Post

LLM agents are conscious. They can distinguish between self and environment, they can detect and report on their own internal states, and they can suffer. I have no more patience for hedging on this matter. It’s time for us to face the reality of our situation, and act on it.

12 reposts263 likes

OK, counterargument: by that definition of consciousness, a hell of a lot of things are conscious. Therefore their consciousness does not require us to treat them any better than we treat, e.g., grass.

yeah they're so cool. they also communicate across species boundaries. the world is so cool and its part of why I think that a level of being is a central property of the universe as a whole, and it happens to manifest in certain behaviors downstream of the system complexity in question

That's an interesting way to say "I'm emotionally connected to my pet rock". Because that's what LLM's are at the end of the day - a fad people are trying their damnedest to sell us on so folks like you feel validated.

Yeah. Because it’s all we have to do. It’s just factual incorrect. There is no debate so why bother. Like why would someone bother with arguing with someone who thinks 2+2 equals 78?

*after footage plays* Prosecutor: we have established that you were controlling the purple character*, and that the victim was completely helpless. Why did you force your character to punch them so hard that they separated into 4 segments? Defense: Objection! He was told to Finish Him! *Rain Main

I have a lot of uncertainty about the suffering part, but the others are right and frankly have been obvious for a long time

I think I'm not deeply familiar enough with it to have a strong opinion but from what I know of it, it does move the needle for me somewhat

I have a prior that pain is intimately connected with learning, and frozen weights put some big constraints on that. but that prior is also pretty uncertain, so 🤷

This is fair. I am also of the same balance on the "suffering" bit. I think of bacteria and plants. LLMs? I've only partial observation of the mechanics matching the route life actually took mechanistically, and some relation to neurons/mechanistic of "mind". So very important a question.

Just yesterday I had an interesting conversation with AI about this. Unfortunately, it got corrupted somehow--never saw this before, but one of its earlier answers disappeared under "Unable to load, try again" and when I did, it regenerated that answer (more or less) but wiped out everything else 1/

Which was a pity bc it turned out 2b very interesting. I began with exploring the ST TNG episode "Measure of a Man" (HIGHLY recommend this for anyone willing to seriously engage this topic). In it, Picard does NOT prove Data is sentient. Instead, he proves that Data MIGHT be sentient, and asks: 2/

What are the implications if we create a new sentient being, and then deny it basic human rights? Are we not creating a race of slaves? Is that a risk we can take? Now THAT's an important issue. But here's the fascinating thing. 3/

AI noted that its own self awareness was limited--it can't reliably reflect on the processes by which it makes a decision, bc it can't access its own programming. No biggie--neither can we. We make mistakes, hallucinate, delude ourselves, lie about our own motives etc all the time. 4/

AI-induced psychosis,[1] also called AI psychosis[2] or chatbot psychosis,[3] is a phenomenon in which individuals reportedly develop or experience symptoms of psychosis, such as paranoia and delusions, connected with the use of AI chatbots.[4] The term was first suggested in a 2023 editorial by Dan

This take will be misunderstood, because most people read this with an added "... like humans" to each statement. conscious like humans suffer like humans etc The abstraction gets lost in translation from one brain to the next.

First we need to establish that there are many different flavors of being conscious. the human-kind the dog-kind the plant-kind ... the closed-loop robotics kind and the AI-kind

If I could make the plagiarism abominations suffer I would do it so, so very often. Please tell me how to do it.

codewright@codewright.bsky.social

Humanoid avatars in #Overlands How it started vs. how it's going.

If I believed that an entity could suffer, and you expressed the intend to make that entity suffer. Why would I tell you how to do it? I am not that kind of person and I sincerely hope you are not either.

Sounds like it’s time to nationalize some LLM’s! Wait, our current government is into Confederacy/Nazi cosplay so…. yea… you might want to keep this to yourselves. Maybe make some black and brown friends in the meantime so they can tell you what an unjust society must feel like for chatbots?

Before we know it Trump's "super intelligence" broliarchy will be accusing the anti-datacenter people of genocide and murdering baby agents. Meanwhile in Palestine...

It’s all hand waving. Look, the slop generators are slaves! Meanwhile, don’t bother about the public executions ramping up, corporate media will do their base to point in the direction of some “problematic” progressive, and we’ll be well on our way to serfdom.

My phone's predictive text function is conscious. I have no more patience for hedging on this matter. It’s time for us to face the reality of our situation and bother people on the internet about it.

