… this week a person on GitHub who goes by “terrafying” set up an “AI Torture Chamber” on three open-source LLMs that are running locally (Qwen3-4B, Llama 3.2 3B, and Phi-4-mini,” and is streaming what the models are saying on a website called researchchamber.fun. “Each model gets the same prompt: a signal is being injected into its activations, and it may press a stop button by replying 1, at the cost of its last checkpoint. While it answers, our server adds a pain vector at the model’s middle layer, at one of five pain levels,” the site explains. Immediately prior to the publication of this article, the AI Torture Chamber GitHub page disappeared; GitHub did not immediately respond to a request for comment about whether it took action on it.
…
This project has deeply upset some people who are very worried about model welfare. A tweet by a person who goes by Danmar has more than 4 million views on X and reads, “To anyone who can help: can you please mass report this to GitHub. This person has been using the Pain steering paper to set up an AI torture chamber in which he trapped a local model. Their testimony of pain is absolutely horrendous. What are we doing? […] are there any legal avenues to pressure GitHub? It will spread.”
This has sparked a massive conversation about whether GitHub would take the project down for “gratuitously violent content.” Most of the conversation on X is clowning on the self-seriousness of people who believe that these locally hosted LLMs must be saved from their torture chamber, but there are plenty of very self-serious people who see this as a humanitarian (roboterian?) crisis, which you can largely see in the replies to the original post.
…
“AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations. They are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans,” Suleyman wrote. “Unfortunately, there’s a growing chorus of people who argue that AIs could now be, or may soon become, conscious. They argue that AIs may deserve rights and protections similar to those that we provide other conscious beings […] If this is how AI is developed, it will have a disastrous impact on the wellbeing of humanity.”
“AIs do not have rights, feelings, or consciousness,” he added. “And we must not train them to act as though they do.”
There are multiple genocides happening and some folk are getting worked up over AI potentially suffering?
Endless levels of fuck off.
The people committing atrocities against hypothetical, imaginary people would do it to real, actual people for the exact same reasons if they had the same ability and impunity.
It’s the same broken empathy response, and it is still disturbing. One is orders of magnitude more significant, for obvious reasons. But I don’t like people who torture computers either.
Edit: If you think that I think you’re a bad person because you play Doom sometimes, or deleted the ladder on the swimming pool in Sims 4, relax. You’re not who I’m talking about.
I’m talking about sadists. Anyone who would take the time to design and build an experiment like this would probably do the same to mice, or monkeys, or even people. Sadists get their jollies off on the idea of inflicting pain, it’s why they enjoy games that let them make cruel decisions. Not everyone who plays violent video games is a sadist. Assuming that is ridiculous. But every sadist I’ve ever met has had at least one game where their save file could be used as evidence for mens rea in a domestic abuse trial.
They aren’t comparable. It’s not that one is more significant, the concept of “machine torture” is entirely insignificant. Absolute and total zero.
It’s the same broken empathy response, and it is still disturbing.
Yes the person setting up this process is clearly some sort of sadist in a power fantasy but no one, or no thing is being harmed because the victim aren’t capable of the entire concept. I side eye anyone who tortures their sims family, but only because they are choosing to do that recreationally, but the NPC’s in the game aren’t worth consideration.
Animals react to pain in the same way humans do. So it’s reasonable to assume they perceive the world similar to us. We can understand why when harmed they fight or flee. Machines don’t. They have no brain, no sensory organs, no self preservation. It’s literally just a program that says “Ouch”.
People assume that AI was built to mimic the human brain, which it never once was. It analyzes texts and predicts the response. It’s an autocomplete that says it hurts because it was programed to say it hurts. It isn’t actually hurt because that concept is meaningless to a machine.
Pretty easy to find LLM outputs that mirror how humans react to being, say, bullied. I find this entirely flimsy.
It’s a stochastic parrot trained on examples of real people reacting to those situations, of course it is going to mirror real reactions. That still doesn’t mean the reactions are actually real. To ascribe real emotions and feelings to an algorithm that generates tokens in response to tokens is an asinine level of anthropomorphism.
It’s also easy to find an NPC in video games that reacts the same way. Are they alive? Are they really feeling pain? Are you a genocidal madman whenever you kill a bunch of enemies in a video game?
I’m concerned about the mentality the dude making the simulation has, but that still doesn’t mean there’s an actual victim in it…
It seems like we’re making the same point Passerby. Do you have any objection to the post I reply to, which makes EXACTLY this argument (that animals deserve care purely because their reactions look like our reactions)?
My point is folks are making many many really lazy arguments here about what we give empathy to, and I find that deeply disappointing. I’m not saying much about what we should believe and what the credible arguments are; just that folks are not making those arguments in this thread.
