This is pretty cool. My bet is that we have more LLMs running locally when its possible, either thru "better hardware as default" or some new tech that can run the models on commodity hardware (like apple silicon / equivalent PC setup).