Comment on Arrows of Time for Large Language ModelsparentComments−frotaur2yIn the paper it is mentioned that the LLMs predicting the previous token are indeed pre-trained in this way, so it is not true that the difference is obvious.
Comments
In the paper it is mentioned that the LLMs predicting the previous token are indeed pre-trained in this way, so it is not true that the difference is obvious.