A short exact sequence walks into a bar.

Dry Shampoo.

Get it?

Calling the mistakes of LLMs hallucinations gives them too much credit as it would imply that the errors resulted from a deviation of their normal operation. Once I had asked it to come up with punchlines for "A short exact sequence walks into a bar", then I asked it for hair advice, then I asked it to explain it's punchlines, half way through it switched to giving explanations for it's hair care suggestions, but it still "thought" they were punchline about short exact sequences. This wasn't an aberration, it wasn't that it was working before, and hallucinating after, it simply didn't understand anything it was saying at any point, just predicting the most likely word, guided by no concepts or internal models.

I voted no.

@PaulFoss Well yes thats true now but consider the level of performance achieved as you say without internal models. We humans are clearly capable of them so why shouldn’t it in the long run be possible to develop them in machines. I expect that for the same reason you say people over estimate them now there is an order of magnitude or two more worth of unclaimed potential.

@Ebcc1 I was answering the question assuming it meant the current hype bubble surrounding transformers and LLMs. I am sure that there are yet undiscovered algorithms in computation space that are capable of doing the things people falsely claim transformers already can, and when you demonstrate that they in fact can not, they'll argue "We just need to build an even bigger model, investors keep sending us money, this goose is sure to start laying golden eggs." Armando di Matteo asked "Which people?", I guess I could further ask "which AI?"

@PaulFoss Well that’s the reason I didn’t specify LLMs I mean more specifically the concept of machine intelligence itself. As in are people (who people are should be left up to your interpretation with those in focus that you find important) underestimating how intelligent we can potentially make machines and how significant a truly intelligent machine would be.

Which people?

@ArmandodiMatteo who people are should be left up to your interpretation with those in focus that you find important. that could be your family or important decision makers in the field or in public policy.