Re:Inductive or Deductive? Rethinking the Fundamental Reason...
by godelski on Monday September 02, 2024 @11:05PM
(#41430855)
> AGI exists now
AGI exists if we redefine AGI to mean what we have now. But otherwise, no, it doesn't. What we have today does not meet the criteria that have been standard for the last half century[0]. I know people have a hard time keeping up, but no, the Turing Test is not the accepted threshold and hasn't really ever been. It was a starting point, and I should mention we beat it since at least 1967 (<20 years since Turings proposal). Nor is Chess, which we succeeded at in 1997.
What I hate about these conversations is that people learn the basics but there's been a lot of advancements since the 50's. We've learned a lot since the 90's too! Yes, the metrics keep changing, but they are not in response to the current state of the art in a specific effort to make the systems not AGI[1]. But rather the criteria keeps changing because we keep learning. It would be ludicrous if it didn't! Because WE STILL HAVE NOT CONCRETELY DEFINED CONSCIOUSNESS, REASONING, THINKING, and many other similar terms. The criteria are generally guides created by best guesses. But until we have formal definitions, expect it to change (and when we do have formal definitions, be sure as hell to understand their assumptions!).
But I can confidently tell you that LLMs do not exhibit generalization in the way we've been using it for decades. This is easy to demonstrate but if you're a true believer then there's no convincing. And in that case, just don't bother arguing either.
I really do encourage people to take their passion and use it to drive gaining a deep expertise on these subjects. You're passionate right? I hope that passion is not about "being right" or "having answers" and is actually about "understanding." And if you can't differentiate that, I suggest starting at the beginning, since these topics are covered there.
[0] Sure, you can find examples of people stating criteria, but I can find you samples of idiots, even among scientists.
[1] There are some bad faith people who do this but they're terrible people too. You may freely disregard their opinions.
What I hate about these conversations is that people learn the basics but there's been a lot of advancements since the 50's. We've learned a lot since the 90's too! Yes, the metrics keep changing, but they are not in response to the current state of the art in a specific effort to make the systems not AGI[1]. But rather the criteria keeps changing because we keep learning. It would be ludicrous if it didn't! Because WE STILL HAVE NOT CONCRETELY DEFINED CONSCIOUSNESS, REASONING, THINKING, and many other similar terms. The criteria are generally guides created by best guesses. But until we have formal definitions, expect it to change (and when we do have formal definitions, be sure as hell to understand their assumptions!).
But I can confidently tell you that LLMs do not exhibit generalization in the way we've been using it for decades. This is easy to demonstrate but if you're a true believer then there's no convincing. And in that case, just don't bother arguing either.
I really do encourage people to take their passion and use it to drive gaining a deep expertise on these subjects. You're passionate right? I hope that passion is not about "being right" or "having answers" and is actually about "understanding." And if you can't differentiate that, I suggest starting at the beginning, since these topics are covered there.
[0] Sure, you can find examples of people stating criteria, but I can find you samples of idiots, even among scientists.
[1] There are some bad faith people who do this but they're terrible people too. You may freely disregard their opinions.