GPT-4 Trained on 10x More Data, Can Write 60k Word Books

by baptiste313on 1/11/23, 6:42 PMwith 1 comments
by minimaxiron 1/11/23, 6:56 PM

This misinformed GPT-4 take is new levels of comically bad because it correlates number of model hyperparameters with the quantity of training data.