What is going to happen is that groups of people will pool resources to fine tune released LLMs on specific data sets, then release low rank approximations that

What is going to happen is that groups of people will pool resources to fine tune released LLMs on specific data sets, then release low rank approximations that can be blended into a base model. This is already happening in Stable Diffusion land and it works great, I'm sure we'll see it with LLaMa soon.