What makes GPT4 AGI that makes GPT3.5 not AGI? And what makes GPT3.5 AGI that makes GPT3 not AGI? And what makes GPT3 AGI that makes GPT2/smaller models not AGI?
What is the "hard line" that makes something AGI or not AGI? Because IMO it looks like GPT4 is somewhat AGI, but also the older models possibly all the way down to even Markov chains: it's just that this AGI is nowhere near human-level.
Abstraction and reflection have been posited in the past as pre-requisties for intelligence. "I am the thinker that is thinking." How do we prove whether an AI has this capability or not? I'd say it's nearly impossible.
However, I think we can certainly prove when it doesn't. For instance, the fact we need an external plugin to get the model itself to return text claiming 1+1=2, tells me that GPT4 cannot reason about numbers in the abstract, and therefore lacks abstraction ability.
That's a rather interesting line to me, as someone with a young child, they cannot perform that abstract reasoning on mathematics either. At the same time, I feel extremely confident they're a thinking and intelligent being.
I think we're so strongly biased against a deeply uncomfortable reality that there may not be a hard line, we don't even want to consider the alternative.
Models have trouble with discrete spaces often, partly because of the internal continuous space representations, but also in this context because the transition from the probabilities in natural language to mathematics may not be as stark as it should be.
But to make matters worse, or more muddied, 1+1=1 can be a valid mathematical statement. It simply depends upon the set and group you have, or if you're doing modular arithmetic, etc. Sometimes you're given a unital magma. So, there's still a heavy dependency on context for the problem setup, but the underlying discrete and deterministic rules that are applied to the context is less malleable than other context switches in NLP LLMs do well in (such as language styling).
The inability to fully define a thing doesn't invalidate all attempts to set its outline. At least, it's easy to conclude that an intelligent being has to be able to reliably perform basic reasoning (given all the necessary information is properly acquired). The current GPT models all fail at this, and neither the token length nor the network size can fix this.
We don't exactly know what intelligence means and if we are always intelligent.
An example about ping pong players. The pace of their movements is too fast for conscious thinking, so it's all trained reflexes with some overall strategic planning trying to keep up with events. There is no time to think about anything. Is intelligence suspended there, at least the general one? Then the same person stops playing and gives an interview about the game and full general intelligence turns on again.
I’d say that any technical limitation that doesn’t apply to humans is not AGI. Context windows are the most apparent ones; humans don’t have a stroke after reading N characters.
AGI is not human intelligence. Cats are generally intelligent for example. A cat level AGI is worth billions.
Humans certainly have context windows. Try asking your CEO about some lines of code in your work. Humans have a fairly large one, I give you that and it is fuzzy.
Well if it walks kind of like a duck, and quacks almost like a duck.. it may be a prototype robotic duck but it's still a duck.
It's pretty intelligent and rather general too, so at least by the definition that doesn't include mandatory consciousness it would mostly fit. And consciousness is pointless for a robotic system because it doesn't add anything practically useful. Just because agency has to be provided by the user doesn't make it any less of an AGI I'd say.