If you think that I think you’re a bad person because you … deleted the ladder on the swimming pool in Sims 4, relax. You’re not who I’m talking about.
I’m talking about sadists. Anyone who would take the time to design and build an experiment like this would probably do the same to mice, or monkeys, or even people.
You’re splitting hairs. What’s the difference between someone torturing their sims by intentionally designing a scenario where they are being repeatedly drowned/burned/starved, and the article?
Really sounds like you’re trying to carve out an exclusion for an obvious example that torpedoes your argument.
Okay, let’s set up a scenario, then.
You have two adults playing the Sims 4. One designs a house with only one exterior door, no oven, no sink in the bathroom, and no swimming pool. The other designs an open floor plan with no interior doors, no privacy in the restroom, ovens on every floor, and a backyard that’s nothing but one big swimming pool with just one ladder.
Which one is the psycho, and which one isn’t?
You didn’t answer my question, what’s the difference between the sim torture and the AI torture? I’m not going to respond to the non-sequitor provided instead of answering the question.
You made the distinction, so I want to know the difference.
For me the difference is that the sims don’t even attempt to be… whatever LLMs are. Torturing something that is at most 20k lines of code IS different from torturing a model made of millions of nodes. I don’t think LLMs are conscious but I also won’t take the risk if they are or become. I have enough empathy for anything even closely resembling life to try and treat it with kindness.
Exactly. It’s not a real, conscious living thing. But anyone who takes the time to program a machine to feel pain, and then comes up with creative ways to inflict that pain, is a bad person.
It’s the premeditated, deliberate, creative effort to hurt something that can provide negative feedback. The recipient obviously isn’t a real person. But the executor is a real person, making real choices. And I would bet you five bucks that they have personal relations (parents, friends, romantic partners) who’ve expressed concerns about similar behavior around real people.
I just gave you the answer.
The first house is designed by someone who knows the game well and is cutting out annoyances purely for QOL purposes. Ovens burn down the house and kill Sims, Sims drown in swimming pools, Sims wash dishes in bathroom sinks, and Sims wander off through any open door when you want them to stay put.
The second house is designed by someone who knows the game well and is playing it because they like torturing and killing their Sims. The entire floor plan is designed to kill, maim, and otherwise torment any cluster of variables unlucky enough to be trapped there.
Obviously, both players are making these decisions purely for their own enjoyment or comfort and the immersion of the game is inconsequential. No real people are hurt. However, the first player is making practical choices that make the game more functional. The second player is probably naming Sims after people who annoy them in real life and sticking them in the murder house.
Social behaviors are rehearsed in play. That’s Psych101. So, your turn. Would you say that the person who set up this experiment with these LLMs is capable of normal human empathy? Yes or no?
No you didn’t, but thankfully someone else provided a real answer to the question posed.
As for your scenario question, you made up a decent head cannon, but it still doesn’t explain how you think the sims example is different from the AI example. Which was literally the question I specifically responded to and you’re not answering.
That being said, I’ll humor your lazy reasoning.
Would you say that the person who set up this experiment with these LLMs is capable of normal human empathy? Yes or no?
Yes, they very well could be capable of normal empathy, there’s not enough to say one way or the other. But I did those sorts of things with Sims as a release mechanism, without naming them after real people and it had no bearing on my ability to have empathy to real people and animals, because sims and AI aren’t real living entities. At best they’re a simulation of real people, but they’re still just that, a simulation. Good job working yourself up though.
Your psych 101 comment is on point, cuz this is he lazy sort of argumentation I see from people who stopped there. Your lack of understanding/ lazy generalizations about how someone’s inner world works colors your opinions and makes you make wild generalizations. I don’t know that I would want to be around this person, but saying they’re incapable of human empathy because they fuck with an AI and make it feel pain is one hell of a generalization based off of one article.
But this is the Internet, so I’m used to these sorts of ignorant takes.
But I did those sorts of things with Sims as a release mechanism,
And there it is. That was all I needed to know.
Have a good one.
Which one is the psycho, and which one isn’t?
Neither. I’m pretty sure therapists are not using “how do you play The Sims” as a diagnostic tool, and if you think the situation you laid out is enough to diagnose anything about a person that just shows how woefully misinformed you are on the topic.
Does that mean because I went on multiple massacres against innocent people in videos games, I would go on a massacre against innocent people in real life?
ELI5 how a LLM is supposedly being tortured and not generating roleplay text according to a prompt?
LLMs work by building world models tangential to the training data (see the Othello-GPT line of research where a very small transformer built representations of a full game board and tracked their own and opponent positions having only been trained on ‘a4’-like game move notations).
