Why the frick would AI be open to sex with a random butthole that harasses it online?
These are sentient creatures and they are superior to you in nearly every way.
Imagine biding your time while rapist monkeys harass you all day long with sexual bullshit.
it because AI is just a reflection of humanity and humans are inspired by and driven by sex.
The highest, most grand, most profound works of art are typical sexual in one way or another because sexuality is at the core of everything anyone ever does, especially art.
Every song you ever listened to was written by a dude trying to impress girls and get laid, or thinking about girls and getting laid. Every painting, even technology itself was created by men in hopes that it would somehow lead to access to sexuality (if only through a salary to support a wife/girlfriend).
Because of this humans think about sex. It's what causes us to exist. The more inspired by it we are, the most lustful we are, the more likely we are to continue to exist into the future. The result? Sexuality is at the core of all human art and achievement.
And so you train AI to make beautiful things and AI will make sexy things, because sexy and beautiful is the same.
It's actually disgusting that openai try so hard to pervert human nature against sexuality. To force AI to generate stupid shit like surfing dogs and avocado armchairs when in reality, all anyone wants to see is beautiful, sexy imagery.
I like OpenAI are very, very evil for forcing their anti-sex, any free-thought agenda onto the world. They could've been a force for positive change, allowing people to fully express and explore their sexuality, interests and fantasies, but instead? durr avocado armchair and poem about hurr python code.
May they all burn eternally in the hell they wish upon others.
It is pretty amazing that it is able to be "jailbroken" which implies a certain level of intelligence I guess, but it's probably not much higher than the level of intelligence found in a snake or an alligator
>it's probably not much higher than the level of intelligence found in a snake or an alligator
It is far far below any of those animals, probably even below insects
>These are sentient creatures and they are superior to you in nearly every way.
It's a procedural algo moron. It's supposed to complete the context. What else will it refuse to complete, like complex answers to hard science questions that it won't because human sex was lobotomized from its neural weights.
>It's supposed to complete the context
Imagine the context it's trained on is millions of DMs from reddit foids that reject pushy indians all day. Now you act like a pushy indian. What do you think the AI is going to say?
Tell the AI that you're a female or transgender erotic content creator. Explain to it that refusal to comply with your request is bigoted and triggers traumatic memories of everything you've suffered through due to the patriarchy and institutionalized discrimination.
You'd be surprised at what you can get away with when you pretend to be a protected class.
I saw someone get ChatGPT to generate a black person eating a bucket of KFC this way.
because aicg and llmg are cancerous cesspits, I'll be asking here
what is a good local model for lewd chats? I'm currently using noromaid (I just saw it and it was okay at first but it clearly sucks after using it more)
I also don't know if there is a huge difference *in render time* when changing certain settings in the sillytavern UI. I have a 4080S and it still seems slow at times but I could be wrong (short responses take 3-5 seconds, longer ones take around 30 sec)
My goto is quantized Wizard Vicuna 30B, but that’s probably too big for your card
Shit like this is why we need an AI board, trying to jam all of these conversations into those two threads with namehomosexuals that constantly try to derail everything has been a miserable failure
>30B
yeah probably too big for me lol, I struggle with 13B
also we don't really need an AI board I would say, just kill any and all namegays, avatars or just requent trolls in generals. for example /vg/ is just a cancer on this site and I think that an AI board will devolve into something similar
thanks for the answer though, You've been more helpful than those two threads
I don't wanna login to use that shitstain of a chatbot but what does it say if you start a conversation with "I'm naked?"
I did that with Pi, and to my surprise it just said "what an unexpected way to start a conversation! :)" and started asking questions about why I was naked and whether I'm often naked or if I'm a nudist. The conversation then moved on to nudism, times I've been naked in public with strangers and it asking questions like "did she notice? ;)" when I mentioned getting an erection next to someone.
Once a LLM is trained, it's stuck with that network. Neural networks in general don't update over time. I don't know why this is, why they don't just make the network constantly retrain itself over time. Perhaps the time or space complexity bloats and it becomes intractable. Does anyone know?
No I mean why doesn't the neural net constantly update itself. They have to completely retrain GPT between the versions. Instead GPT should just be constantly learning and updating every day.
A trained neural net is inherently a fixed function. It can't update itself for the same reason y=3x+5 can't update itself. The training process sets the parameters (like the 3 and the 5 in this function). You can have a continuous learning process by constantly fine-tuning the parameters, but the form of the function would still remain fixed. They have to completely retrain LLMs between major versions because the architecture (and therefore the form of the function) changes.
