That's hard to prove one way or the other. Watch Ilyas' latest talk where he talks about how next token prediction is a much more fundamental problem than peopl

That's hard to prove one way or the other. Watch Ilyas' latest talk where he talks about how next token prediction is a much more fundamental problem than people give it credit for. Empirically you can easily see that gpt-4 has a world model and that it does abstract logical reasoning. Maybe it's a vastly different intelligence than our own but the fact remains that it can perform well on an extremely broad range of multidisciplinary tasks. You can argue the epistemology of this all day. But as a pragmatic programmer, gpt-4 does certainly seem to fall into the "AGI" category.