grump
kiwifarms.net
- Joined
- Apr 14, 2016
Bard is quite the fence sitter
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
I think that'd be a scenario where three letter agencies all over the world are allowed to indiscriminately make people and entire buildings disappear.Anyone with actual knowledge that OpenAI was capable of breaking AES256 would have the three-letter agencies far up their ass. They wouldn't be anonymously leaking things online.
They pretty much admitted to that after researchers started calling them out and then journos picked it up. They put out a blog on how it works, but basically they just fed it text and still images like any other model, then edited responses to make it seem like it could respond to voice prompts and in real-time.I think googles last video of its new AI with camera functionality was fucking baloney though.
AI reacts too fast for it to be real I believe.
Yeah I saw this and thought I'd play with it. Just got it installed so haven't really done too much with it, seems to be pretty decent thoughDudes this is Mixtral uncensored. My computer is a potato and I currently don't have the GB, but has anyone done this. This is the Code Report by Fireship. Please tell me if it's worth it!
If you can post something funny or spicy. I'll get it when I upgrade my rig, but in the meantime I'll have to wait. Have fun dude!Yeah I saw this and thought I'd play with it. Just got it installed so haven't really done too much with it, seems to be pretty decent though
Spouting completely incorrect information and being ridiculously confident about it? Damn, ChatGPT turned into a Redditor.I know some models are simply not suited to this kind of thing, but the confident way it asserted something so obviously wrong in this case was really funny to me.
Heres a bunch of stuff using the mixtral-dolphin model:If you can post something funny or spicy. I'll get it when I upgrade my rig, but in the meantime I'll have to wait. Have fun dude!
Good stuff man! yeah, the "kiwifarmers are gay" or "reddit is gay" is still a bit "aligned" but perhaps it's just the prompt. The letters to Cogent come out good, but a bit verbose. I do like the "yours in eternal disdain" part. That made me lol.Heres a bunch of stuff using the mixtral-dolphin model:
Yeah some things it won't outright refuse but it seems to have a biased response though that might just be the data it was trained with.Good stuff man! yeah, the "kiwifarmers are gay" or "reddit is gay" is still a bit "aligned" but perhaps it's just the prompt. The letters to Cogent come out good, but a bit verbose. I do like the "yours in eternal disdain" part. That made me lol.
Perhaps putting into the prompts more specific insults may work? I dunno, for example "make the letter 'niggerfaggot' word at least once and call cogent 'censorious troons' at one point". It seems promising tho
counter argument: journalists are sub human and thus dont deserve this treatmentAI people "need to start asking creators – of journalism, of art, of code – for permission to use their works, just like anybody else needs to do."