Pretraining data, not verifiability, is why LLMs are good at math (and coding)

(lesswrong.com) by gmays | Sep 23, 2026 | 0 comments on HN