Reproduce “Know What You Don't Know: Agentic Data Synthesis by Evolving Generation Space for Closed-Loop Learning” 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
Wei Yang, Qing Ping, Youzhi Luo, Yawei Wang, Abhishek Kumar, Panpan Xu, Jun Huan
EVOSYN uses learner feedback to adapt synthetic training tasks for interactive agents. Its Reflector reallocates effort toward weaknesses represented in existing generation controls, or proposes new factors when current controls cannot express them; verified tasks and trajectories train the agent in repeated rounds. Across four benchmark settings, EVOSYN-Final-SFT averages 77.8% task success versus 64.5% for AutoPlay Style, under these evaluations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.