2026-09-02 科技日报

扫描 567 篇候选内容 · 覆盖 650 个信息源 · 纳入 556 篇

📌 今日要点

• Introducing agentic video understanding with Gemini
• BenchMIRT: What are LLM benchmarks actually measuring?
• Introducing @huggingface/kernels: 200+ WebGPU Kernels for Local AI
• Free Transcription with Speakr
• 7 Common Python Mistakes to Avoid in AI Workflows
• The latest AI news we announced in August 2026

📝 科技简讯

• Introducing agentic video understanding with Gemini — Google DeepMind
• BenchMIRT: What are LLM benchmarks actually measuring? — Hugging Face Blog
• Introducing @huggingface/kernels: 200+ WebGPU Kernels for Local AI — Hugging Face Blog
• Free Transcription with Speakr — KDnuggets
• 7 Common Python Mistakes to Avoid in AI Workflows — KDnuggets
• The latest AI news we announced in August 2026 — Google AI Blog
• Try Google Pics: Easy image creation and editing in Google Workspace — Google AI Blog
• Whistle Synth Mac App — LessWrong
• “Collusion” is just cooperation that you don’t like — LessWrong
• What happened to impact markets? — LessWrong
• A nightwatchman on every probe: Superintelligent surveillance to prevent galactic anarchy — LessWrong
• No Bush, No Potato — LessWrong
• 刚刚，A 社自曝「最坏 Claude」！150 人紧急转岗，新品开发全停 — 新智元
• Runway 把代码干掉了！首个界面世界模型，UI 自己长出来 — 新智元
• 奥特曼播客曝猛料：Astra 操作电脑已达人类水平！ — 新智元
• 奥特曼直言：四年太久，大学两年就够！ — 新智元
• DeepMind AlphaEvolve 再破纪录：矩阵乘法指数 40 年最低！ — 新智元
• Introducing Claude Fable 5.1 on AWS — AWS Machine Learning
• From theory to delivery: How Atos upskilled 400 engineers in agentic AI — AWS Machine Learning
• Tokenomics at scale: How Jamf built real-time spend enforcement for Amazon Bedrock — AWS Machine Learning
• Securing Amazon Quick from POC to production: Agents, Flows, and Spaces — AWS Machine Learning
• How t54 built a trust layer with Amazon Bedrock AgentCore payments — AWS Machine Learning
• How ZS democratized secure ad-hoc analytics with Amazon SageMaker — AWS Machine Learning
• How Boomi Scribe streamlines documentation using AWS — AWS Machine Learning
• ERR+: Sequential Entropy Resolution for Efficient and Decisive LLM Reasoning — ArXiv ML (cs.LG)
• Unsupervised Latent Space Alignment with Hyperspherical Geodesic Matching — ArXiv ML (cs.LG)
• Curvature Cryptanalysis of Smooth Transformer Feed-Forward Networks — ArXiv ML (cs.LG)
• Equivariant Sheaf Neural Networks: Learning Geometric Transport on Graphs — ArXiv ML (cs.LG)
• The Halt Vector: Internalizing a Causal Steering Intervention for Efficient Reasoning — ArXiv ML (cs.LG)
• Conservative Hybrid Graph Networks for Process Systems with Learned Routing — ArXiv ML (cs.LG)
• Off-Policy Evaluation for Semantic ID Recommenders: Does the Model’s Own Code Hierarchy Help? — ArXiv ML (cs.LG)
• Training a Misaligned Reward Seeker — AI Alignment Forum
• AfterQuery reportedly becomes Y Combinator’s fastest-ever unicorn, now valued at $3.2B — TechCrunch AI
• OpenAI’s Astra model is on the way — and very good at breaking into computer systems — TechCrunch AI
• Google’s Android update tackles motion sickness, accessibility, and more — TechCrunch AI
• Anthropic’s new Fable release is cheaper, less restrictive — TechCrunch AI
• Google’s answer to Canva is an AI tool where you prompt instead of design — TechCrunch AI
• ChatGPT Health adds Epic integration for clinicians to import patient data — TechCrunch AI
• Sequoia-incubated Empirik launches with $21M to predict outages before they happen — TechCrunch AI
• Your JSON Is Valid but Your Data Is Wrong: Five Failure Modes LLM Structured Outputs Won’t Catch — Towards Data Science

📊 数据概览