Research Notes
Fan Wang's Blog
关于基础模型、智能体、强化学习与通用智能的长篇思考。
Long-form notes on foundation models, agents, reinforcement learning, and general intelligence.
OpenClaw爆火的迷思：关于基模、脚手架和工具
Big Model vs Big Harness: 底层逻辑
OpenClaw的爆火再次超出预期，结合近期关于Big Model与Big Harness的争论，我个人也深感技术发展之迅速，并对此感到些许迷茫。近期与一些业内人士交流后，针对目前AI发展最重要的三个要素：基模、脚手架和工具，我大致形成了一些看法，在此不妨分享。
REGEN: Recycling Expert Experience to Train a Generalist
Reusing replay memory with offline reinforcement learning to consolidate expert capabilities
[Paper] [Code]
Teaching Linear Attention to Remember: Stateful Training in StateLinFormer
Persistent memory across batch boundaries for embodied navigation
[Paper]
Misconceptions Behind OpenClaw's Explosion: Foundation Models, Harnesses, and Tools
Big Model vs. Big Harness: The Underlying Logic
OpenClaw’s explosive popularity has once again exceeded expectations. Alongside the recent debate over Big Model versus Big Harness, the speed of technological development has l...
Large-Scale Meta-Learning Elicits In-Context Reinforcement Learning
Agents Learn to Iterate Through Closed-Loop Context in Random Worlds
[Paper] [Code]
A Benchmark for General-Purpose In-Context Learning
Give AI a Fish, and You Feed It for a Task; Teach AI to Fish, and You Enable It to Learn
[Paper] [Code]
Information Bottlenecks and Plastic Recurrent Neural Networks
A Different Scaling Law: From Parameter Scaling to Memory Scaling
[Paper] [Code]