Stephan Wolfram talks about how we'll soon have LLM psychologists, that if you add to the prompt "take a deep breath before answering" you get better answers.

Stephan Wolfram talks about how we'll soon have LLM psychologists, that if you add to the prompt "take a deep breath before answering" you get better answers. He also speculates that if after the answer you ask it "is what you wrote correct?" you give the LLM a chance to get out of path dependency - one badly generated token and the LLM is now forced to follow it through.