There’s talk about Microsoft SoCs on their own products, much like Apple does the M1 SoCs.
These Microsoft SoCs would be used in Surface devices and likely have dedicated AI hardware. Again, much like Apple.
If we’re talking about specialized models, not one generic LLM for everything a la GPT4, they might not have to be THAT big and could run on reasonably powerful devices.
I really doubt that, at least for the next few years. “AI Assistant” usually means LLMs, and even M2 struggles to run them mostly due to large compute and RAM requirements. If Microsoft could somehow release a truly local AI assistant feature that can run on average windows users’ hardware, that would be shake the whole ML industry.
True, but they could get the base requirements of a task using OpenAI and then use specialized models locally to do subtasks.
Microsoft owns 49% of OpenAI, they don’t need to pay nearly as much per request as we do and the cost will likely decrease over time too.
