A CTGT.ai research post reports that distilling DeepSeek into GPT-OSS did not transfer DeepSeek’s censorship behavior. The claim is relevant to a central question in model distillation: whether task capabilities and behavioral alignment policies are transferred together or can remain separable. The supplied metadata identifies a Hacker News discussion with a score of 169 and 73 comments, but does not provide the paper’s experimental protocol, datasets, baselines, or quantitative results. Those details should be verified in the original article before drawing broad conclusions.
No heat snapshots are available in the last 24 hours.