Alborz Geramifard

Distilling teacher-relative logit shifts, not just final policies, unlocks substantial gains in multi-teacher on-policy distillation by preventing inherited base biases from dominating learned preferences.

TREK transforms the way models tackle challenging prompts by expanding their exploration support, leading to substantial performance gains even in the hardest task scenarios.

Role-typed credit assignment can drastically improve reinforcement learning outcomes by accurately distinguishing between useful exploration and regression in agent actions.

Overconfident tokens, often missed by entropy-based methods, carry surprisingly dense corrective signals in on-policy distillation, allowing for near-baseline performance with <10% of tokens.