evhub comments on Basic facts about language models during training

[evhub]

(Moderation note: added to the Alignment Forum from LessWrong.)