Yeah if people want to use LLMs long term, they better learn using smaller models and less tokens, because once investors stop subsidizing tokens, those who can't maximize results will be fucked.