Even if all the big ideas are gone and we are entering a new part of the curve, there is still an enormous amount of improvement possible. Just iterating on dat

Even if all the big ideas are gone and we are entering a new part of the curve, there is still an enormous amount of improvement possible. Just iterating on data mix/quality etc, training pipelines, reward functions, specific ways of reasoning (which i guess is mostly just data still) for the next 20 years will yield a looooot. And that's just the models. The harnesses/application layers/whateveritgetscallednext space has 20 years of progress to make.