Until any FSD solution delivers, it's far too early to vindicate the mobileye CEO. As it stands, even with additional (expensive) sensors, it seems very likely that ML will be required regardless. Running purely off sensors is not going to be feasible when braking distances alone are greater than the sensors will be able to accurately measure.
He did give a second talk a few years later explaining how the purely cognitive aspect may be handled by ML, and the total system probably going to end up being a mix.
But his fundamental opposition to ML for things related to safety was, IMHO absolutely correct.
