Daniel Kokotajlo comments on [Link] Training Compute-Optimal Large Language Models

[Daniel Kokotajlo]

Don’t you mean the dataset size was much too large for the smaller models and maybe too small for the largest models?