6:10: People seem very credulous when it comes to LLM "introspection". Like the LLM produces an output, then the user asks "how did you do it?" and the LLM then describes essentially the creative process a human would use to produce the output, and the user is like "wow, so smart, it reasons like a human".

An LLM doesn't "know its own mind"! It probably just summarised the WikiHow to you