← 返回最新日报
2026-07-08· 当日存档
AI 日报
AI内部机制与模型新势力
Anthropic揭示Claude内部工作记忆，可读性突破；Meta发布Muse图像模型，性能逼近GPT-4o；中国AI模型使用率攀升，DeepSeek自研芯片应对出口管制。
论文研究
1
Claude's hidden inner monologue is now readable thanks to Anthropic's new Jacobian Lens
Anthropic新工具J-Lens可读取Claude内部思维，AI可解释性重大进展
The Decoder深挖 →
2
Interesting stuff. And the visualization at the end is worth trying: https://www.neuronpedia.org/qwen3.6-27b/jlens
语言模型内部存在类似人脑的全局工作空间，可视化工具值得尝试
@emollick深挖 →
3
4
Cortex: A Bidirectionally Aligned Embodied Agent Framework for Long-horizon Manipulation
Cortex框架弥合高层规划与底层执行鸿沟，推动长时域机器人操作
arXiv深挖 →
5
AgentGym2: Benchmarking Large Language Model Agents in De-Idealized Real-World Environments
AgentGym2在非理想化环境中评估LLM代理，更贴近真实世界
arXiv深挖 →
模型动向
6
Meta’s new Muse Image model can pull other Instagram users into AI photos
Meta Muse可拉取Instagram用户生成AI照片，社交与AI深度融合
The Verge AI深挖 →
7
8
9
10
11
NVIDIA and Hugging Face Bring New Models and Frameworks to LeRobot for the Open Robotics Community
NVIDIA与Hugging Face为开源机器人社区带来新模型和框架
NVIDIA深挖 →
12
This is what latency optimization looks like below the API 👇 Together ATLAS, NVIDIA Blackwell, CUDA, TensorRT-LLM, Dynamo, and custom kernels all w...
Together ATLAS与NVIDIA软硬件协同实现毫秒级推理延迟
X · List深挖 →
产品上新
13
Anthropic's Claude Cowork AI agent is now available on mobile and web
Claude Cowork AI代理扩展至移动端和网页端，支持后台持续运行
The Decoder深挖 →
14
Run AI workloads on any cloud, store on Hugging Face: zero-egress storage with SkyPilot
SkyPilot与Hugging Face集成，实现零出站费用的多云AI存储与计算
Hugging Face深挖 →
15
From Hugging Face to Amazon SageMaker Studio in one click
Hugging Face一键部署至Amazon SageMaker，简化模型上线流程
AWS ML深挖 →
16
How AI could enable autonomous robot workers in workplaces—and maybe homes
AI赋能机器人从工厂走向家庭，自主性成为关键
Ars Technica AI深挖 →
行业动态
17
Copilot goes cheap as Microsoft phases out OpenAI and Anthropic models to cut costs
微软用自研MAI替代OpenAI/Anthropic模型降本，Copilot性能或受影响
The Decoder深挖 →
18
Facing US export controls, China's DeepSeek plans to make its own chips
DeepSeek自研芯片应对美国出口管制，减少对英伟达和华为依赖
Ars Technica AI深挖 →
19
China eyes export curbs on its top AI models, and Europe is caught in the middle
中国拟限制顶级AI模型出口，欧洲开源模型捷径或关闭
The Decoder深挖 →
20
Chinese AI models regularly pass 30 percent on OpenRouter as cost gap widens
中国AI模型在OpenRouter使用率超30%，成本优势显著
The Decoder深挖 →
21
OpenAI and Anthropic are giving away millions in computing power to attract startups
OpenAI和Anthropic提供数百万美元免费算力争夺初创公司
The Decoder深挖 →
22
AI Innovators Adopt NVIDIA Vera — Why Max Single-Threaded CPU at Scale Matters
NVIDIA Vera CPU以最强单线程性能驱动AI智能体时代
NVIDIA深挖 →
23
The first American autonomous ground vehicles are fighting in Ukraine
美国无人地面车辆首次在乌克兰实战部署超百辆
TechCrunch AI深挖 →
24
25
26
Apollo economist warns AI profit gains outside tech could take "well beyond" what Wall Street expects
非科技行业AI利润增长或远超预期，但需数年时间
The Decoder深挖 →
实用技巧
27
YC CEO says he ships 37K LoC AI code per day. A developer looked under the hood
YC CEO声称日写3.7万行AI代码，开发者实测揭露真相
Hacker News深挖 →
活动会议
28
I got in on last round and I just applied again. I am not sure if they will do it twice though. If you maintain an open source project, I highly recom...
开源项目维护者可申请免费6个月Claude Max 20x使用权
X · List深挖 →
快讯
·
Deepseek is designing its own AI chip
The Decoder