I doubt facial recognition is a switch turned "on", rather its vision capabilities are advanced enough that it can recognize famous faces. Why would they build in a separate facial recognition algorithm? Seems to go against the whole ethos of a single large multi-modal model that many of these companies are trying to build.