[seaucre]

This is interesting, and I have always thought this approach worth exploring given the "bitter lesson" in other ML domains, but I think we should be skeptical until we see such models deployed and operating effectively on real-world vehicles.