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

[Daniel Kokotajlo]

(In terms of the neural network model, this means lowering our estimate for how many parameters will be needed.)