evhub comments on LLMs Sometimes Generate Purely Negatively-Reinforced Text

[evhub]

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