I think as people get increasingly aware of the nebulous consistency of hosted LLMs, local models will become more valuable