[jejeyyy77]

Actually the improvement graphs are still scaling exponentially with training/compute being the bottleneck. So there isn't yet any evidence of diminishing returns.

source: https://youtu.be/zjkBMFhNj_g?feature=shared&t=1545