Post

Louder for those at the back: once you start to think about LLM outputs as optimising for *plausibility*, it's easier to put into context what LLMs can/can't do and how to use them properly. #databs #datasky