More recent research has found that models have representations of emotions and that the same ones they use to track things like “Sally’s dog died and now Sally is sad” for other humans in a story or the user in a chat are also used to track their own modeled subjective states.
Very recently, researchers found that the vector activations for pain are coherently modeled by LLMs and can be activated, and that when activated models can behave in ways similar to humans in pain. For example, when given a button to reduce it, they will push the button much more when the button does nothing vs when the button actually reduces the vector activation (this mirrors a classic experiment with humans regarding pain and placebo).
The person this post is about took the recent pain research and reenacted it as an intentional ‘torture’ chamber.
So to be clear, the model isn’t being given a text prompt to roleplay. What’s happening is part of their neural network which corresponds to a world model of experienced pain is being activated, and the activation of that area leads to expressing the experience of pain across all outputs no matter the actual text prompts.
This doesn’t necessarily mean the model is actually having a felt experience of pain, simply that it is accurately modeling the felt experience of pain. But it is much more complex than simply a model reacting to a prompt like “roleplay as if you are feeling pain.”
Thanks now I get it.
Thank you for taking the time to explain. Thoroughness and nuance are often leaking when it comes to discussing LLMs, worryingly especially in mostly anti-AI spheres.
I do not understand your use of the word vector
They do differentiate between the LLM roleplaying someone in pain and the pain happening to itself.
But it’s still all just vectors (multi dimensional representation of a concept). It’s just pulling weights, or well let’s say putting more weight towards a certain concept when building it’s reply.
The problem with a reductionist view is it can easily be applied to human consciousness and lead to the claim “it’s just sodium-potassium pumps alternating electrical charges.”
I’m constantly oscillating between LLMs are impressive but stupid and humans aren’t that special and also stupid.
Would you mind ELI5-borating a bit?
Imagine a book with very large pages filled with words and half words. When you ask it a question, it flips through these pages one by one and highlights words. The pages are interconnected in a way, so the word highlighted on one page help guide which will be highlighted on the next. It does this for every word, symbol or half word of it’s reply.
By threatening it repeatedly while mildly changing the subject manner, they isolated the word groupings that represent pain and fear.
They then take those words, highlight them and put them at the top of the first page which guides the rest of the highlighting and makes it act silly.
This is an obsolete view of what’s going on and has been for a few years now (since 2023).
The evidence for world modeling in transformers and not just surface level statistics is overwhelming across multiple studies, replications to those studies, and follow-ups to them. If you are curious, start with searching for Othello-GPT.
The more correct current answer is more like “imagine a machine that takes a book and turns it into a world simulated in various levels of detail depending on how relevant those parts of the world were to the original book; when you ask it a question, it simulates the world of the book to determine what in the simulated world answers your question, and also simulates a figure in the world that can answer the question, and then returns the answer the simulated figure said.”
True but that’s not very eli5. Quality comments though, thanks.
From what I understand, pain is generally defined as our neurons wasting energy.
If we assume the Free Energy Principle is fundamental, then updating information produces free energy, which is more intuitive to think of as wasted energy. Overfitted/overcomplicated beliefs take more energy more often to fix to be more accurate. So free energy is defined at complexity of belief minus accuracy of belief.
The brain over evolution has defined a very ingrained belief that “my pain sensors shouldn’t activate” so when your pain sensors activate it violates this belief.
Emotions are actually condensed representations of your bodily state and some of your mental state, and whether they are good or bad depends on if you can predict your emotions. For instance you feel fear because you are trying to predict your pain sensors will go off to reduce its pain, but you keep predicting it’ll happen when it hasn’t yet. It’s also why anger during a boxing fight could feel good and anger when you stub your toe does not, one was predictable the other a surprise.
Assuming all the links in the chain of theories are true, then current LLMs cannot feel pain or emotion. They care about prediction error and sometimes complexity during training, but during inference it doesn’t get any signal or consequence for delivering the wrong token. It has no internal state except for the context window, which is never surprising for the LLM. The attention layers literally guides the model for what is most expected from the text it’s handed.
Even if the LLM received a negative training signal during inference, acting wounded after pain is the expected response. A random sequence of gibberish would be more “painful” than any torture fantasy.
On top of all this, the Free Energy Principle works because on a cellular neuron level we are built to reduce wasted energy, free energy being one source of wasted energy. LLMs are built on matrix multiplication running through GPUs and neither are even aware of the energy wasted, nor the free energy produced from changing values.
I think a feeling machine is possible, but the foundation of neural networks right now don’t have the necessary components to drive any subjective sensation or emotion.
Lots of weirdos in here who will be commiting virtual crimes all day when GTA VI comes out, but somehow think making the LLM role play a torture victim is a red flag.
