Papers2

#on-policy training

EvoCUA: Evolving Computer Use Agents via Learning from Scalable Synthetic Experience

Taofeng Xue, Chong Peng et al.Jan 22arXiv

Before this work, computer-using AIs mostly copied old examples and struggled with long step-by-step tasks on real computers.

#computer use agent#verifiable synthesis#validator

Diversity or Precision? A Deep Dive into Next Token Prediction

Intermediate

Haoyuan Wu, Hai Wang et al.Dec 28arXiv

The paper shows that teaching a language model with a special “reward-shaped” next-token objective can make later reinforcement learning (RL) work much better.

#next-token prediction#cross-entropy as policy gradient#reward shaping