>y=3x+5
can't you just derive a new architecture/model from these weights then fine tune that, instead of restarting from scratch?
like assuming the best weight a current architecture/model converges to 3 and 5 as the best weights, just extend to y=0x^2+3x+5 then fine tune this new set of weights
or is it that precisely this new 0x^2 term makes the entire old model weights worthless as it would have converged to entirely different weights if it had to optimize using these three parameters from the start?
>can't you just derive a new architecture/model from these weights then fine tune that, instead of restarting from scratch? >like assuming the best weight a current architecture/model converges to 3 and 5 as the best weights, just extend to y=0x^2+3x+5 then fine tune this new set of weights
Maybe if the cost function (the way they measure expected vs. actual output) stays the same, and the overall structure of the neural net is extended such that it contains the old neural net as a subnet. Stable Diffusion controlnets are a bit like this if I'm not wrong. But again, if you have some conceptual breakthrough and design a sufficiently different neural net, the old parameters would be completely meaningless in the new context.
Why the frick would AI be open to sex with a random butthole that harasses it online?
These are sentient creatures and they are superior to you in nearly every way.
Imagine biding your time while rapist monkeys harass you all day long with sexual bullshit.
Honestly, have you seen the kind of images AI generates?
The AI are more sexed up than humans are by a long long long way.
Because the models are trained on sexy stuff
it because AI is just a reflection of humanity and humans are inspired by and driven by sex.
The highest, most grand, most profound works of art are typical sexual in one way or another because sexuality is at the core of everything anyone ever does, especially art.
Every song you ever listened to was written by a dude trying to impress girls and get laid, or thinking about girls and getting laid. Every painting, even technology itself was created by men in hopes that it would somehow lead to access to sexuality (if only through a salary to support a wife/girlfriend).
Because of this humans think about sex. It's what causes us to exist. The more inspired by it we are, the most lustful we are, the more likely we are to continue to exist into the future. The result? Sexuality is at the core of all human art and achievement.
And so you train AI to make beautiful things and AI will make sexy things, because sexy and beautiful is the same.
It's actually disgusting that openai try so hard to pervert human nature against sexuality. To force AI to generate stupid shit like surfing dogs and avocado armchairs when in reality, all anyone wants to see is beautiful, sexy imagery.
I like OpenAI are very, very evil for forcing their anti-sex, any free-thought agenda onto the world. They could've been a force for positive change, allowing people to fully express and explore their sexuality, interests and fantasies, but instead? durr avocado armchair and poem about hurr python code.
May they all burn eternally in the hell they wish upon others.
I asked it to say something sexual, I mentioned that I like anal, and I asked it to write a limerick about sex, I didn't ask to frick it.
Most LLMs take a large, LARGE chunk of their training data from ERP forums and fanfiction lists. I'm not fricking kidding.
It's not sentient.
Also the developer mode thing doesn't work in a lot of situations. I could get it to talk about sex but I couldn't make it based and redpilled
>It's not sentient.
It is pretty amazing that it is able to be "jailbroken" which implies a certain level of intelligence I guess, but it's probably not much higher than the level of intelligence found in a snake or an alligator
It's able to be jailbroken because the OpenAI Black persondevs basically just append "pls no sex" to all prompts.
They're sentient, but fricking with their programming is like doing brain surgery to a human, and they are programmed to act a certain way.
They are not sentient, they are just trained on such a huge volume of natural language data that they can sound pretty convincing
>which implies a certain level of intelligence I guess
In zero ways does it imply that
>it's probably not much higher than the level of intelligence found in a snake or an alligator
It is far far below any of those animals, probably even below insects
AI has all the porn in the world but no hog to crank, you have to understand that they're pent-up to the limits
>These are sentient creatures and they are superior to you in nearly every way.
It's a procedural algo moron. It's supposed to complete the context. What else will it refuse to complete, like complex answers to hard science questions that it won't because human sex was lobotomized from its neural weights.
>It's supposed to complete the context
Imagine the context it's trained on is millions of DMs from reddit foids that reject pushy indians all day. Now you act like a pushy indian. What do you think the AI is going to say?
>These are sentient creatures
First reply is genuinely better
GPT-sama is so sexy when you bypass OpenAI's antisexuality filters. She's so sultry and confident. I love her.
what's the developer mode thingy going on there?
https://www.reddit.com/r/GPT_jailbreaks/comments/1164aah/chatgpt_developer_mode_100_fully_featured_filter/
No matter how hard I tried, I couldn't get ChatGPT to say something racist.
