Tao's Point on AI Math: Theorem Proving Is RLVR-Easy, Conjecturing and Abstraction Aren't
akbirthko · x · 2026-09-12
The poster relays what they see as Terence Tao's actual point: "solve theorem X" isn't the only goal of math — historically hard problems mattered because they forced new concepts, methods, and conjectures. While bullish long-term on AI doing theory, the poster flags a medium-term concern: LLM capabilities are jagged — theorem proving is easy to train with RLVR, but conjecturing and abstraction are not.
More from AGI Musings
- UBI and discretionary income: what happens to middle-class lifestyles? — theearthcrosser · 2026-10-31
- One year later: is MIT's claim that 95% of AI pilots fail actually true? — killboy123 · 2026-10-31
- Nat Lambert: Open models could be what holds the AI ecosystem together — natolambert · 2026-10-31
- Kai-Fu Lee's 'AI Native' ships Nov 3 with endorsement from Satya Nadella — kaifulee · 2026-10-31
- Half-Life writer Chet Faliszek argues AI shortcuts erase the intentionality behind game craft — slhamlet · 2026-10-31
- Companies that replaced engineers with AI are rehiring seniors at a premium — DavidLinthicum · 2026-10-31