Claude is More Anxious than GPT; Personality is an axis of interpretability in language models — LessWrong