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

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