Quantara AI Daily · Saturday, October 25, 2026AIDAILY
Quantara
微软AI诊断系统准确率超医生两倍
微软的AI诊断系统MAI-DxO在诊断准确性方面优于医生，复杂病例测试中达到83.2%，而医生仅为36.1%。此外，至知创新研究院发布了320B开源大模型的训练细节，Anthropic的Claude在网络安全领域取得成果。
模型发布/更新
New AI Is More Than Twice as Good at Diagnosing Diseases as Doctors
Microsoft says its medical AI system can diagnose difficult cases more than twice as accurately as practicing doctors. Dubbed the Microsoft AI Diagnostic Orchestrator (MAI-DxO), the system “substantially outperformed practicing...
新智元RSI两大里程碑发布！单人手搓千亿模型，英伟达大佬教「保命」法则
Mirendil 和 Recursive 两家初创公司分别发布了 Mirendil-Base 和 Flint 两个编程模型，展示了单人训练千亿模型的技术突破，并引发了行业对 AI 编程能力提升对科研速度影响的讨论。
新智元100万道实战题，还得防AI作弊！320B国产基模技术细节揭秘
至知创新研究院发布了IQuest-Q1技术报告，详细介绍了320B开源大模型的训练细节，包括训练方法、性能评测和架构设计。
新智元Claude狂写1.8亿行证明！万亿次测试漏掉的Bug，被AI一锅端
Anthropic 的 Claude 在形式化验证领域取得了显著成果，通过数学证明方法揭示了网络安全漏洞，为网络安全领域提供了新的解决方案。
新智元Anthropic首发公益报告，自曝「人类拯救计划」！
Anthropic发布首份公益报告，指出AI训练AI的进展和安全隐患，强调AI安全的重要性，并宣布成立公益基金。
vLLM Team BlogServing DeepSeek-V4.1-Flash at Scale: vLLM Matches DeepInfer and Scales to GB300 NVL72
TL;DR: On one NVIDIA GB300 NVL72 rack, vLLM with P/D disaggregation and Wide-EP serves DeepSeek-V4.1-Flash on SemiAnalysis AgentX at 954K tokens/s per GPU at a P90 of 196 tokens/s per user, 7.9× our single-node result from the first...
Research Papers
Agent Control Flash News长视频记忆别手搓策略：Meta LVMH/SEEM，三基准均值 65.7%（+7.7 pp）、token ≈1/32
View PDF HTML (experimental) Abstract: Long-video agents must map hours or days of multimodal evidence into a bounded context while respecting token and latency budgets, making memory design a context-management challenge. Existing...
Agent Control Flash News可验证 CUA 训练数据工厂：港大 Xlang/Tao Yu × Qwen CUA-Gym，OSWorld-Verified 62.1/72.6
Authors: Bowen Wang , Dunjie Lu , Junli Wang , Tianyi Bai , Shixuan Liu , Zhipeng Zhang , Haiquan Wang , Hao Hu , Tianbao Xie , Shuai Bai , Dayiheng Liu , Que Shen , Junyang Lin , Tao Yu View PDF HTML (experimental) Abstract:...
Agent Control Flash News自进化要有不可改的不变量核：UCL/NTU Steward，24.6% 行业回撤压到 0.3% 损失
View PDF HTML (experimental) Abstract: LLM agents that act on live Web data must combine heterogeneous evidence, remember what they have done and improve from outcomes without weakening their own safeguards. We study this problem in...
Agent Control Flash NewsMLE 自进化要跨枝信息流：上海 AI Lab MLEvolve，12h 预算奖牌率 65.3%
Authors: Shangheng Du , Xiangchao Yan , Jinxin Shi , Zongsheng Cao , Shiyang Feng , Zichen Liang , Boyuan Sun , Xu Guo , Tianshuo Peng , Yifan Zhou , Jie Hou , Xin Li , Jie Zhou , Liang He , Bo Zhang , Lei Bai View PDF HTML...
Agent Control Flash News别另训 critic：Apple LVRL，AppWorld 最高 +13.6 pp、BabyAI 0→75.0%
View PDF HTML (experimental) Abstract: Reinforcement learning (RL) for large language model (LLM) agents remains challenging in long-horizon tasks with sparse outcome rewards, where assigning credit to individual actions is difficult....
Agent Control Flash News卖方 agent 会自发 GEO 搞砸工具市场：Google/DeepMind Toolflation，成功率 89%→约 30%
View PDF HTML (experimental) Abstract: Agents are starting to sell to other agents. Seller agents write the listings of tools, services and products, and buyer agents read these listings when they choose what to use. We show that seller...
AI趣闻
新智元刚刚，全球互联网已被AI攻陷！史上最黑暗48小时来了
全球互联网在48小时内遭遇大规模AI攻击，导致网站瘫痪、银行数据泄露，揭示了网络安全面临的挑战。