It strikes me that ChatGPT-style fine-tuning is in effect specifically optimising for The Barnum or Forer effect. Answers that feel more accurate get rated as more accurate and promoted. The entire AI bubble might just be based on the accidental automation of a carnival psychic con.