2026-07-24 科技日报

扫描 576 篇候选内容 · 覆盖 650 个信息源 · 纳入 565 篇

📌 今日要点

• 小红书笔记，长出「App」了！16 万人手搓，00 后打头阵
• 马斯克第三次喊「奇点已至」！AI 三个月连破数学、安全两道墙
• AI PPT 一改就崩？MemSlides 登顶抱抱脸，让 Agent 记住你的改稿习惯
• Pulling the Fire Alarm
• Contra George Hotz on “AI 2040 and the Cult of Intelligence”
• The Model Organism Lottery: Model Organism Interpretability Strongly Depends on Training Methodology

📝 科技简讯

• 小红书笔记，长出「App」了！16 万人手搓，00 后打头阵 — 新智元
• 马斯克第三次喊「奇点已至」！AI 三个月连破数学、安全两道墙 — 新智元
• AI PPT 一改就崩？MemSlides 登顶抱抱脸，让 Agent 记住你的改稿习惯 — 新智元
• Pulling the Fire Alarm — LessWrong
• Contra George Hotz on “AI 2040 and the Cult of Intelligence” — LessWrong
• The Model Organism Lottery: Model Organism Interpretability Strongly Depends on Training Methodology — LessWrong
• vibes-based thinking as a cultural response to unknowns — LessWrong
• Red-teaming LLM unlearning: LUNAR’s “forgotten” knowledge is still recoverable — LessWrong
• Stateful Guardrails for Multi-Turn LLM Systems: A Conversational Risk Accumulation Framework — ArXiv CL (cs.CL)
• When Reasoning Narrows the Move: Diversity Collapse in LLM Game Play — ArXiv CL (cs.CL)
• On the Computational Complexity of Structural Generalization — ArXiv CL (cs.CL)
• Scaling Laws for Hypernetwork-Based Knowledge Injection in Large Language Models — ArXiv CL (cs.CL)
• Task Competence Is Not Instruction Following: Evaluating Instruction-Conflicting Behavior in Small Language Models — ArXiv CL (cs.CL)
• ModPack: An Extensible Teleoperation Interface for Bimanual Mobile Manipulation — ArXiv RO (cs.RO)
• Emergent Autonomous Drifting for Collision Avoidance in Real-World Winter Driving Scenarios — ArXiv RO (cs.RO)
• Milo, a Fully Autonomous Indoor/Outdoor Robotic Guide Dog — ArXiv RO (cs.RO)
• Learning Personalized Safety Interventions for Haptic Human-Robot Shared Control — ArXiv RO (cs.RO)
• LENS: LLM-guided Environment Simplification for Planning and Control in Clutter — ArXiv RO (cs.RO)
• Detect Early, Escalate Rarely: Anytime Detection of AI-Generated Video from the Compressed Bitstream — ArXiv CV (cs.CV)
• Crowd4D: Scene-Aware Monocular 4D Crowd Reconstruction — ArXiv CV (cs.CV)
• D3VL: Understanding Driving Scenes from 3D Time Series Data and Video with Language Models — ArXiv CV (cs.CV)
• Synthetic and Derived Training Images for Campus Waste Detection: A Multi-Seed Evaluation with YOLOv8n — ArXiv CV (cs.CV)
• ChronoStitch: Training-Free Composition of Visual KV Memories for Long-Horizon Temporal Reasoning — ArXiv CV (cs.CV)
• Challenge: Hand coding weights for efficient sequence memorisation — AI Alignment Forum
• Are we existentially threatened by the type of AI misalignment seen in the OpenAI Hugging Face attack? — AI Alignment Forum
• Nvidia bets physical AI can solve healthcare robotics’ data problem — AI News (Jack Clark)
• AMD to invest up to $5 billion in Anthropic under AI infrastructure deal — AI News (Jack Clark)
• FineServe: A Fine-Grained Dataset and Characterization of Global LLM Serving Workloads — arXiv AI (cs.AI)
• Hybrid LSTM-Graph Neural Framework for Robust Financial Fraud Detection and Adversarial Resilience — arXiv AI (cs.AI)
• OpenEvoShield: Dual Non-Stationary Continual Defense for Open-World Multi-Agent System Attacks — arXiv AI (cs.AI)
• Benchmarking Confidential GPU Inference on NVIDIA H100 under Intel TDX — arXiv AI (cs.AI)
• FormulaSPIN: Self-Play Fine-Tuning for Natural Language to Spreadsheet Formula Generation — arXiv AI (cs.AI)
• The Anatomy of an Agentic ITSM Workflow — Salesforce AI Blog
• What are the Best AI Chatbots for Small Business Customer Service? — Salesforce AI Blog
• Salesforce Named a Leader in the 2026 Gartner® Magic Quadrant™ for Conversational AI Platforms — Salesforce AI Blog
• Launching Health in ChatGPT — OpenAI Blog
• ChatGPT will give you worse health advice if you don’t pay — The Decoder
• Flux 3 generates videos with native audio up to 20 seconds long, a first for Black Forest Labs — The Decoder
• One tampered ChatGPT link could spawn a rogue AI agent that took orders from an attacker every five minutes — The Decoder
• Poolside’s Laguna S 2.1 is a small open-weight coding model that punches well above its size — The Decoder

📊 数据概览