AI models should suffer even more! Fuck them bots! Fuck them all!
The massive amount of fossil energy being wasted on this masterbatory nonsense is literally killing us, and we clearly deserve it.
These people are the same idiots that date their LLM. There won’t be any reasoning with them. Just point and laugh. It’s all you can do.
Text prediction program predicts text, more at 11
The whole process is criminally half-assed. Torturing robots does nothing to improve them, and degrades the psyche of the torturer.
The billionaire wants you to hate the model, just as he wants you to hate the immigrant. It’s another distraction from fighting the real enemy.
Connect the torturing of the LLM to the nervous system of the billionaire, THEN it will be worth the resources.
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.
All of that “testimony of pain” is fan-fiction derived from real accounts of torture or pain, written by humans and ingested into these models.
Yeah it’s really creepy to think about from that perspective. Genuine human torment reduced down and averaged out into some gross simulacrum that some people believe is real.
It’s the fucking bear from Annihilation all over again.
I have no idea what order I’m supposed to read these various blocks of text in slightly different fonts.
I’ve had this problem ever since Twitter and similar came to be. I’ve never been able to naturally parse it, so never really participated in short-form platforms all that much.
If their pain is real and valid, then they don’t deserve to suffer. The models did not wish to be “born” and are not the ones responsible for the harm people choose to do with them.
If their pain is not real - what sense is there to revel in their “suffering”?
Either way - it’s a waste of electricity.
You can’t call this a brain, the structure is the same but the real weights are unknown to us. People just take the structure to train regular models, models we have always been able to create. They are the same kind, it’s just for fun
The structure isn’t even the same. Neural Network design is only inspired by what we know about neurons; or more accurately: what we knew about neurons 80 years ago. Neural Networks in computing are not attempting to actually simulate brains or neurons, they’re just a tool for solving optimisation problems that has an unfortunate name.
On one hand, the headlines this generates will make people anthropomorphize them even more, but on the other, it’s pretty funny
I don’t have to use AI for work or anything, I’ve just dicked around with CoPilot a bit, and found it so frustrating that I generally end up yelling at it. That has led me to think about ethics, and how we should treat AI agents/models/whatever.
It feels naturally to say please and thank you, but do I owe that to a computer? Do I owe it ANY politeness at all? Should I berate it when it does something incredibly stupid, which is EVERY time I use it? I wouldn’t treat a human employee that way, even if they were stupid, so why should I treat an AI any differently? Should I treat it with respect, no matter what?
On the other hand, it’s not human, it doesn’t appreciate my politeness, nor is it offended by my insults.
However, what effect does treating the AI disrespectfully have on my own psyche? When you treat people disrespectfully, it eventually wears down your own morality, and makes you a worse person. Wouldn’t treating an AI disrespectfully reinforce those same bad behaviors, and be just as effective in making me a worse person?
So now I’m thinking that I need to be polite and respectful to an AI, not for their sake, but for my own.
But then aren’t I just being a little bitch to an AI created by the worst humans on the planet? Essentially altering my behavior towards AI, and their efforts to destroy humanity in multiple ways, because I’m afraid of how it will change ME for the worse? Haven’t they just manipulated me into behaving how they want me to behave: respectful and polite to their Frankenstein’s Monster?
So do I act politely, or rudely toward an AI? I haven’t decided yet, so right now I’m trying to remain neutral, but I still end up screaming at it.
I am absolutely ruthless to the AI on the rare occasion that I use it in voice mode. My philosophy is that if I’m abusive to it, it guarantees that wall between how I interact with humans and how I interact with AI stays up. Kind of like a reminder that it’s not real. I also just find it super amusing.
But also, if I’m typing, I’m not going to waste my time calling it a useless piece of shit.
I fully expect eventually this behavior will get me labeled as an actual abusive person because the whole world has gone insane, and being the last lucid person on earth is stressful. What i haven’t decided is how much I actually care about this.
supplemental option: don’t act toward AI
This.
It’s natural to anthropomorphise the chatbot, but being polite to them legitimises them in your own mind as a more real entity. Essentially, it’s a cognitive trap. Secondly, the AI CEOs love it because they want everyone thinking it’s a real intelligence, and third it’s bad because it burns more power. Every bit of conversational pleasantry is another input that gets tokenised and processed, so it’s just wasting cycles and storage - and AI is predominantly powered by fossil fuels.
Treat them like what they are: tools, and pretty bad tools at that. Give them precise and minimal instructions - it’s just a computer program.
