> I think cloud hosted models will probably always be far better for most tasks It might depend on just how good you need it to be. There are lots of use-cases

> I think cloud hosted models will probably always be far better for most tasks It might depend on just how good you need it to be. There are lots of use-cases where an LLM like GPT 3.5 might be "good enough" such that a better model won't be so noticeable. Cloud models will likely have the advantage of being more cutting-edge, but running "good enough" models locally will probably be more economical.