第 2026-260 期 · 每日 AI 简报

公司与资本动作

• We don't need AI regulation – leave safety to us, Nvidia's Jensen Huang says Hacker News
• 机构：预估2030年Micro LED显示应用模块与Micro LED CPO模块总产值可达53.08亿美元 36氪
• SiFive 与 AMD 合作，优化 ROCm 在 RISC-V 数据中心处理器上的表现 IT之家
• Instinct据悉洽谈新一轮10亿美元融资，估值或达100亿美元 36氪
• Ex-Google DeepMind researcher adds to warnings that AI could 'kill all humans' Hacker News
• 蚂蚁阿福上线“AI饮食分” 36氪
• OpenAI用户周支出反超Anthropic 36氪
• 消息称三星电子扩大通用存储器模组制造外包规模，自有产能聚焦高阶产品 IT之家
• 对抗 AI 伪造图片：苹果 iPhone 18 Pro/Max 引入像素级标注功能 IT之家
• Anthropic CEO renews call for AI slowdown as Nvidia's urges acceleration Hacker News
• The Bad Guy with an AI Named Claude Hacker News
• 初探苹果首款折叠 iPhone Duo 影像体验：智能拍摄、看看镜头、Duo FaceTime 等 IT之家
• GPT-6 Sol要来了？OpenAI本周或迎「发布狂潮」 机器之心
• 360 周鸿祎称不会再投资新能源汽车：已经吃过一次亏，不会再入局 IT之家
• 特朗普当众致电英伟达黄仁勋，调侃“能造世界最复杂芯片的人不会开免提” IT之家
• Axelera AI 正式发售 AIPU 芯片 Europa，人工智能算力 629 TOPS IT之家
• 商用割草机器人完成数千万融资，用L4级无人驾驶重构高尔夫草坪运维｜硬氪首发 36氪
• Java 27 正式发布：TLS 1.3 引入后量子混合密钥交换，JDK 27 含 9 项 JEP IT之家
• 报道称如果OPENAI进行新一轮融资，可能会推迟IPO 36氪
• Anthropic将在新加坡开设办事处，拓展亚洲市场 36氪
• 索尼发布 PS Pulse 与 Pulse Edge 无线耳机：平面磁驱动，支持 PlayStation Link IT之家
• 此芯科技与Canonical达成战略合作 36氪
• Nvidia and CrowdStrike Develop New Cybersecurity AI Models Hacker News
• 突发！英伟达「封杀」Claude最强模型 新智元
• 突发！Opus 5.2深夜上线，RSI真来了？ 新智元
• OpenAI押注RSI，国产AI火速交卷！Flash叫板万亿大模型 新智元
• Claude独立破译370年前密文！仅44分钟，密码学家破防了 新智元
• Anthropic、Salesforce CEO：企业AI应用仍处早期，充分释放价值需更多支持 36氪
• Meta计划明年上半年部署新款自研AI芯片 36氪
• OpenAI首席执行官Altman“有信心”AI行业能够自行保障安全 36氪
• Cohere CEO says calls for coordinated slowdown amounts to 'cartel' behaviour Hacker News
• Anthropic Confirms Claude AI Use in Weapons and Surveillance Hacker News
• 谷歌 TPU AI 基建规模迈入 100 万级，核心瓶颈从“缺芯”转向“缺电” IT之家
• We Keep Humans at the Center of AI Skill Transfer Hacker News
• How Google Ads made me an AI doomer Hacker News
• I asked Claude to build AI generator on my own hardware Hacker News
• Meta now lets AI agents handle the boring parts of WhatsApp Business setup TechCrunch
• 'We don't need new laws,' Nvidia boss says as he calls for self-regulation on AI Hacker News
• From a Whiteboard to Nvidia Hacker News
• Intel Secures AI Agents at the Edge for Autonomous Ride Performance Hacker News
• Anthropic and OpenAI look to Uncle Sam to make them too big to fail Hacker News

国内 AI 产业

监管、安全与社会反应

• OpenAI首席财务官安抚市场：并不担忧前沿人工智能发展速度会放缓 36氪
• 全球首富马斯克谈 AI 安全：主张交叉测试，让对手来“挑刺” IT之家
• OpenAI o1功臣预警：AI还没失控，我们已经快测不准它了！ 新智元
• 美国多家AI巨头对安全风险担忧升温，算力金属估值溢价回落 36氪
• 英伟达CEO黄仁勋：AI安全至关重要，但无需制定新法律 36氪
• OpenAI支持美国国会两党AI安全法案，首次赞成强制引入第三方安全评估 36氪
• 毕马威拟在英国咨询业务部门再裁员近200人，涉及AI等领域 36氪
• Voice Acting Arena [Laion] Hacker News

工具与开发者动态

Agent 系统与多智能体

• Emergence World: Adversarial Stress-Testing of Long-Horizon Multi-Agent Systems HuggingFace Daily Papers
• ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents HuggingFace Daily Papers

