Zhaoyi Li

OPD transfers reasoning skills rather than answers, revealing a complex interplay between teacher-student origins that can either enhance or hinder model capabilities.

Uncover a model's "digital DNA" – its pretraining data mixture – from its outputs alone, even without access to the training data.

LLMs can learn to reason *worse* from seemingly better training data: models trained on CoT data with lower loss can generalize poorly due to inheriting inefficient, divergent reasoning patterns.