> If you can demonstrate that, I would put it to Strominger and his colleagues, and I imagine they would be obligated to cite your contribution in the peer-reviewed publication.

There's one little problem: OpenAI isn't actually open and doesn't reveal which dataset they used for training.