Comment by AnthonyMouse
9 days ago
I've made a significant number of edits to Wikipedia over the years. I probably have an account but generally don't even bother to sign in because I don't care about credit or a dynamic IP that will change in a week being recorded in the edit history, which they've apparently stopped doing anyway.
My most recent edit (a minor addition to a technical article) was instantaneously reverted as "suspected vandalism" by a bot, an unambiguous false positive. The bot seemed to think I was going to follow its instructions if I thought it was a false positive instead of finding that irritating and concluding that I should stop making edits if having them actually go through requires me to fight with a broken AI.
You don't have to report the false positive, the link to the place to do it is just included in explanatory edit summary so that you can conveniently use it if you want to. (The reported ones eventually are reviewed by multiple other editors, and then, true or false, are included in the training data to improve the accuracy). The retrained bot is measured against the human-verified vandalism and non-vandalism data so that the bot is expected to generate 1% false positives of all the reverts it does.
By the way, the bot will only revert an edit once, so you can undo that revert and the edit goes back in (at least until a human editor decides it should be reverted). The bot has available to it not just the change text and its placement in the existing article text, but also meta information such as the editor's account information (and I believe logged-out edits happen to get dinged more often simply because those are the major source of vandalizing edits).