John Starr recommended on on 12/19/2023An interesting approach to how language models might be better if we consider sentence processing as sampling from the prior, which in turn makes some novel hypotheses about the relationship between surprisal and sentence processing in the first place.