I love how the takeaway from this is apparently not anything about the actual point about AI safety, but pointers to places I get can hold of an LLM that's willing to be erotic.
this post was submitted on 23 Jan 2024
53 points (100.0% liked)
Technology
37800 readers
393 users here now
A nice place to discuss rumors, happenings, innovations, and challenges in the technology sphere. We also welcome discussions on the intersections of technology and society. If it’s technological news or discussion of technology, it probably belongs here.
Remember the overriding ethos on Beehaw: Be(e) Nice. Each user you encounter here is a person, and should be treated with kindness (even if they’re wrong, or use a Linux distro you don’t like). Personal attacks will not be tolerated.
Subcommunities on Beehaw:
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.
founded 2 years ago
MODERATORS
Well, pretty much only businesses and parents really worry if an AI is too lewd. For everyone else, it's probably a plus.
The world isn’t gonna end if ChatGPT is willing to talk with you about titties, no. The point is that it’s easy to deploy an AI system that’s capable, but real hard to deploy one that you’re sure will do exactly what you want it to. And, as they get more capable, that’s gonna be a bigger and bigger problem so we better be working on it now. Avoiding talking about sex things or criminal things or what have you is important partly because it’s a good specific way to be working on that.
I personally love the meme the other day of someone asking a Honda dealership chatbot to write python for them and it did perfectly.
These LLMs are really only good for big dogs like Google, Microsoft, and the like to put on their search engine for a general AI. It's going to be reeeeal difficult to limit to such a narrow range. I'd argue that they're better off using their existing "chat" bots
anyone wanna give me access to their compute cluster.. for science?
KoboldCPP now pretty much can run everywhere, even without a GPU. CPU only without Cuda on a 9th Gen i7 takes around 60 seconds to create a response, depending on complexity. That is actually an alright time for not hardware accellerated usage.
Run whatever model you wanna run, I won't judge. A lot of people use the software for mostly porn anyway alright.
You can set a prompt on Poe with GPT-4 that allows for lewd RP. Will still occasionally throw “I’m sorry, I can’t”s but you just gotta remind it to follow the prompt and stay in character after deleting the message.
I’m not posting the copy pasta for it. Easy enough to find.
Worth $20 dollars on a Privacy card to try out, IMHO.
Will still occasionally throw “I’m sorry, I can’t”s but you just gotta remind it to follow the prompt and stay in character
This sounds like a kink scene with bad consent
Yes, for the record: I am solely interested in the lewd stuff, nothing else.
Novel AI is uncensored and better than GPT2 and runs from $10-25/mo. Fair warning, it doesn't take long to plumb the depths of language to achieve "maximum fetish". It's like eating nothing but ice cream for breakfast lunch and dinner.
GPT-2 medium will run on a toaster and generate replies in a second
The True Story of How GPT-2 Became Maximally Lewd [then it no longer was]
Oh well.
I wonder about The True Story of How YouTube Video Titles Became Maximally Clickbait...
Do you know who Rob Miles is?
I don't even necessarily disagree with you, but having an awareness of who he is and his history of AI-safety advocacy was pretty effective at turning me from "wtf clickbait" into "oh cool Rob Miles is using clickbait now b/c it's more effective."
I mean the people in this thread definitely seem fascinated by the lewdness aspect 🙂. And obviously I clicked on it, too.
Oh, I missed that part, clicked on the thumbnail directly to see what BS was this going to be... and was only mildly disappointed.
But indeed, this being Rob Miles, changes things. I guess most of the video can be seen educational in disguise. A sad moment for YouTube video titles though, when even minimally decent stuff needs to hide behind clickbait.
Haha yeah. He released all these videos just talking about how it's a problem, and I can sort of picture him trying to work out, okay I tried just explaining the problem clearly, now how can we make the videos reach a wider audience. I know! (a) sex is involved in the story (b) cute little animations of the computers in a glowy neon style (c) clickbait title. Okay, check, let's do a/b/c then, and see how it works.
This video is recommended by Tournesol community:
[30🌻] Rational Animations: The True Story of How GPT-2 Became Maximally Lewd
#Tournesol is an open-source web tool made by a non profit organization, evaluating the overall quality of videos to fight against misinformation and dangerous content.