State of AI in 2026: LLMs, Coding, Scaling Laws, China, Agents, GPUs, AGI | Lex Fridman Podcast #490
47:13of model performance degradation?
But I do think... I mean, there are alternatives popping up to the
transformer. There are text diffusion models—47:21a completely different paradigm. And
there is also... I mean, although text diffusion models might use
transformer architectures, it's not an2:29:02diffusion models—just a different
exploration of text diffusion. Can you talk about what that is and
the possibility it holds?









