Zhaoyi Li

OPD excels at transferring reasoning skills over specific answers, revealing a nuanced relationship between teacher-student origins that can complicate multi-teacher setups.

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.