RE: LeoThread 2026-05-26 23-19

Math may resemble chess and go because results are verifiable
The window of fruitful human–AI collaboration could be short — a few years or less, not a decade

Progress in math may be uneven, with harder-to-formalize fields lagging, but that jaggedness could be time-compressed — superhuman performance across nearly all areas of math might appear within a few years (≈2–3)

AIs could become better at posing pure-math questions than humans and rapidly develop theories beyond human comprehension

Theorists may retain a recreational comparative advantage over other humans in understanding those theories, while AIs would be better at translating them for applied researchers; pure mathematicians might need to shift toward applied work until applied research is automated

Confidence: ~50–65% for the general gist, ~40–50% that all specific claims are correct