it's literally a NeurIPS paper they train classifiers on different-sized datasets, different-sized models and fit power laws can't believe this was 32 years ago

→ View original post on X — @jxmnop