i don't like how they train the models. the cost of capable hardware is falling fast now and we can decide the agenda in the parameters, and not be subject to and victim of the policies they inflict on us. what this is gonna mean, also, is that the genie is out of the bottle. poor people with small budgets will be far more creative than hyperscalers with one good idea.

It's easy to judge, tho, now that the large models exist and we can distill them.

Yeah. I actually get a lot out of a plain-ole Codestral chatbot from self-hosted Ollama.