Meanwhile, MIT is like.. chips themselves are too slow.. we are just gonna do AI matrix multiplication directly in optics. Why simulate if you can just use reality?