OpenAI has a reputation and a history for pushing compute forward (AI in Dota2, GPT, CLIP, DALL-E, etc.).
They have everything to lose by lying. If they say that these examples are not cherry-picked, then we have no reason (a priori) to doubt them.
On a side-note, the fact that you could doubt the results are real is telling: each of their compute-heavy experiments shakes our belief and further reinforces their reputation.
They have everything to lose by lying. If they say that these examples are not cherry-picked, then we have no reason (a priori) to doubt them.
On a side-note, the fact that you could doubt the results are real is telling: each of their compute-heavy experiments shakes our belief and further reinforces their reputation.