I think most of the latency issues will improve as token generation does as will accuracy. The device is just a conduit, it represents the state of LLMs today,

I think most of the latency issues will improve as token generation does as will accuracy. The device is just a conduit, it represents the state of LLMs today, you can't ask for more from a I/O device.