The main takeaways I got from the Tristan Buckmaster talk were:
- It is obvious that OpenAI participated in academic theft
- The solution to Navier Stokes was in the convex hull of known facts when the race started
- AI massively accelerated the discovery of the key insights the experts were making over the course of the few months before
- But without those experts, it is not likely that OpenAI could have solved Navier Stokes on their own
@markusde are there any new details on the academic theft part
@maxsnew I don't know what is new or not, but Tristan explained what they were doing that was new, and that OpenAI published the exact same construction two days after he got in contact. I don't do PDE's obviously but hearing the details of the construction really quashed any idea that this might have been an instance of parallel thinking imo.
@markusde link?
@markusde oh, was it just an in-person seminar?
@highergeometer @markusde YouTube randomly recommended me a recording of, I presume, this talk. It's hard to read the slides, but it's better than nothing, I guess