Dylan Patel and a co-host discuss whether current AI models might already be approaching AGI. They compare different model generations like Llama 3 and reference upcoming architectures like Blackwell and Reuben, speculating casually about how long today's models will remain competitive.
Chapters
I think more than once I had this thought in my mind. Like if someone tells me that this is AGI, I think it wouldn't be something, a statement that I have like a strong pushback against because it's like, so it's gotten so good. And to think that it's still on like current generation hardware before like the, you know, Blackwell vintage, Reuben vintage models to look forward to before they come in. It's frankly like crazy to think about. Vintage, man.
That's such a good word for these models trained on these things. I haven't heard of the— Yeah, yeah, yeah. I'm gonna put these weights aside from this 2025 vintage on that BF-16 Hoffer model and let it age for a little bit. See, yeah, see what Llama 3 tokens taste like in a few years' time after they've had some time to, to age, you know? I love that.