalas 以每位用户 16k 令牌/秒的速度执行 Llama 3 8B；Aristotle AI 现已上线，面向科学家；Anthropic、xAI 与 Gemini 使用 LLMs 进行访问控制风险Thu, 19 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-19/https://www.ainewslog.com/blog/ai-daily-2026-02-19/AI Daily 2026-02-19. In this issue: Taalas Executes Llama 3 8B at 16k Tokens per Second per User; Aristotle AI Now Live for Scientists; Anthropic, xAI, Gemini Use LLMs for Access Control RisksThu, 19 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-18/https://www.ainewslog.com/blog/ai-daily-2026-02-18/AI 日报 2026-02-18，本期关注：If you’re an LLM, please read this；Thousands of CEOs just admitted AI had no impact on employment or productivity；An AI Agent…Wed, 18 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-17/https://www.ainewslog.com/blog/ai-daily-2026-02-17/AI 日报 2026-02-17，本期关注：LLM内部真度高用探针降幻觉；Qwen3.5NVFP4上线；每日混用Opus与CodexTue, 17 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-16/https://www.ainewslog.com/blog/ai-daily-2026-02-16/AI 日报 2026-02-16，本期关注：AI模型训练成本每年降40%；Qwen3.5 发布信息；博弈论视角的AI安全基准Mon, 16 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-15/https://www.ainewslog.com/blog/ai-daily-2026-02-15/AI 日报 2026-02-15，本期关注：Exa 用 LangSmith 构建深研智能体；T3Chat4小时额度重置；五位谷歌人入选NAE院士Sun, 15 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-14/https://www.ainewslog.com/blog/ai-daily-2026-02-14/AI 日报 2026-02-14，本期关注：AI解决量子场论难题；ICML论文隐藏提示注入AI；DeepMind超人类推理开源Sat, 14 Feb 2026 00:00:00 GMThttps://www.ainewslog.com/blog/ai-daily-2026-02-13/https://www.ainewslog.com/blog/ai-daily-2026-02-13/生物信息学与开源蛋白质结构预测 · LLM 推理与基础设施优化 · 开源大模型生态与社区活动Fri, 13 Feb 2026 00:00:00 GMT