

I have no idea. For me it’s a “you recognize it when you see it” kinda thing. Normally I’m in favor of just measuring things with a clearly defined test or benchmark, but it is in the nature of large neural networks that they can be great at scoring on any desired benchmark while failing to be good at the underlying ability that the benchmark was supposed to test (overfitting). I know this sounds like a lazy answer, but it’s a very difficult question to define something based around generalizing and reacting to new challenges.
But whether LLMs do have “actual intelligence” or not was not my point. You can definitely make a case for claiming they do, even though I would disagree with that. My point was that calling them AIs instead of LLMs bypasses the entire discussion on their alleged intelligence as if it wasn’t up for debate. Which is misleading, especially to the general public.
Damn, she’s grown quickly!