Hi @courtneyb, the short answer is that you are fixing it (and other contributors) by patiently submitting records for it to make mistakes with. Like other AI programs, Carbon learns from its mistakes, when moderators correct them. Some things are quick but depending how visually similar the alternatives are, it can take an awful lot of images to train AI. IMO, Carbon has performed astonishingly well with commonly photographed things like wombats, echidnas, macropods, rabbits, rakali, platypus, brush tailed and ring tailed possums, etc, so I don't really have to look at those any more. But gliders will take more time, and Yellow Bellied may need a decade or more. Please just keep sending lots of records!
@DonFletcher thanks for fixing this one and for the info! It was interesting that CarbonAI was confident enough to confirm this one rather than suggest (usually it’s the latter for the GG sitings).
@courtneyb yes that is a concern and I'm surprised that it was verified. I understand the trigger for verification is 99% correct, but I have corrected at least a dozen GGs misidentified as P. notatus, so Carbon seems far from ready to be confirming P notatus. @AaronClausen how say you?