4 ms·
Faces can’t be easily be “faked” in person but they can definitely be misrecognized or recognized correctly against mistagged data. Anything like this needs to
by acdha 1y ago
Faces can’t be easily be “faked” in person but they can definitely be misrecognized or recognized correctly against mistagged data. Anything like this needs to be designed with the assumption that errors will be frequent enough that the finding has to be validated before doing something serious.