I keep having problems with the automatic labelling of species with words pulled from the names of photo files as well, mostly when I label something as “orange bug” or “orange mushroom,” in which case it gets labelled as Citrus x aurantium (=orange, the fruit). But the oddest one I’ve seen so far was this:
Somehow there’s a hash table that defines “skunk” not as Mephitis mephitis or its relatives but as a synonym for cannabis. (Does “weed” in file names also yield species labelled as Cannabis sativa?) Anyway, if the AI drew from both the name of the file and the content of the image, this problem would probably not exist – at least not in the same way.