Reproduce “Verify by Doing: Co-Evolving Agent Execution and Evaluation without External Feedback” 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
Yitong Lin, Jianda Chen, Yihao Hu, Pan Wang, Hang Wang, Zhihao Wen, Xiaoxue Ren, Wei Wu
The authors introduce Verify by Doing (VbD), a method that lets an agent’s Evaluator inspect task artifacts and state, then uses its findings to update both task-execution and checking procedures. On AutomationBench, VbD reaches a 37.0% overall pass rate versus 33.7% for RQGM. This three-run mean does not establish fewer attempts, cross-task transfer, or sustained improvement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.