I'm saying they're not an advancement in the tech in the way GPT 1 through 3 were. They're a different kind of improvement.
And as such the rate improvement cannot just be extrapolated into the future.
All interesting conceptual breakthroughs came after GPT3: RL and reasoning being the main ones.