Oddbean new post about | logout
 Friendly reminder the transformer model is only as good as the data it’s trained on