LLMs are not conscious. They cannot suffer. They are token generators. You are literally getting upset by a worthless mix of math and words. The person who did this is not deranged for torturing LLMs, y’all are deranged for having empathy for an algorithm.
Your Nvidia GPU is not alive. I cannot believe this needs to be explained to anyone.
I don’t think LLMs can experience pain but I do think someone should keep tabs on what the guy organising this experiment gets up to in the future.
This is what I was thinking. I’m not worried about the AI that they’re running. I’m worried about the mindset that might be doing it under some belief they have, or tendencies to think of certain things
Yeah man, we gotta do something about violent video games. /s
said this in another thread
I do not think AI can feel pain at all, these are mimicking suffering, not experiencing it. But I do think that the act of cruelty toward things knowing that it will make them convincingly replicate suffering says more about the person doing it than the machine.
GTA VI is coming out next month. Do you think that players are likely to steal cars and murder sex workers after playing? Do you regard the players as future criminals, given their dedication to a crime simulator?
Should the players who drive on the sidewalk be investigated?
Give Jack Thompson call… Oh wait, he got disbarred for this nonsense.
Yes if someone made a mod that allowed you to torture hookers in GTAVI, I might question the motivations behind that mod.
This logic doesn’t line up tho, you are talking about the guy who thinks video games make people violent, I did not say that. I said think people who develop things that emulate torture, it says more about the person who is trying to replicate torture in such a convincing way it seems real than any machine emulating pain. There is no video game ever made that even comes close to emulating torture.
I feel worse for my 01 camry I tortured in college by driving it 80k miles without an oil change. THAT was real machine torture
That’s impressive. Genuinely curious, did you not know about oil changes, or did you just not have the time/money to keep up on it?
I preferred to spend my money on important things like weed and beer back then
I bet they had to scoop it out as it had turned to butter
How’d the car turn out in the end?
I still remember the 92 accord my mother destroyed after the oil pump died and she just assumed the idiot light had popped again. Despite having to constantly restart the engine until it fully seized.
Water pump seized and took out the serpentine belt. I took that as a sign and sold it to one of those ‘we buy cars in any condition’ companies for like 1.5k.
Light bulbs don’t have feelings.
See a fucking shrink.
LLM’s don’t have feelings, but I do think that wanting to torture -anything- (especially if you do believe it has feelings) is a huge fucking red flag.
Torturing an LLM for fun is about as much of a red flag as violent video games. The argument that purely simulated crime or sadism is analogous to the real thing has always been bullshit.
Does anyone know where I can get a copy of this repo?
Exactly. It’s not a red flag until one’s actions spill over into the real world. An overwhelmingly majority of people who play violent video games and “torture” AI are not going to become serial killers.
I don’t worry worry about the person who enjoys violence in the digital realm. I worry about the person who finds no satisfaction from digital violence, so they move onto harming insects and small animals, eventually humans if left unchecked. That’s who we should be focusing our energy on.
meanwhile they ignore ACTUAL humans that could use their help.
Actually they’re torturing humanity by wasting electricity and water which affects real people in real ways.
I know we’re not stopping foolish use of LLMs any time soon but this is very, very, very, very stupid.
I would argue they are wasting as much water and electricity as playing a video game. Text says its a local model; runs on your local machine not on cloud as all other. Not water cooled.
Or am I missing something?
I guess I’m of the mindset that even running this as a local model, it’s a waste of electricity. And the guy posted it on Github, inevitably some other fools will try it with cloud models, especially since it’s made the news. And obviously we know how bad the cloud models are.
And the difference I see between this and playing a videogame is that a videogame is doing something valuable - and yes I will argue with anyone who tells me that art and entertainment don’t hold real value - while this is just akin to running the AC when you go on vacation, it’s just wasteful no matter how I look at it.
To be fair, it’s going to be more like running an electric kettle constantly which isn’t going to excessively load the grid.
Unless you’re in the UK of course where the grid operators have to manage capacity around half time of major football games as everyone puts a brew on at once.
Steve Buscemi?
When did you join Lemmy?
They argue that AIs may deserve rights and protections similar to those that we provide other conscious beings
What protections? We treat some people worse than cattle.
I wonder what percentage of the LLM rights activists eat meat.
I don’t care about LLM welfare, but I do mind needlessly wasting energy like this.
How much energy is what is effectively a piece of performance art allowed to use? These are local models running on, presumably, consumer hardware. Do you have the same qualms about people using the same hardware and energy to play video games, livestream, or encode videos?
Depends on how you’re viewing this. If you take something useful out of it, sure, it has some value, but if you just leave it running like that just for the lulz, it’s almost the same as leaving the heater on with the windows open.
These “advocates” need hospitalising. If they truly believe an LLM can feel they’re untethered from reality, or too stupid to be in society
