2026-09-09 科技日报

扫描 524 篇候选内容 · 覆盖 650 个信息源 · 纳入 505 篇

📌 今日要点

• Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic
• AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome
• How good are slop-vestigators?
• Allow Carriers on Planes
• Is there an acceptable way to store clothes?
• FLOP Around and Find Out: LLM Training Workload Size Estimation With Power Monitoring

📝 科技简讯

• Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic — Hugging Face Blog
• AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome — Google DeepMind
• How good are slop-vestigators? — LessWrong
• Allow Carriers on Planes — LessWrong
• Is there an acceptable way to store clothes? — LessWrong
• FLOP Around and Find Out: LLM Training Workload Size Estimation With Power Monitoring — LessWrong
• A Conceptual Framework for Reasoning about Exploration Hacking — LessWrong
• From RAG to Agentic AI: Building the Next Generation of Intelligent Enterprise Systems — KDnuggets
• Is ArrowJS Really the UI for the Agentic Era? Here’s What I Found — KDnuggets
• 5 Ways I Access Coding Models for Free — KDnuggets
• Latest open artifacts (#24): Motif-3, GLM-5.3, Hy4-preview and open model licenses — Interconnects (Nathan Lambert)
• Hackers are stealing Claude tokens from subscribers — TechCrunch AI
• Cognition hits $48B valuation, signaling investors believe AI coding is far from a winner-take-all market — TechCrunch AI
• Meta debuts its Muse AI agent. Will consumers trust it? — TechCrunch AI
• OpenAI fought dirty on career-making math problem, says NYU mathematician — TechCrunch AI
• Google Cloud races to catch up in the AI deployment wars with Accenture deal — TechCrunch AI
• Chrome is now shipping updates every 2 weeks as AI changes the security landscape — TechCrunch AI
• Mistral raises €3B as sovereign AI becomes big business — TechCrunch AI
• How GPT-5.6 Sol helps run quantum computing experiments — OpenAI Blog
• The Work Now Within Reach — OpenAI Blog
• Introducing ChatGPT Images 2.5 — OpenAI Blog
• On the Navier–Stokes Millennium Prize Problem — OpenAI Blog
• Funding grants for new research into AI and teen development — OpenAI Blog
• Training on probes: Research ideas — AI Alignment Forum
• Training on probes: What’s going on — AI Alignment Forum
• 北大博士团队杀回来了，要做「游戏界的红果」！AI 造梦片场永不加赋 — 新智元
• 突发，OpenAI 下一代 Bel 爆出攻克千禧年难题！ — 新智元
• 刚刚，国产 2B 小钢炮开源！逼近人类水平，跑出端侧通用 Agent 雏形 — 新智元
• 他让 GPT-6 管 GPT-6，一周搭出曼哈顿！ — 新智元
• 系统综述 259 项工作：AI 创作如何从「会生成」走到「能交付」？ — 新智元
• OpenAI researcher allegedly pressured mathematician to drop Anthropic co-author from math breakthrough paper — The Decoder
• Patagonia has what AI data centers want, including no resistance so far — The Decoder
• ASML locks in TSMC, Samsung, and Intel while Huawei races to break its grip — The Decoder
• Meta drops AI usage from engineer performance reviews after “tokenmaxxing” backfires — The Decoder
• Mistral AI raises 3 billion euros in Europe’s largest-ever tech funding round despite lagging behind rivals — The Decoder
• How to Maximize GPT-6 Astra — Towards Data Science
• The Model Validation Playbook for GenAI: Lessons from Banking — Towards Data Science
• A Beginner’s Guide to World Models — Towards Data Science
• Introducing ShipAI — Towards Data Science
• Context Windows Don’t Know What’s Still True — I Built a Validity Layer That Does — Towards Data Science

📊 数据概览