2026-09-22 科技日报

扫描 556 篇候选内容 · 覆盖 650 个信息源 · 纳入 540 篇

📌 今日要点

• SoftBank to borrow over $11 billion in risky bonds for OpenAI stake
• Bristol researchers say medicine already knows how to handle black boxes and AI could learn from it
• US and China agree on AI dialogue with security mechanism ahead of Trump-Xi summit
• Amazon blocks Meta’s AI agent Muse from online shopping
• WM-VS: Progress-Aligned World Models for Closed-Loop Visual Servoing
• AeRove: A Compact Bimodal Aerial-Terrestrial Drone with Rapid Bistable Reconfiguration for Close-Range Pipeline Inspection

📝 科技简讯

• SoftBank to borrow over $11 billion in risky bonds for OpenAI stake — The Decoder
• Bristol researchers say medicine already knows how to handle black boxes and AI could learn from it — The Decoder
• US and China agree on AI dialogue with security mechanism ahead of Trump-Xi summit — The Decoder
• Amazon blocks Meta’s AI agent Muse from online shopping — The Decoder
• WM-VS: Progress-Aligned World Models for Closed-Loop Visual Servoing — ArXiv RO (cs.RO)
• AeRove: A Compact Bimodal Aerial-Terrestrial Drone with Rapid Bistable Reconfiguration for Close-Range Pipeline Inspection — ArXiv RO (cs.RO)
• Shake to Learn: Dynamic Interrogation of Hidden Object Physics for Robotic Manipulation with Physical Reservoir Computing — ArXiv RO (cs.RO)
• ForeTac-VLA: A Forecasting-Based Tactile-Vision-Language-Action Model for Contact-Rich Robotic Manipulation — ArXiv RO (cs.RO)
• PIVOT: Physically Informed Vision-Language Off-Road Traversability for Field Robot Navigation — ArXiv RO (cs.RO)
• TAPe+ML: A Compact Structured Representation for Multi-Task Computer Vision — ArXiv CV (cs.CV)
• MemeTAG: Keyword-Driven Meme Classification through Tag Embedding Reconstruction — ArXiv CV (cs.CV)
• Image-Derived PM10 Estimation in Cattle Feedlot Using Machine Learning: Addressing Concentration Ranges Beyond Existing Digital Imaging Methods — ArXiv CV (cs.CV)
• Fragment-Aware Vision Transformers for Fresco-Fragment Style Classification — ArXiv CV (cs.CV)
• MAGIC: Marginal-Guided Compression with Optimal Transport for Efficient Visual Document Retrieval — ArXiv CV (cs.CV)
• RBS-Attention: Radius-Bounded Sparse Prefill for Long-Context Large Language Models — arXiv AI (cs.AI)
• Attention-Aware Routing: Coupling Routing and Attention in MoEs — arXiv AI (cs.AI)
• CaLR: Causal Latent Revision for Robust Diffusion Reasoning — arXiv AI (cs.AI)
• LoRA Enhanced Contrastive Learning with SAS Vision Transformers — arXiv AI (cs.AI)
• Detecting Hallucination in LLMs: Tracing the Topological Signatures of Impaired Context Sharing — arXiv AI (cs.AI)
• Agent Optimizer: A Faster Path to Better Outcomes — Salesforce AI Blog
• Dreamforce Takeaways: Small Teams, Big Summit — Even Bigger Future — Salesforce AI Blog
• SkyProduction 天工工作台中秋国庆特惠第二波：Seedance 2.5 720P 低至 0.27 元/秒，全网地板价！ — 量子位
• 啊啊啊 GPT-6 Astra 这么不安全！这次马斯克都瘫坐了 — 量子位
• 长三角安全人工智能安徽省实验室发布星界、星驭、星鉴三大 AI 安全解决方案 — 量子位
• 国产数据库跑出 AI 新能力！OceanBase 登顶国际 Data Agent 榜单 — 量子位
• Building standards for the next phase of AI — OpenAI Blog
• Expanding OpenAI Academy with new learning paths — OpenAI Blog
• An opinionated primer on Causal Decision Theory (CDT) and Evidential Decision Theory (EDT) — LessWrong
• Acausal interactions: What they are, why they matter, and what to do about them — LessWrong
• Story Imprinting: AI Assistants Absorb Traits from Human Characters They Resemble — LessWrong
• Why advance automated scientific labs rather than slow them? — LessWrong
• What if not Circuits? — LessWrong
• Improving synthesis prediction of small molecules at scale with RetroChimera — Microsoft Research
• A New Kind of Model for AI Decision-Making? — Towards Data Science
• GPT-6 Astra Just Hit OpenAI’s Highest Cybersecurity Risk Level — Towards Data Science
• Google Offered $10M for a Dying Airline’s Data. How Can You Value Yours? — Towards Data Science
• Your AI Assistant Wrote the Code. Who Checked the Defaults? — Towards Data Science
• GPT-6 真敢杀人啊！推下天台，挥刀扎人不眨眼 — 新智元
• OpenAI 研究员惊人言论：物理隔离无法阻止失控 AI！ — 新智元
• 工具、技能、记忆连成网，北大等提出自进化程序图 PG，越用越聪明 — 新智元

📊 数据概览