← back to paper
arxiv: 2607.05394 · 2 revisions
Weak-to-Strong Generalization via Direct On-Policy Distillation