I’m glad that your spider clean-up efforts have a positive effect on the CV!
Unfortunately, with bees, it seems like each CV update only means that it is making different wrong suggestions than the ones it did before. It is incredibly frustrating, precisely because my efforts seem to have very little effect – I can correct a hundred IDs today and next month I will still be doing the same thing because the proportion of wrong IDs will not have noticeably decreased.
The wrong CV suggestions are exacerbated by user error (e.g. the first broader CV suggestion is more likely to be correct but users will often pick something wrong from the more specific lower-confidence suggestions instead), but here, too, there is the problem that if I educate one user about why they should be cautious about using the CV suggestions for bees or even give them pointers on general bee identification, next week another new user will come along and I have to do it all over again. And it seems to be increasingly common to get users who are using iNat for some pollinator project and manage to upload a few hundred observations with bad CV IDs before they even see the feedback from IDers asking them to crop their photos, put all photos of the same individual in a single observation, be critical of the CV suggestions, etc.
So changes to the CV training that would make it better at handling difficult taxa would be at the top of my list of things that would make IDing a more rewarding experience. As would better onboarding for users to help them make better observations and use the CV suggestions in a more conscious, thoughtful manner.