Raw LLM Responses

It does not help that there is indeed so much real slop on facebook etc. that I wonder what kind of models they use, well it probably is something dirt cheap... But even dirt cheap/free things today would generate something more believable on maybe 3rd try or if you prompted better. But it is indeed how it is, even two months ago on an evening show in my country they had an actor and then read some facts about him as if by ''ChatGPT'' which had some incorrect things they laughed about. I immediately went there and tried to recreate it... nope, no hallucinations, no fake facts. Neither GPT 5, nor Gemini 2.5 at the time hallucinated anything. It was an issue with older models though. So I am not sure if they just invented that, or they had it as part of the show but turned out newer models are better so they still... invented that. By no means the models are perfect and great at everything but they are constantly getting better, like Gemini 3 reading handwriting scribbled on a page in a hurry like pro now.