AI Chatbots Remain Overconfident — Even When They’re Wrong: Large Language Models appear to be unaware of their own mistakes, prompting concerns about common uses for AI chatbots.

I'm getting pretty sick of this daily discovery, for years, that statistics machines for spitting out words by copying patterns aren't intelligent, thinking beings with ethics and an understanding of their world. It's like the news reporting every day that car seats may not be mammals after all. Makes me feel a bit crazy.

God, the coverage of machine learning in general is absolutely infuriating. I sometimes listen to the BBC daily news podcast while I shower, and today they thought it was newsworthy that a soccer "AI" could predict where a penalty kicker would put the ball 52% of the time. Fifty fucking two percent. A straight up coin flip. They were acting like it was some huge achievement. They wrote that, recorded it, edited it, broadcasted it, and at no point did anyone say, "wait, this is fucking stupid, cancel this useless story."

I need to find a different global news broadcast to listen to.

Edit, just because I felt like I was going crazy, I decided to look up the story again. The text article says it was 64% accurate, and in the opening news summary the BBC actually also said 64%, but then in the main story the number they presented was 52% (at about 16:20 in the episode titled "Israel to open humanitarian corridors into Gaza"). So I dunno, maybe that's supposed to be some separate stat somehow, but anyway it's still crystal clear this shit so ass

appear to be unaware of their own mistakes

Of course, they are not aware of anything. If they realised it was a mistake they wouldn't have said it first time.

They’re not aware of anything at all, it’s software, it only does what it’s programmed to do. They so not know, think, or understand anything, point blank. LLMs and diffusion models are not AI.

they dont know anything they are just Markov chains on steroids.