Frontier models are not as good as the navier stokes solution would lead you to believe

I’m generally optimistic about AI, but the hype surrounding the recent Navier-Stokes proof announcement misrepresents where frontier models actually stand.

​Yes, AI formalisation tools like Lean have advanced, and seeing models assist human mathematicians with real breakthroughs is genuinely impressive. But claiming the core model itself has reached human-level mathematical brilliance ignores what actually happened behind the scenes.

​This wasn't a single super-smart AI having a lightbulb moment. It took 10,000 concurrent agents running for 88 hours, exchanging 2.7 million messages and burning through roughly 130 billion tokens.

​If you talk to a frontier LLM directly right now, it isn't giving you these kinds of insights natively. What happened here wasn't a leap in base reasoning capacity; it was an industrial-scale tree-search. The orchestrating framework spun up a massive web of parallel loops to kind of brute-force possibilities, prune dead ends, and cross-pollinate the few branches that didn't fail.

​That is a triumph of massive compute infrastructure, automated verification, and cluster orchestration—not an indicator that the base neural network possesses human-level domain intuition.

​Anyone who thinks AI won't keep improving is blind to the trajectory. But we need to separate agentic scale from model intelligence. Throwing a century's worth of parallel human work-hours at a single problem until the math compiles in Lean isn't AGI; it's just raw compute applied to an automated search space.

Edit When I say brute force, I don't mean it in the literal sense as AI does have some form of intelligence that's obvious, just not at the level where these models can solve millennium problems.