The more subtle point is that this cannot be corrected via what appears to humans as “conversation” with the LLM. Because it is more plausible that a confident liar keeps telling tall tales, than it is that the same liar suddenly becomes a brilliant and honest genius.
A human on the internet loves to argue, stand and prove a point, simply because they can. Guess what the AI's were trained on? People talking on the internet.
