3 ms·
Nowadays you can just use an LLM for translation and rewriting offensive or harassing messages and removing innuendo
by casey2 2y ago
Nowadays you can just use an LLM for translation and rewriting offensive or harassing messages and removing innuendo
- Cthulhu_ 2y agoIn theory, yes, in practice it sounds like it could be a very expensive content filter. Then again, a simplified and cheap / optimized single purpose model would work. Given how creative people can get and how many different languages etc there are, it'd be interesting to see. But it would need to be a model that not only knows all the bad words and intents, but also learns the workarounds, like how Roblox users have invented creative phrases like "go commit die", "yeetus yeetus commit self deletus", or "go commit cease vital functions necessary for the prolonging of one's physical being".