Is there an equivalent to Moore’s Law for language models? It feels like every week an even bigger (and supposedly better) model is announced.