2026-08-04 科技日报

扫描 552 篇候选内容 · 覆盖 650 个信息源 · 纳入 532 篇

📌 今日要点

• Demo 的高估值时代结束了，具身智能开始按生产力重新算账
• 刚刚，阿里 Qwen3.8-Max 来了！冲进全球第一梯队，模型表现直逼 Claude
• 阿里 Qwen3.8 正式发布，编程与办公再进化，推理更快更稳定
• 阿里“千问办公”开启公测
• 赛博义父 Tibo 爆料：谷歌早一年就做出了 ChatGPT，硬是没敢发！
• AI 不再用完即忘：华为诺亚开源 MindMemOS，记忆和 Skill 一起进化

📝 科技简讯

• Demo 的高估值时代结束了，具身智能开始按生产力重新算账 — 量子位
• 刚刚，阿里 Qwen3.8-Max 来了！冲进全球第一梯队，模型表现直逼 Claude — 量子位
• 阿里 Qwen3.8 正式发布，编程与办公再进化，推理更快更稳定 — 量子位
• 阿里“千问办公”开启公测 — 量子位
• 赛博义父 Tibo 爆料：谷歌早一年就做出了 ChatGPT，硬是没敢发！ — 量子位
• AI 不再用完即忘：华为诺亚开源 MindMemOS，记忆和 Skill 一起进化 — 量子位
• Attackers Can Subliminally Implant a Backdoor at Low Sample Count Without Prompt Access — LessWrong
• Selective Identity — LessWrong
• III. Anthropic reasoning has issues with infinite worlds; D-SIA can fix this — LessWrong
• Why biological weapons are scary, and what we can do about it — LessWrong
• Deliberate Alignment Faking as a Defense Against Model Poisoning — LessWrong
• Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face — AI Alignment Forum
• 刚刚，Qwen3.8-Max 首发上岗！「千问办公」正式开测 — 新智元
• OpenAI 连破 10 道数学难题，Fable 24 小时「复现」5 道 — 新智元
• 让沉默三千年的甲骨开口：首个模仿人类专家工作流的辅助释读系统 — 新智元
• 马斯克宣布：Grok 学会看片了！无字幕看懂陶哲轩菲奖级难题 — 新智元
• AI 证伪百年数学猜想被打假！Lean 证明惊现漏洞，哥大教授破防了 — 新智元
• Can LLMs Really Understand Item Difficulty Levels? Implications for Automated Item Generation Using LLMs — ArXiv CL (cs.CL)
• Imbalanced Data Clustering via Targeted Data Augmentation Using GMM and LLM — ArXiv CL (cs.CL)
• Chain-of-Models: Cross-Model Auditing for Bias-Robust LLM Judges — ArXiv CL (cs.CL)
• ZeroR@CHiPSAL 2026: Two-Stage Vision-Language Adaptation with Contrastive Learning for Nepali Meme Classification — ArXiv CL (cs.CL)
• Learning Stateful Predictive Knowledge From Experience — ArXiv CL (cs.CL)
• Advances, challenges, and opportunities for legged robots — ArXiv RO (cs.RO)
• ST-WAM: Semantic-Temporal World Action Model for Robust Manipulation under Visual Distribution Shifts — ArXiv RO (cs.RO)
• Receding-Horizon Next-Best-View Planner for Autonomous Leaf Surface Reconstruction — ArXiv RO (cs.RO)
• D-VLC: Decentralized Vision-Language Collaboration for Heterogeneous Embodied Multi-Robot Systems in Unknown Environments — ArXiv RO (cs.RO)
• DART: Dual-Axis Airborne Reachability-Gated Torque-Reaction for Off-Road Vehicle Jumps — ArXiv RO (cs.RO)
• ReLoop-UME: Recurrent Depth with Learnable Retrieval Registers for Universal Multimodal Embedding — ArXiv CV (cs.CV)
• SCMA: Structure-Conditioned and Metal-Aware Flow Matching for CT Metal Artifact Reduction — ArXiv CV (cs.CV)
• WaiT for the Signal: Simple Frequency-Aware Flow-Matching — ArXiv CV (cs.CV)
• Uncertainty-Aware Deepfake Detection via Multi-View Structural Learning — ArXiv CV (cs.CV)
• Do Medical Foundation Models Generalize on the African Brain? — ArXiv CV (cs.CV)
• Alibaba’s new Qwen model is also taking your job, but this time it’s great — The Decoder
• IBM finds 92% of companies hit by AI security breaches lacked basic access controls — The Decoder
• Interpol says AI has become the “core operational driver of cybercrime” across Africa — The Decoder
• China’s MiniMax H3 is the first open model to top an AI video ranking — The Decoder
• Unicorn, pelican, Middle-earth: OpenAI co-founder Karpathy is looking for the next AI vibe test — The Decoder
• Two teams solved the same quantum crypto problem using GPT-5.6 just three hours apart — The Decoder
• Alibaba’s open-weight Qwen3.8-Max takes on long-horizon AI tasks with 2.4 trillion parameters — The Decoder
• OpenClaw and Ollama in Agentic AI: Toward Fully Autonomous and Scalable AI Agent Systems — arXiv AI (cs.AI)

📊 数据概览