Reproduce “A Good Self-Teacher Meets the Student Where They Are: Joint On-Policy Learning and Teaching” 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

Randy Ardywibowo, Arnav Dalal, Jiantao Jiao

JOLT trains a shared policy as both privileged teacher and unprivileged student. The teacher learns from outcomes under token-level KL regularization toward the student, while the student receives dense feedback on its own responses. On MATH, JOLT reaches 72.14 ± 0.90%, versus GRPO at 68.36 ± 3.73%. At matched accuracy, token budgets are estimated lower, but do not measure elapsed time.

Summary by alphaXiv.

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