Show HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it
Summary
The article reports on a study that distills knowledge from a censored Chinese model into an American open-base GPT-OSS model. It finds that censorship in the teacher does not transfer to the distilled student across finance-related prompts, and presents a detailed evaluation of self-distillation versus teacher-taught distillation, along with the release of LineageEval and open weights.