As for Google Photos, it makes mistakes, false positives mostly (as a funny example, for my photos so far it consistently thinks cats are dogs, although it also recognizes the same cats correctly) and sometimes it's too “eager” (“cars” will produce results where cars should be but are not actually in the photo). Sometimes it makes bizarrely trivial mistakes considering how smart it is generally: it categorized photos of human cadavers and anatomy specimens I made at the local university institute of anatomy as “Christmas”, I guess because they were made on Christmas day. But it will correctly find true positives. It also correctly visually recognizes notable places globally (that is, without geotags) and various things, concepts etc. even if they're a completely marginal subject in a photo. One thing that impressed me was a cat video it edited out of multiple clips with almost perfect music some 5-10 minutes after I uploaded the first items and started using it.