People keep mistaking Mojo as good syntax for writing GPU code, and so imagine Nvidia's Python frameworks already do that. But... would CuTile work on AMD GPUs and Apple Silicon? Whatever Nvidia does will still have vendor lock-in.