2026-07-25 科技日报

扫描 545 篇候选内容 · 覆盖 650 个信息源 · 纳入 532 篇

📌 今日要点

• Spatially Grounded Concept Bottleneck Models for Trustworthy Breast Ultrasound Diagnosis
• What is Good? Extracting and Testing Implicit Theories of Literary Quality from LLM Reasoning Traces
• Knowledge Injection Exists in MoE? Exploring Expert-Aware Contrast Decoding in MoE for Mitigating LLMs’Hallucinations
• Is MoE Routing a Huffman Code? Discovering the Frequency-Diversity Law in Chain-of-Thought
• Human-in-the-Loop Large Language Model Framework for Identification of Cutaneous Immune-Related Adverse Events
• More Is Not More: What Matters for Diversity in LLM Opinions?

📝 科技简讯

• Spatially Grounded Concept Bottleneck Models for Trustworthy Breast Ultrasound Diagnosis — ArXiv CV (cs.CV)
• What is Good? Extracting and Testing Implicit Theories of Literary Quality from LLM Reasoning Traces — ArXiv CL (cs.CL)
• Knowledge Injection Exists in MoE? Exploring Expert-Aware Contrast Decoding in MoE for Mitigating LLMs’Hallucinations — ArXiv CL (cs.CL)
• Is MoE Routing a Huffman Code? Discovering the Frequency-Diversity Law in Chain-of-Thought — ArXiv CL (cs.CL)
• Human-in-the-Loop Large Language Model Framework for Identification of Cutaneous Immune-Related Adverse Events — ArXiv CL (cs.CL)
• More Is Not More: What Matters for Diversity in LLM Opinions? — ArXiv CL (cs.CL)
• Anthropic claims its new Claude Opus 5 delivers near-Fable 5 performance at half the token price — The Decoder
• Microsoft’s open-weight AI push is so obviously an Azure play it hurts — The Decoder
• Sakana claims its AI model router Fugu Ultra v1.1 now beats Fable 5 without even including it in the pool — The Decoder
• German AI consortium releases Soofi S, an open 30B model that tops benchmarks in both English and German — The Decoder
• Claude’s voice mode now runs on Anthropic’s most capable models across all platforms — The Decoder
• Kimi K3 trails frontier US models by a wide margin on cyber exploits, and distillation may explain why — The Decoder
• Why Marketers Need to Move Beyond Attribution and Embrace Agentic Optimization — Salesforce AI Blog
• Employee Agents Vs. Customer-Facing Agents? Spot the Differences — Salesforce AI Blog
• Real-Time Engagement with Time Spent Tracking in Data 360 with Web SDK — Salesforce AI Blog
• AINTMA: Agentic AI Architecture for Autonomous Test Management with Generative Intelligence, Secure Cloud Communication and Adaptive Quality Analytics — arXiv AI (cs.AI)
• Marking the Wrong Symptoms: Evaluating LLM Watermarks in Medical Texts — arXiv AI (cs.AI)
• ClickGuard: Detecting and Spoiling Clickbait News with Informativeness Measures and Large Language Models — arXiv AI (cs.AI)
• Stochastic Sampling is Epistemically Shallow: The Dimensionality Gap Between Temperature Variation and Model Diversity in LLMs — arXiv AI (cs.AI)
• JAXBench: Benchmarking Autonomous TPU Kernel Optimization — arXiv AI (cs.AI)
• Are BuzzBallz actually recyclable? — BBC Science
• It’s the cocktail taking young drinkers by storm - but the packaging is ‘unrecyclable’ — BBC Science
• Firefighters struggle to contain Madrid wildfires, as more than 140,000 evacuated in France and Spain — BBC Science
• ‘Buck Moon’ set to light up sky — BBC Science
• Rain returns for some parts of the UK — BBC Science
• The Long (Self-)Correction — AI Alignment Forum
• 合肥又押中 AI 独角兽：多模态赛道，3 个月融了 21 亿 — 量子位
• 国产世界模型登顶李飞飞团队榜单！适配国产昇腾算力、代码权重全开源 — 量子位
• 菲尔兹奖得主王虹，也发过 NeurIPS — 量子位
• 抢跑！全球化新品牌进入 AI 时刻，何以定胜负？ | 2026 全球化新品牌 AI 竞争力大会实录 — 量子位
• 领克 07GT 重磅上市，以越级驾趣 + 全场景拓展生态掀翻旅行车市场 — 量子位
• 近百名玩家涌入具身数据：一年融资 44.7 亿，谁能真靠“卖数据”赚钱？ — 量子位
• 世界模型“六小龙”在 WAIC 吵起来了！行业红利就在非共识里 — 量子位
• 亚太唯一！阿里云跻身 Gartner 可观测魔力象限“挑战者”象限 — 量子位
• 长内容创作者苦 AI 失忆久矣，这个新 Agent 漂亮填坑！门槛低到只需要会用键盘打字 — 量子位
• OPC 创业者看过来，最高 10 万美元云资源等你拿 — 量子位
• HERMES: Heterogeneous Edge-Relational Multi-Head Embedded SSM Attention for Traffic Conflict Prediction at Signalized Intersections — ArXiv RO (cs.RO)
• PhysCoRe: Physics-Corrected Residual World Models for Material-Aware Deformable Dynamics — ArXiv RO (cs.RO)
• Scalable Low-Cost Laboratory Automation: A Digital Twin-Integrated Robotic Platform for Autonomous Liquid Handling (RAINBOTTM) — ArXiv RO (cs.RO)
• Safe and Scalable Multi-Drone Payload Transport via CBF-based Reinforcement Learning with Zero-Shot Sim-to-Real Transfer — ArXiv RO (cs.RO)

📊 数据概览