Chain of thought is what LLMs report to be their internal process--but they have no access to their internal process ... their reports are confabulation, and a study by Anthropic showed how far they are from actual internal processes.