2026-08-17 科技日报

扫描 514 篇候选内容 · 覆盖 650 个信息源 · 纳入 505 篇

📌 今日要点

• Nvidia investing $1.5B in SoftBank data center developer behind OpenAI project
• Wispr raises $280M at$2B valuation as it looks beyond dictation
• Does a Language Server Save Tokens for Coding Agents? A Measurement Methodology and Preliminary Study
• Think in Latent, Explain in Language: Self-Explainable Latent Reasoning
• Not All Tokens Are Equal: Inflation-Aware Routing for Agentic LLM Systems
• BCMT: Blockwise Causal Memory Transformer

📝 科技简讯

• Nvidia investing $1.5B in SoftBank data center developer behind OpenAI project — TechCrunch AI
• Wispr raises $280M at$2B valuation as it looks beyond dictation — TechCrunch AI
• Does a Language Server Save Tokens for Coding Agents? A Measurement Methodology and Preliminary Study — ArXiv CL (cs.CL)
• Think in Latent, Explain in Language: Self-Explainable Latent Reasoning — ArXiv CL (cs.CL)
• Not All Tokens Are Equal: Inflation-Aware Routing for Agentic LLM Systems — ArXiv CL (cs.CL)
• BCMT: Blockwise Causal Memory Transformer — ArXiv CL (cs.CL)
• Jais 2: A Family of Arabic-Centric Open Large Language Models — ArXiv CL (cs.CL)
• What to Preserve, Where to Adapt: A Depth-Wise Analysis of Forgetting in Continual Gynecological Image Segmentation — ArXiv CV (cs.CV)
• Multiphase-Diff: Diffusion-Based Generative Modeling for High-Contrast Multiphase Physical Systems with Sharp Interfaces — ArXiv CV (cs.CV)
• PROVE: Training-Free Prompt Recovery using Verifiable Evidence — ArXiv CV (cs.CV)
• MedPlex: Deep Vision-Language Co-Adaptation for Clinically Grounded Medical Segmentation — ArXiv CV (cs.CV)
• Limitations of Synthetic Data Generation in Specialized Data-Scarce Domains — ArXiv CV (cs.CV)
• You Can’t Iterate to Trustworthy AI Code Without Understanding — LessWrong
• Do your readers get annoyed if you post every day for 30 days? — LessWrong
• What 27 AI Safety Generalists Are Building This Summer — LessWrong
• What gives you away: how LLMs form opinions of you — LessWrong
• Evaluating Chain-of-Thought Monitorability is Still an Open Problem: Comments on OpenAI’s Monitorability Evals — LessWrong
• Adjacency-Based Spectral Proxy Control of Mobile Communication Agents — ArXiv RO (cs.RO)
• hint$^2$: Hierarchical World Models for Inference-Time Temporal Logic Guidance — ArXiv RO (cs.RO)
• Graph-MambaNav: Spatial-Temporal Graph Mamba Leveraging Object-Relation Knowledge for Object-Goal Navigation — ArXiv RO (cs.RO)
• Knowledge-Data-Dual-Driven Reinforcement Learning for Autonomous Vehicle Control in Mixed Traffic — ArXiv RO (cs.RO)
• Ontology-Grounded World Models for Failure Diagnosis and Closed-Loop Repair in Physical AI Systems — ArXiv RO (cs.RO)
• AirTag reveals how Amazon destroys rare books for AI training — The Decoder
• OpenAI signs record Ohio data center lease with Nvidia backing up to $105 billion — The Decoder
• AI video market has bounced back from Sora’s false start — The Decoder
• Same Cluster, 33 Points More Utilization: What Changed Was the Order — Hugging Face Blog
• L-FNO: Lorentzian Fourier Neural Operator for Stochastic Event Dynamics — ArXiv ML (cs.LG)
• Don’t Claim Benchmark-Oriented Optimization Improves General Coding Capability — Diverse Evaluation Is Required — ArXiv ML (cs.LG)
• Robust XGBoosting for Regression — ArXiv ML (cs.LG)
• Training-Free Knowledge Transfer Across Model Scales through Activation-Guided Pruning — ArXiv ML (cs.LG)
• Hard Cases, Bad Labels: Testing Error Exposure and Error Location in Uncertainty Sampling Under Bounded Label Noise — ArXiv ML (cs.LG)
• Robust Dual-Model Collaborative Random Vector Functional Link Network — ArXiv ML (cs.LG)
• Contrastive Learning for Interpretable Anomaly Detection at Collider Experiments — ArXiv ML (cs.LG)
• What Can I Actually Do with a Small Language Model? — KDnuggets
• 5 Python Libraries That Make Data Cleaning More Enjoyable — KDnuggets
• NVIDIA Nemotron 3.5 Lightning now available in Amazon SageMaker JumpStart — AWS Machine Learning
• Build OpenClaw agents that transact with Amazon Bedrock AgentCore payments — AWS Machine Learning
• Get closer to the game with Gemini and Pixel — Google AI Blog
• Teaching Everyone to Fish for Tokens — Interconnects (Nathan Lambert)
• 从 Claude 隐形水印回望 GenAI 时代信息隐藏：嵌入水印如何保证无损？ — 新智元

📊 数据概览