-
@jessefuchs @metasynthie if it's trained specifically, then it's like, do an extra layer of training on top of the Internet-training to skew towards the stuff they've written before. and that stuff can be more interesting, tbh! still got a lot of the other concerns, tho.Permalink Look on archive.org
❤️ 2 Favorites Mood +6 🙂
-
@jessefuchs @metasynthie but also a lot of stuff that promises that kind of thing just cues it within the prompt. given this, what's likely to come next.Permalink Look on archive.org
-
@jessefuchs @metasynthie I'd also say that it is a lot less dangerous in this scenario than when you plug an end user directly into the LLM. especially if you wanna accomplish a task with it, not just... make some text of variable quality.Look on archive.org