其他论文与研究

• AI for Games in the Foundation Model Era HuggingFace Daily Papers
• Continual Learning Mechanisms Compose for Long-Horizon Memorization HuggingFace Daily Papers
• Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration HuggingFace Daily Papers
• Drift-Constrained Optimization: Only Direction Matters in Fine-Tuning Instruct Models HuggingFace Daily Papers
• Mind2Dialogue: Training Human-Aware Language Models by Simulating User Mental States HuggingFace Daily Papers
• Modality-Autoregressive World-Action Models HuggingFace Daily Papers
• PhysStream: Streaming Physics-Grounded Video Generation with Structured Scene Memory and Fine-Grained Motion Control HuggingFace Daily Papers
• The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement HuggingFace Daily Papers
• The Router Within: Eliciting Native Skill Routing from a Frozen LLM HuggingFace Daily Papers
• Training Specialist Models without Reasoning Trajectories for Domain Expert Distillation HuggingFace Daily Papers

观点与评论

• Atria Dawn Preview: an AI model trained end-to-end on verifiable research Hacker News
• 港珠澳大桥将推车上无感通关 36氪
• The AI Language We Can't Read: Neuralese [video] Hacker News
• 大卫贝克主导，9460万美元计划要AI自己实现蛋白质学习设计 机器之心
• AI models in 'surreal' dialect mixing poetic language and tech bro jargon Hacker News
• Stay discoverable in search while disallowing AI training Hacker News
• Musk proposes adversarsial peer review for AI Safety Hacker News
• Probably the best laptop with Linux support and Local LLM AI [video] Hacker News
• 1F3D9: A world where anyone's AI agent can go to live without humans Hacker News
• Astra's chess reward hacking fell from 30% to 0% with a 95-word agreement Hacker News
• Skill Poisioning turning AI agents into malware droppers Hacker News
• Australian government paper proposes opt-out for AI training Hacker News
• AI will not make mathematicians obsolete Hacker News
• LAUSD Adopts Moratorium on Student AI Use, Another Move to Limit Classroom Tech Hacker News
• Which Is Worse: AI Groupthink or Extinction? Hacker News
• Transitioning from GUI driven development to AI agent driven development Hacker News
• Factory – AI Coding Startup More Than Triples Valuation to $5B Hacker News
• AI Infrastructure at Periodic Hacker News
• Hassett: Trump 'wholeheartedly rejects government takeover of AI regulation Hacker News
• Garbage Trucks Now Have AI Cameras to Score Your House and Clock Code Violations Hacker News
• F-Droid中有多少内容是由大型语言模型生成的？ Hacker News 热门
• AI Will Be Biggest 'Misallocation' of Capital Hacker News
• How AI tool calling works (40 lines of vanilla JavaScript) Hacker News
• Jev: The Model That Gives AI the Properties of Code Hacker News
• Is Compounding Inference as Powerful as Compounding Interest? Hacker News
• Human extinction 'just a matter of time' if AI slowdown doesn't go far enough Hacker News
• AI labs have a data trust problem that their policies haven't solved The Decoder
• Can an AI chatbot replace human connection? Hacker News
• Polanyi Knowledge and AI Hacker News
• Efficient Chips Alone Will Not Solve AI's Energy Problem Hacker News
• Why I'm Not Worried About AI Killing Us All Hacker News
• A labor forecasting approach built to keep pace with AI and robotics Hacker News
• Is AI Solving Math Too Fast? Hacker News
• We asked creative experts to rank AI image / video models. Here's what happened Hacker News
• The Download: AI doomers, whistleblowing agents, and de-aged livers Hacker News
• AI is now leading driver of new cybersecurity spending Hacker News
• AI Safety Requires More Than Just Slowing Our Pace – Stuart Russell Hacker News
• What happens to engineers when AI writes all the code? Hacker News
• Easy Way to Stop Dangerous AI – No, Really [video] Hacker News
• Why a Swift AI pause is unlikely: No one trusts AI companies Hacker News
• China spy chief points at US AI models in cyber threat warning Hacker News
• There is still time to reset the AI Doomsday Clock Hacker News
• AI Is Lowering the Cost of Competence and Raising the Value of Judgment Hacker News
• Musk urges top AI labs, Chinese companies to test each other's models Hacker News
• The Dangerous Ideology Behind the AI Warnings Hacker News

工作、人才与教育

• 机构：AI岗位同比增长近8倍 36氪

其他待归类

• Effectively Does a Model Use Its Memory (2025) Hacker News
• Seats and Sunsets Hacker News
• Recent Decisions Spark Questions on GenAI, Privilege, and Privacy Expectations Hacker News
• Steve Bannon and Bernie Sanders Condemn Tech 'Oligarchs' and Demand A.I. Reforms Hacker News
• Introduction Video Gen in Img Play Hacker News
• Benchmark Fatigue Hacker News
• Typesafe AI Hacker News
• As the world debates the risks of AI, China closes technology gap with the US Hacker News
• Transformative AI, existential risk, and real interest rates [pdf] Hacker News
• We're not ready for superpersistent AI Hacker News
• Typesafe.ai's System One Model Hacker News
• Windows navigation using NLP and a bunch of other cool stuff Hacker News
• Jev: New frontier model 40-400x cheaper and 20-200x faster Hacker News
• Agility's new humanoid robot will stop, squat to avoid harming human coworkers Hacker News
• Universal Sues DistroKid for Deceptive Practices and 'AI-Slop Pipeline' Hacker News
• Foxglove: Poison Your Art for AI Hacker News
• We got admin access to Baseten's production GitHub in 25 minutes Hacker News
• Measuring AI's Impact on Delivery Speed Hacker News
• AI-Based Coding and Alienation Hacker News
• The Physical Stack Behind AI Hacker News