a case @masnick.com is making is that returns to scale on LLMs may top out, that beyond a certain level, the difference between a managed-at-scale frontier model and a self-hostable open model may be small enough that the latter is good enough for most users and applications. mb wrong. but possible!