For a while I did say that if they actually cared about solving math they would, and not just rely on the Navier Stokes' recognizability. It's nice to see that they are putting their effort into this now that Lean can be used as a checking machine to game.

About the actual impact of the advances:

• For new discoveries there are always humans interacting with other humans and there is cross-pollination
• For agent discoveries we need to figure out how to do this properly.
• Perhaps some kind of agentic conferencing?
• Or maybe a continuously updated ledger of knowledge that agents from other fields rapidly build from?

Thinking about the OpenAI math results, they’re clearly a big deal in terms of substance but it feels deploying more AI in an IPO sports rivalry rather than advancing AI. This feels like the same advances from a month ago just being put to work with enough compute on more fodder, cranking through. The narrow superintelligence hasn’t gotten more super. It’s just doing more work.

Math is a curiously playable domain, but unclear how much of this is interesting beyond mathematics. Navier-Stokes is kinda interesting though I’m not sure whether it will open up interesting new fluid mechanics. In this new crop, the quasi Riemann Zeta result is obviously huge, but I don’t know if it matters outside math. Much of this feels ignorable the way the Super Bowl is ignorable. I think we need to revisit how we math relates to reality.

from Venkatesh Rao's Blyg ↗ · snapshot of v1