Distillation works: we can get models leaner, better performance with lesser and less parameters, more efficient. I think Apple gets it too, and we'll see the result with M7 Max/ultra chips and I reckon this will be one of the titular apps when M7 lines of chips are introduced.
I guess it explains the haste Anthropic and OpenAI are in to IPO, cash out and leave the the retail investor holding the bag.