I'm not sure what you mean by "everyone knew what AI meant"? It's always been an extremely vague term, since its inception — and LLMs do, actually, enable computers to do things I, at least, would've called basically AGI in 2020.
There's nothing vague about it. Intelligence is understanding. I think you're just motivated to be confused because you want to fit this thing into that box and it doesn't really fit. So maybe output is an that matters, are we auto completers, can you be intelligent without knowing, etc etc.
I'm sorry, but very many systems have been called AI that don't have "understanding," and been very useful without that. I'm not sure when that was ever part of the definition? Even Wikipedia gives a definition similar to my own, and I think Karen Hao makes a very good historical case that the definition of AI has never been super clear in Empires of AI, and she's no fan of LLMs (or the field of AI in general, it seems), that there never really has particularly been a clear definition.
> Everyone knew what AI meant before LLMs came along
Please do tell.
> Why is everyone on the ground floor trying to defend what the guy in the penthouse is telling the investing public?
I'm talking about observable capabilities of the technology. There are facts of the matter there that have nothing to do with what tech CEOs say. Pointing out the reality of the situation is not "defending" anyone.
If you take a "team sports" approach to reality where truth is a function of which side you're on, then you're not going to able to participate meaningfully in rational discussion.
There's rational discussion and then there's motivated obfuscation. If your definition of AI is so loose that it applies to a system that can't count objects reliabily, then what is there to discuss? Is a broken clock intelligent because it tells the time correctly twice a day?
