>You already said "only by training on far more data", which is different than "more parameters" being the only option. I never said more parameters was the on

>You already said "only by training on far more data", which is different than "more parameters" being the only option. I never said more parameters was the only way to increase performance. I said the training data required to reach any arbitrary performance x reduces with parameter size. It's literally right there in what I wrote. >a 50 billion parameter model will far outperform a 5 billion one TRAINED ON THE SAME DATA.