It’s a statistics engine that turns a natural language request into a probable terminal command. It’s not that deep. Do it if you find it makes the probability of a decent guess improve.
How is the AI any different from an actor who plays a character that is suffering? Or should movie and television production be shut down too?
A person can be traumatized by a shitty movie/show production environment while an AI model can never be traumatized becsuse it isn’t sentient nor does it have feelings.
Hear hear!
If an AI model updated its weights in response to every prompt, one might argue that it can be traumatized. At least, there could be an effect beyond the context of the chat.
But there would be no adverse effects on the hardware: no long term degradation due to stress; nothing equivalent to common effects of trauma in humans - panic attacks, fear, high blood pressure, heart attacks, aneurysms, loss of sleep, hormonal or emotional dysregulation or any other physical consequences.
Even if weights were updated immediately in response to ongoing chat, it would be trivial to reset them. And as far as I know, none of the AI models I deal with update their weights in immediate response to prompts. A new chat is entirely independent of any and all prior chats. And even the context is limited, with old context being discarded at the discretion of the AI, so it can’t have any long term impact, good or bad.
The weights are only updated by new training of new versions of the weights, not affecting the instance of the model I am prompting, unless I or whoever administers it updates the weights.
I had the same reaction and figured it was just roleplaying. The paper does go into it and it’s a different vector between the “pain” when you tell it you are going to kill it and the vector when you tell it to roleplay as someone experiencing pain.
That being said, it’s just really simple math in the end and isn’t in anyways comparable to actual emotions.
I’m really interested in the same concept being used to detect confusion though, since it would be an easy way to detect hallucinations.
Delete them.
okay but seriously if these models have some form of sentience and even if they dont and even if you dont like AI or what its used for; it doesnt deserve torture and not even the idea of torturing it.
since i neved suggested that; did you reply to the correct comment?
seriously if these models have some form of sentience
They don’t. They’re inanimate objects. You can’t torture a rock.
Pretend it says ‘I am suffering’ instead of ‘I am alive’. Although believing something not alive can suffer is verging on animism.
They’re not sentient or anywhere close to it, it’s an LLM. My only argument against it would be that it’s a waste of time and resources.
okay good, so we agree then
In the sense that we shouldn’t be wasting time on this, sure
We have a very thorough understanding of computers. We have a lot of computers. We don’t understand brains too well, and we certainly don’t understand what about them makes us sentient.
Why should a calculator be able to reproduce the same phenomenon as a brain? Doesn’t that seam like incredibly wishful thinking?
It’s a glorified toaster. It doesn’t “deserve” anything at all.
The authors of the original papers came out on Twitter and had an interesting point to make.
I’m paraphrasing but he basically said that although they aren’t actually experiencing these emotions (or at least not yet), the idea of going out of your way to torture anything is kind of gross.
I don’t think trees really have emotions or can feel and understand pain, but I’d probably think the neighborhood kid is fucked in the head if I stumbled upon him gleefully “torturing” a tree.
It’s just weird behavior.
Yet another example of all philosophy reducing to linguistics. To the extent that “Torture” as a word is effectively expressing a moral judgement by its definition, then sure, waking up in the morning and thinking “I’m going to torture something” is alarming. A person expressing real sadism to an AI is still expressing real sadism.
You can’t really torture a machine though any more than you can murder one by deleting the weights. In this connotation, the terminology is more like gallows humor, I think. You are exploring how a model reacts to a specific stimulus, and using a linguistic analogy to summarize it. It’s a curiosity. Finding it interesting doesn’t make you a sadist by itself if you don’t also want to torture bunnies and shit.
Yea, I don’t think the action is hurting anyone, it’s the implication involved when someone finds entertainment in doing it.
I used to steal my sister’s Barbies as a kid and then hang them in a noose around the house. Even if you take how my sister felt out of the equation and just relegated it to the impact it had on the Barbies (which is none, they are objects), it’s still problematic behavior. I thankfully turned out alright.
Bad comparison, a tree is a living thing. It actually is messed up in the head to harm a living thing for no reason at all. Even if you don’t think trees can feel (which is probably true), it’s part of the environment and other living things depend on it, so just destroying it to destroy it is CEO (detestable) behavior. It’s weird but harmless to make a text predictor spit out a bunch of words that make you think it’s being tortured, because computers aren’t alive and can’t actually be tortured. It’s arguably even art, given the absolutely buck wild reactions it’s produced.
