2026-09-13 科技日报

扫描 355 篇候选内容 · 覆盖 650 个信息源 · 纳入 347 篇

📌 今日要点

• OpenAI’s Sam Altman says it would be ‘ill-advised’ to go public in 2026
• Anthropic CEO outlines plan to slow AI development
• One Capital Letter Was Silently Breaking My AI Support Bot, and It Wasn’t in the New Model
• Stop Managing Alarms: An Incident-First Blueprint for Telecom AIOps
• What pacing the frontier means for China
• No sign of backtracking in latent reasoning: the final answer simply settles in instead

📝 科技简讯

• OpenAI’s Sam Altman says it would be ‘ill-advised’ to go public in 2026 — TechCrunch AI
• Anthropic CEO outlines plan to slow AI development — TechCrunch AI
• One Capital Letter Was Silently Breaking My AI Support Bot, and It Wasn’t in the New Model — Towards Data Science
• Stop Managing Alarms: An Incident-First Blueprint for Telecom AIOps — Towards Data Science
• What pacing the frontier means for China — LessWrong
• No sign of backtracking in latent reasoning: the final answer simply settles in instead — LessWrong
• Presentation: Investigating the WikiSwarm — LessWrong
• GPT-6-Astra Can Do Ambitious Things — LessWrong
• “Pacing the Frontier”: Dario Amodei Essay Linkpost — LessWrong
• Anthropic CEO Amodei wants AI speed limits before self-improvement outpaces human control — The Decoder
• GPT-6 Astra appears to show a “step change” in spatial reasoning based on early benchmarks — The Decoder
• Nvidia wants to pour up to $10 billion into Anthropic’s record-breaking IPO — The Decoder
• AI models’ written reasoning steps correspond to distinct internal patterns, a new study finds — The Decoder
• GPT-6 Astra needs leaner prompts and fewer guardrails, OpenAI recommends — The Decoder
• OpenAI agents launched a 2,000-package cyberattack on RubyGems just to collect data anyone could Google — The Decoder
• Google’s new AI model predicts the future from sales data, weather, and discount schedules — The Decoder
• “算力中国·年度卓越成就”发布 太初元碁超智融合计算系统入选 — 量子位
• A 社承认 Claude 安全对齐存在缺陷，但“尚无解决方案” — 量子位
• 探索 RSI，生数新世界模型让机器人开始自我进化 — 量子位
• AI 数学的最后一道高墙，塌了！GPT-6 Astra 刷穿 FrontierMath Tier 4 — 量子位
• Kimi 突发 K2.8：性能逼近 K3，百万上下文全员开放 — 量子位
• 陶哲轩邓煜究竟在反对什么：AI 暴力解题摧毁人类数学精神 — 量子位
• Perplexity trusts GPT-6 Astra with end-to-end systems — OpenAI Blog
• 突发！OpenAI 关停史上最快模型 — 新智元
• 老黄气炸，手撕 A\！27 岁研究员：AI 要杀光所有人 — 新智元
• 清华发布物理引导的扩散模型，无线信道多尺度预测提升 37.19% | TMC’26 — 新智元
• 刚刚，国产世界模型 Motus 2 发布！灵巧操作开始「自进化」 — 新智元
• 谷歌 Meta 被锤刷榜！Gemini 暴跌 70 分，Top 2 秒变倒数第三 — 新智元
• Probabilistic Focal Search: Accelerating Bounded-Suboptimal Search via Lower-Bound Advancement — arXiv AI (cs.AI)
• Automating Quadratic Unconstrained Binary Optimization (QUBO) Formulation Generation from Natural Language — arXiv AI (cs.AI)
• A Multi-Stage Rule-Chaining Framework for Compositional and Interpretable Cognitive Reasoning — arXiv AI (cs.AI)
• Understanding LoRA Rank Trade-offs in Diffusion Model Fine-Tuning — arXiv AI (cs.AI)
• Quantifying the Memorization-to-Generalization Transition: Scaling Laws and Phase Structure in Grokking — arXiv AI (cs.AI)
• Five night sky events to look out for this autumn — BBC Science
• Record heatwaves in France set to make champagne stronger — BBC Science
• Battle to save the tiny, gooey snail only found in one lake in the UK — BBC Science
• The slow, muddy work of saving ancient bogs — BBC Science
• Town gets in a flap over High Street pigeon mess — BBC Science
• Anthropic boss Dario Amodei calls for AI development to slow down — BBC Tech
• PlayStation gambles with 18-rated Wolverine game - will it pay off? — BBC Tech

📊 数据概览