2026-09-25 科技日报
3 min
扫描 561 篇候选内容 · 覆盖 650 个信息源 · 纳入 544 篇
📌 今日要点
- Introducing Gemini 3.8 Live with Live Avatar
- Accelerating vision-language models with LFM2.5-VL-DSpark
- The Drift Contract: Spectral Updates for Depth-Robust Local Learning
- Signal2Symbol: Neuro-Symbolic Temporal Reasoning for Explainable Physiological Time-Series Anomaly Detection
- HARN: Hierarchical Associative Resonance Network for Event-Driven Multi-Timeframe Forecasting
- What Makes a Terminal-Bench Task Hard? Separating Genuine Hardness from Fake-Hardness on an Adjudicated Agentic Corpus
📝 科技简讯
- Introducing Gemini 3.8 Live with Live Avatar — Google DeepMind
- Accelerating vision-language models with LFM2.5-VL-DSpark — Hugging Face Blog
- The Drift Contract: Spectral Updates for Depth-Robust Local Learning — ArXiv ML (cs.LG)
- Signal2Symbol: Neuro-Symbolic Temporal Reasoning for Explainable Physiological Time-Series Anomaly Detection — ArXiv ML (cs.LG)
- HARN: Hierarchical Associative Resonance Network for Event-Driven Multi-Timeframe Forecasting — ArXiv ML (cs.LG)
- What Makes a Terminal-Bench Task Hard? Separating Genuine Hardness from Fake-Hardness on an Adjudicated Agentic Corpus — ArXiv ML (cs.LG)
- LWCal: Loss-Weighted Calibration for Tabular Classifiers with Noisy Calibration Labels — ArXiv ML (cs.LG)
- A Leakage-Aware Multimodal Evaluation Framework for Early Intraoperative Acute Kidney Injury Prediction — ArXiv ML (cs.LG)
- COPE: Continual Personalization of LLMs under Sparse User Feedback via User Embeddings and Self-Evaluation — ArXiv ML (cs.LG)
- Speaker-labeled transcription with WhisperX on SageMaker AI — AWS Machine Learning
- Build a multi-account AI agent with AgentCore Gateway and MCP — AWS Machine Learning
- Aderant builds intelligent ticket triage with Amazon Nova — AWS Machine Learning
- Continual learning might make your blocking monitors nearly useless — AI Alignment Forum
- MCP Explained in 5 Minutes — KDnuggets
- What I’ve Learned About DeepSeek Harness — KDnuggets
- Bring your co-founder, partner, or colleague and get 50% off a second TechCrunch Disrupt 2026 pass — TechCrunch AI
- PrismML brings its tiny LLMs to Qualcomm-powered smart glasses — TechCrunch AI
- Oracle sends force majeure notice on its New Mexico Stargate data center — TechCrunch AI
- Meta’s Muse Charm looks like a Tamagotchi, but it’s tapping into a much newer trend — TechCrunch AI
- Google Photos ‘Clueless’-inspired virtual closet is now available on Android and iOS — TechCrunch AI
- TechCrunch Founder Summit 2026: Everything you need to know — TechCrunch AI
- ElevenLabs’ CEO on margins, IPO timing, and telling customers they’re talking to a bot — TechCrunch AI
- What did AI researchers think at the end of 2024? — LessWrong
- Continual learning might make your blocking monitors nearly useless — LessWrong
- The most important problem (you’ve never heard of) — LessWrong
- First-Order Definable Policies — LessWrong
- Overtly misaligned trajectories score highly in RL. — LessWrong
- CPU 重回 C 位!鲲鹏给 Agent 造了一台「超级计算机」,冲到 4096 节点 — 新智元
- 上线 12 天跑赢 ChatGPT!Meta 爆款 AI 自曝老底,灵魂文件来自 OpenClaw — 新智元
- 日榜第二!北大开源,强化学习数据策略接入更简单、对比更公平 — 新智元
- 三榜第一,完胜 GPT-6!中国物理 RSI 杀出黑马 — 新智元
- 刚刚,ChatGPT 能「开口干活」了! — 新智元
- How to Maximize Your Coding Agent Subscriptions — Towards Data Science
- Beyond RAGs: Building Actually Truthful AI Harnesses — Towards Data Science
- Towards Spec-Driven Test Automation: Part 1 — Towards Data Science
- When the Correct Answer Is Nothing, What Does Your Pipeline Return? — Towards Data Science
- AgroBench: A Reproducible Multimodal Benchmark for Weakly Supervised Crop Yield Learning from County Statistics and Pixel Observations — ArXiv CV (cs.CV)
- Cross-Modal Contrastive Learning from Histopathology and CT for Automated Renal Cell Carcinoma Grading — ArXiv CV (cs.CV)
- A 3D Pose-Based Ensemble Framework for Cricket Shot Classification and Automated Biomechanical Analysis — ArXiv CV (cs.CV)
- nnFoundation: 3D Foundation Models for Radiology — ArXiv CV (cs.CV)
📊 数据概览
| 指标 | 数值 |
|---|---|
| 候选内容 | 561 |
| 去重后 | 544 |
| 纳入日报 | 544 |
| 主题分组 | 0 |
| 独立条目 | 40 |
| 信息源数量 | 650 |