Intel and Phison are working to solve a problem plaguing local AI: you can't run the largest models. The solution -- run ...