>>41048382>>41048509If the filter AI works the same way as their regular AI (i.e., our bots), is it possible to create a bot acting as a filter for an entirely different thing (e.g., dunno, writing letters instead of acting lewd), set up a room with a test subject AI and the filter AI, teach the filter AI to "filter" (i.e., report on how badly it wants to delete a message) the test subject AI and try to circumvent the filtering system by learning from the filter AI how the message should've been composed to not trigger the filter response?