Detoxification of Language Models

I don't know, to what extent Roy, are you following this notion of detoxification of language models? You know, because part of it, remember, as you pointed out earlier, opening, I didn't want to release the earlier versions of GPT because of possible negative applications. And now there's this research direction around detoxification. But as best I can tell, that's nascent. And also the results are somewhat mixed, right? Yeah, I mean, I must say, I'm not a big expert on a topic. So I prefer not to come into too much.