LLMs do better with more compute, but it's logarithmic.

They raced up the soft part of the curve, thinking it was the start of an exponential.

I think that's the core issue; they bet a trillion dollars expecting compounding returns, and got diminishing ones.

If I were an employee/shareholder, I'd have been happy to hear the first part of his message ("oh good he finally gets it") and then miserable again when he says he does believe the singularity will happen, just in 3-6 months.