Reproduce “When and What to Teach: Budget-Aware Online Adaptation for Web 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
Jianwei Zhang, Sihan Cao, Pengcheng Zheng, Ya Wen, Pei Ke, Kuien Liu, Shen Gao, Wei Dong and 2 more
The authors present selective teaching for local agents. A gate uses prior teacher outcomes to choose failures for queries, while turn selection retains difficult teacher actions and likely repeated student errors for preference training. Across streams of 125 MiniWoB and 150 TimeWarp episodes, teacher calls fall 22.6% and student compute 52.1% on average, with comparable but not identical first-pass success.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.