2026-08-13 科技日报
4 min
扫描 565 篇候选内容 · 覆盖 650 个信息源 · 纳入 551 篇
📌 今日要点
- Retrofitting Recurrent Depth into a Pretrained Language Model: Installation, Extrapolation, Transfer, and Retention at Two Parameter Budgets
- TRACE Bench: Task-driven Roleplay Agentic Checklist Evaluation
- Lost in Compaction: Evaluating Side-Constraint Loss under Context Compaction
- Diffuse to Compress: Leveraging Diffusion LMs for Lossless Compression
- The builder’s guide to GPT‑5.6
- Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speed
📝 科技简讯
- Retrofitting Recurrent Depth into a Pretrained Language Model: Installation, Extrapolation, Transfer, and Retention at Two Parameter Budgets — ArXiv CL (cs.CL)
- TRACE Bench: Task-driven Roleplay Agentic Checklist Evaluation — ArXiv CL (cs.CL)
- Lost in Compaction: Evaluating Side-Constraint Loss under Context Compaction — ArXiv CL (cs.CL)
- Diffuse to Compress: Leveraging Diffusion LMs for Lossless Compression — ArXiv CL (cs.CL)
- The builder’s guide to GPT‑5.6 — OpenAI Blog
- Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speed — OpenAI Blog
- OpenAI appoints Dali Rajic as Chief Revenue Officer — OpenAI Blog
- Adaptation of Generalist Robot Policies with Minimal Data — ArXiv RO (cs.RO)
- Top-down Traffic Scenario Generation via Joint Initial-Goal Diffusion and Trajectory Infilling — ArXiv RO (cs.RO)
- From Self-Normal-Positioning to Omni-Directional Tracking: Real-Time Surface Modeling Enabled Probe Tilt Control for Robotic Ultrasound Imaging — ArXiv RO (cs.RO)
- Locomotion Variability and User Experience in Smart Wheelchair Human-Robot Interaction — ArXiv RO (cs.RO)
- Herding End-to-End Autonomous Driving via Neuro-Symbolic Safety Guards — ArXiv RO (cs.RO)
- Ilya 的首个模型来了! — 新智元
- Claude 新风口杀疯!海外年轻人上头,刷 App 成瘾 — 新智元
- Opus 5 通关 ARC-AGI-3!Harness 正在变成捆住模型的绳子 — 新智元
- 刚刚,中国 AI 交卷!40 年难题 56 分钟破局,AI 不再「学习」 — 新智元
- 困扰数学圈 22 年的难题,居然被协和实习医生解决了? — 新智元
- Features that current AIs don’t have that future AIs will have — LessWrong
- Some Ways I Think About Evaluating Grant Applications — LessWrong
- Is Alignment Even Falsifiable? Middle Alignment, An Alignment Taxonomy, and Breaking The Problem Down Into Steps — LessWrong
- Concrete Generalist Projects in AI Safety (and how to do them) — LessWrong
- Comparing Congress’s Two AI Emergency Shutdown Mechanisms — LessWrong
- Suno Studio 2.0’s new chat feature lets you talk to your DAW like it’s a bandmate — The Decoder
- Gemini 3.7 Flash lands with coding gains and undercuts its three-week-old predecessor’s price by 50% — The Decoder
- Deepseek ships improved V4 Pro, open-sources its agent software, and raises API prices — The Decoder
- Ling 3.0 Flash is the smartest open model at its size — The Decoder
- Fable 5’s slow adoption suggests corporate willingness to pay for frontier AI has hit a ceiling — The Decoder
- Top AI lab researchers warned about automated AI research, and several of their predicted milestones have already fallen — The Decoder
- Anthropic brings Claude Cowork to its Chrome extension, adding skills and plugins to the browser — The Decoder
- GeoUniPR: A Geometry-Consistent Unified Framework for Cross-Modal Place Recognition — ArXiv CV (cs.CV)
- SegPAR: Class-Centric Decision-Based Sparse Attack for Semantic Segmentation — ArXiv CV (cs.CV)
- CLEAR: Class-wise Expert Aggregation with Structured Sampling for Long-Tailed Classification — ArXiv CV (cs.CV)
- Self-Evolving Code-with-Image Reasoning — ArXiv CV (cs.CV)
- Clinical Feasibility of Low-Magnification Fluorescence Imaging for Breast Cancer Margin Detection Using Texture Analysis and Deep Learning — ArXiv CV (cs.CV)
- 深度体验 DeepSeek Harness,我原谅它涨价了 — 量子位
- 4.8 亿美元砸向端侧算力!Agent 芯片新贵冲出重围 — 量子位
- 马斯克 Grok 4.6 重回一梯队!更低价格反超 Fable 5,这 Cursor 是真没白收购 — 量子位
- 具身数据来了实战派!40 天 2 轮融资数千万,瞄准物理 AI 基础设施 — 量子位
- Claude 一举扫清 2000 阶以下哈达玛矩阵!AI 开始清空数学待解列表 — 量子位
- 中国人能飞!杭州发布全球首款站姿载人飞行器 — 量子位
📊 数据概览
| 指标 | 数值 |
|---|---|
| 候选内容 | 565 |
| 去重后 | 551 |
| 纳入日报 | 551 |
| 主题分组 | 0 |
| 独立条目 | 40 |
| 信息源数量 | 650 |