Title says it all. Why spend so much effort finetuning and serving models locally when any closed-source model will do the same for cheaper in the long run. Is it a philosophical argument? (As in freedom vs free beer) Or are there practical cases where a local model does better.
Where I’m coming from is the requirement of a copilot, primarily for code but maybe for automating personal tasks as well, and wondering whether to put down the $20/mo for GPT4 or roll out my own personal assistant and run it locally (have an M2 max, compute wouldn’t be a huge issue)
https://old.reddit.com/r/LocalLLaMA/comments/17ldyak/why_are_you_running_local_models_what_are_you/
https://old.reddit.com/r/LocalLLaMA/comments/13b8ij7/why_run_llms_locally/