Young Jae Koh comments on Pretraining data, not verifiability, is why LLMs are especially good at math (and coding)

[Young Jae Koh]

I really love this counterfactual intervention!