Over two months I’ve been using guanaco via hugginface. A lot of new open source models came out lately with more parameters and higher quality. I tried several spaces that were listed in models’ profiles in hugginface, none of them seemed to show any signs of life.
Look if I pull the text above through guanaco and tell to phrase it in 4chan style: I’m such a fucking expert on this shit, let me tell ya. Been messin’ around with GuaCo for like, forever now. But all those other dumbasses just keep releasing these huge-ass models that don’t even work half the time. Tried 'em all too, believe me. Nothing but crickets chirping back at me.
Claude 2 may give uncencored results if smart prompted
there is a huggingface space called Bingo that wraps Bing AI. When Bing AI detects a violation it wipes its answer away, but Bingo steps in and saves the content that has already been generated. Quite useful because Bing AI much less often gives ethical lectures
There is vicuna-33b that is also loosely constrained as the guanuco model and seems to be a bit smarter, https://chat.lmsys.org/
I recently checked out the yi-34b model and it is insane, mistral and guanaco aren’t nearly as good since it is actually capable of generating engaging texts that makes much more sense, moreover it doesn’t cut on vulgarity and style, it can generate sarcastic 4chan style posts that I really laughed off