has the possibility of using an LLM to screen posts been considered yet?

taking the global rules and providing them as context for either a cloud based (expensive? prices are coming down quickly though) or locally based (llama or mixtral8b? cheaper, buy a 5090 or something lol and run it there) LLM, then providing it instructions to rate posts based on rulebreaking probability could add another method to get posts flagged to moderation staff without relying solely on user reports. posts above a certain rule breaking threshold as determined by the model could be added to either the report queue or a new queue.

thoughts? apologies if this is a dupe of topics already discussed