"Hackenburg found that models trained to become more persuasive also ended up being less truthful."

It sounds like the arguments were basically AI slop hallucinations.