Comment by ndriscoll

If humans with the background to understand the proof can get the AI to satisfactorily explain itself, then OpenAI doesn't need to publish explanations or release things at a rate that a human can explain. They have an interactive explainer that you can just ask!

It's still of course useful to have some of this distilled understanding be written down, just like I had it generate high level descriptions of our code at work so it doesn't have to rediscover everything every time, but that's more of a cache/optimization.