https://www.ft.com/content/a0dfedd1-5255-4fa9-8ccc-1fe01de87ea6

OpenAI is alleging that DeepSeek is a distillation of the GPT models. Will this become the consensus view?

This question resolved as “yes” if I judge the majority of credible commentators to hold this view. If there’s reasonable disagreement but a clear consensus, it still resolves as “yes”. I’ll weight my judgment of the consensus heavily on what Zvi reports: https://open.substack.com/pub/thezvi

If there’s no consensus by the time this question closes, it resolves as “no”, even if it’s plausible that a consensus may be reached shortly thereafter.

Update 1/30: the spirit of this question is whether distilled GPT data was centrally used to train DeepSeek models. If some GPT-generated outputs were in the training mix, but in trivial amounts, that will not be sufficient for this question to resolve “yes”.