Don't believe - That LLMs alone would be enough to reach AGI.
But what do you believe about agent harnesses? About AIs that have tools that they can call that can do tasks that the AI, by itself, is not good at? Is that enough of "not LLM alone"? I don't know.
They are mostly here to stay.
Still these will need to adapt to the rising pace of acceleration in innovation. There's something new ALWAYS/DAILY/Soon (every second).
Diffusion models look interesting and strange at the same time. But they can't work with tools yet.