Baby steps, but how useful is a 1B model these days? It seems actual domain specific usefulness (say specific programming language, translation, etc) starts at

Baby steps, but how useful is a 1B model these days? It seems actual domain specific usefulness (say specific programming language, translation, etc) starts at 3B models.