I hope this means AI-accelerated frameworks get better support on Mx. Unified memory and Metal are a pretty good alternative for local deep learning development.