Also consider this: what would be cool is to enhance scalability of local llms by exchanging some latency of the hardware level, and makes it more powerful local llm for different usecases. But when will such a usecase arise? Isn’t our data shared across these foundation model’s companies a huge concern?