Remember Andrew Ngs theses that current architectures are already "AI complete" meaning that given more training data all models outperform other models with le

Remember Andrew Ngs theses that current architectures are already "AI complete" meaning that given more training data all models outperform other models with less training data, and a model which seems a bit more efficient with some amount of training data can be less efficient with more data or the other way around.