tl;dr: LLMs appear intelligent not because they reason, but because they exploit the same cognitive bias (subjective validation) that psychics use in cold reading: producing statistically generic responses that users interpret as specific, insightful, and personal. RLHF amplifies this by optimizing for confident, plausible-sounding output rather than factual accuracy, effectively training chatbots to generate Forer statements, shotgun guesses, and other mentalist tactics. The author argues this illusion is accidental but pervasive, and that intelligent, believer-inclined users are the most susceptible—making LLM-based decision-making functionally equivalent to calling a psychic hotline.