Reproduce “Learning from Interaction Experience via Localized Context Distillation” 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

Zhezheng Hao, Yongkang Yang, Xilin Xia, Jiarui Yu, Yi Liu, Fanjunduo Wei, Peijie Dong, Tianfu Wang and 4 more

LoCD transfers feedback-informed corrections into a language model by locating critical spans in student trajectories and having a teacher regenerate each span from the student’s prefix. This combines student-visited states with teacher-generated behavior absent from the student’s attempts. On Qwen3-8B, its mean score across eight math benchmarks was 48.04, versus 42.39 for OPCD in the reported evaluation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.