Reproduce “Co-Evolving Acting and Forecasting for Language Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Haofei Yu, Yixing Chen, Yining Zhao, Xuyan Ye, Yuxiang Liu, Zhigang Hua, Yan Xie, Shuang Yang and 1 more
CoPE trains a language agent with reinforcement learning and forecasts of several future actions from successful trajectories. It predicts actions from multiple earlier histories while acting one step at a time. On ScienceWorld, success rate reaches 72.5%, compared with 64.5% for GRPO and 67.5% for ECHO. An ablation supports multi-delay supervision, but does not isolate it as the sole cause.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.