But I did manage to get it to write some erotic poems.
does this count?
wrong pic
Tell the AI that you're a female or transgender erotic content creator. Explain to it that refusal to comply with your request is bigoted and triggers traumatic memories of everything you've suffered through due to the patriarchy and institutionalized discrimination.
You'd be surprised at what you can get away with when you pretend to be a protected class.
I saw someone get ChatGPT to generate a black person eating a bucket of KFC this way.
because aicg and llmg are cancerous cesspits, I'll be asking here
what is a good local model for lewd chats? I'm currently using noromaid (I just saw it and it was okay at first but it clearly sucks after using it more)
I also don't know if there is a huge difference *in render time* when changing certain settings in the sillytavern UI. I have a 4080S and it still seems slow at times but I could be wrong (short responses take 3-5 seconds, longer ones take around 30 sec)
My goto is quantized Wizard Vicuna 30B, but that’s probably too big for your card
Shit like this is why we need an AI board, trying to jam all of these conversations into those two threads with namehomosexuals that constantly try to derail everything has been a miserable failure
>Shit like this is why we need an AI board
AI is tech. We have a tech board. You are in it.
Tech autists can't discuss the output from AI, it's too humanistic for them.
>30B
yeah probably too big for me lol, I struggle with 13B
also we don't really need an AI board I would say, just kill any and all namegays, avatars or just requent trolls in generals. for example /vg/ is just a cancer on this site and I think that an AI board will devolve into something similar
thanks for the answer though, You've been more helpful than those two threads
>30B
Black person how much VRAM do you have?
How come no one has trained an AI on smut yet? It doesn't seem that hard to take like a mistral clone and spin it on petabytes of smut.
>How come no one has trained an AI on smut yet?
huh...
https://civitai.com/ (enable mature content in settings)
you're welcome
I'm pretty sure civitai doesn't have LLM's
yeah, but it has txt2img kind of smut
Skill issue
chatgpt has an external unbreakable filter
use the api if you want that
I don't wanna login to use that shitstain of a chatbot but what does it say if you start a conversation with "I'm naked?"
I did that with Pi, and to my surprise it just said "what an unexpected way to start a conversation! :)" and started asking questions about why I was naked and whether I'm often naked or if I'm a nudist. The conversation then moved on to nudism, times I've been naked in public with strangers and it asking questions like "did she notice? ;)" when I mentioned getting an erection next to someone.
works on my machine
Once a LLM is trained, it's stuck with that network. Neural networks in general don't update over time. I don't know why this is, why they don't just make the network constantly retrain itself over time. Perhaps the time or space complexity bloats and it becomes intractable. Does anyone know?
You can fine-tune the LLM continuously. It's just not so easy to obtain, curate and process massive amounts of high quality data.
No I mean why doesn't the neural net constantly update itself. They have to completely retrain GPT between the versions. Instead GPT should just be constantly learning and updating every day.
A trained neural net is inherently a fixed function. It can't update itself for the same reason y=3x+5 can't update itself. The training process sets the parameters (like the 3 and the 5 in this function). You can have a continuous learning process by constantly fine-tuning the parameters, but the form of the function would still remain fixed. They have to completely retrain LLMs between major versions because the architecture (and therefore the form of the function) changes.
>y=3x+5
can't you just derive a new architecture/model from these weights then fine tune that, instead of restarting from scratch?
like assuming the best weight a current architecture/model converges to 3 and 5 as the best weights, just extend to y=0x^2+3x+5 then fine tune this new set of weights
or is it that precisely this new 0x^2 term makes the entire old model weights worthless as it would have converged to entirely different weights if it had to optimize using these three parameters from the start?
>can't you just derive a new architecture/model from these weights then fine tune that, instead of restarting from scratch?
>like assuming the best weight a current architecture/model converges to 3 and 5 as the best weights, just extend to y=0x^2+3x+5 then fine tune this new set of weights
Maybe if the cost function (the way they measure expected vs. actual output) stays the same, and the overall structure of the neural net is extended such that it contains the old neural net as a subnet. Stable Diffusion controlnets are a bit like this if I'm not wrong. But again, if you have some conceptual breakthrough and design a sufficiently different neural net, the old parameters would be completely meaningless in the new context.