OpenAI solves Navier-Stokes using on the order of 16M dollars of better-than-Astra internal model power. https://openai.com/index/navier-stokes-solution/ Possibly including side-channel information from Tristan and Levent's work? (See https://cims.nyu.edu/~tristanb/statement.pdf) This is a remarkable result. It's a good example of computationally-assisted math. But is it an example of how to generate new mathematical insight? Messy.