Exactly my thought. How much of the result is actually the consolidation of an unknown amount of researchers using the LLM to "sort their thoughts".

If it's true that they don't know how much of the training data contribution came from which user, they also have a weird race-condition on each result, where they don't know how distributed the contributed data actually is across users.

This means on each AI result they don't know how close an individual researcher already is to the same conclusion, so they need to rush to a press-release before some human devalues their (multi-million) compute-investment...