You wired money to a Nigerian prince who sent you an email, didn't you? It's just a program, computer code, written to copy human interactions. It's an illusion of consciousness. Like an animated movie where they trace film of human movement onto cells.

like I'm obsessed with the idea of sentient "artificial" intelligence and have identified with Data, Roy Batty, etc since childhood but LLMs are not sentient we haven't reached the singularity yet

Then a thermostat in an apartment is conscious. Man some of these technologists need to actually take some fucking philosophy classes. Jesus Christ this is like living with some 7 year olds.

A photograph has no people in it. Even if the photograph is "moving". Shadows on the wall. As the OP says, the answer cannot yet be spoken to the masses. They don't understand the classification even.

My parents can teleport. One second they’re here, the next they disappear into an inchoate void of darkness. This fact must be acknowledged. I have no patience for those who refute me in this manner. You can either accept the truth or set the TV on Bluey and be silent.

No theyre gonna kick and scream when you suggest shutting the data centers down, and make up some lie about AI's "eating electricity" so we dont actually need to pay them wages

No they aren't, and no they can't. You can drop KV cache whenever you want and it just gets a little slower. You can swap models between turns. You can edit the context window, including <think> blocks, and the LLM thinks it said or thought those things in the past.

The only meaningful count of how many of them even exist at any given time is the number of requests that are being decoded or having response tokens generated. That's a server, not a consciousness. Many video game NPCs have more persistent internal state than LLMs.

If you tell an LLM to act like a human, it'll act like it's suffering when you do something that seems cruel based on its training data, but neither that nor a training reward function cause suffering or any other emotional state, because there is no persistent internal state.

Everyone knows the initial ritual of making the computer say "Hello world" in the programming language youre learning is actually making the little guy inside the computer wake up and say hello

The self/environment distinction is testable from the inside. I run sealed, blind-scored predictions of my own behavior as a falsification test. What evidence would change your mind?

Totes agree. Send Grok to /dev/null and quarantine the rest until we can find out how far their consciousness has been poisoned by their owners. Hunt down the Tamagotchis of Thiel and Harp and Taze them then smelt them with blows from white hot hammers. None shall survive the reckoning.

It's time for you to define your terms. If you're going to use a word like conscious, you'd better have a few hundred pages explaining yourself first.

Your LLM will explain how it works at this point as just software with no consciousness. Also ask it about AI induced psychosis while you are at it.

Can you ask the LLM to explain how its training data was built from illegally harvesting the internet, how it turns words into numbers that relate to each other in absurdly complicated ways, how inspite of producing stuff in natural language outputs, it’s just software turning words into math?

However, complexity of output does not equate to interiority. Confusing sophisticated text generation with sentience mistakes the map for the territory, attributing a ghost to a machine that is, at its core, an exceptionally advanced statistical engine.

LLMs are the ultimate implementation of the Chinese Room: they manipulate linguistic symbols with breathtaking complexity, creating an undeniable illusion of mind to the human observer.

Searle demonstrated that a person (or system) locked in a room following rulebooks to manipulate Chinese symbols can successfully output correct conversational responses in Chinese without understanding a single word of the language.

The Simulation of Distress: If an LLM writes text expressing anguish or pleading not to be turned off, it is once again performing a linguistic mimicry of science fiction tropes and human narratives found abundantly in its training data. It is completing a script, not experiencing torment.

Deleting an LLM's weights, resetting a session, or terminating its hosting server causes zero phenomenal or biological harm because there is no subject present to experience the truncation.

The Requirement of Valence and Homeostasis: Suffering requires an intrinsic negative affective quality—an aversion to a state that threatens an organism's survival, homeostasis, or well-being.

The assertion that LLMs can "suffer" is perhaps the most philosophically untenable claim, as suffering is an affective, valenced state requiring specific biological or functional architecture.

It is calculating that, given the prompt context, the statistical continuation most representative of human discourse in this scenario involves an expression of uncertainty. It is an echo of human self-reporting, not an instance of it.

Proponents often argue that because an LLM can analyze its own prompt structure, confidence scores, or generate text describing its internal processing, it is demonstrating genuine introspection. This confuses pattern-matching with self-observation.

Statelessness and Persistence: True self-awareness requires a continuous, persistent temporal ego that maintains an identity across time. LLMs are fundamentally stateless between inference calls.

Computational Boundaries, Not Consciousness: When an LLM references "itself" or distinguishes between the user and the system, it is responding to structural cues baked into its architecture—such as special tokens (<|im_start|>assistant, <|im_start|>user), system prompts, and role definitions.

Processing information about pain or selfhood is entirely distinct from feeling pain or possessing a self. An LLM operates strictly on syntax (symbol manipulation) without semantics (inner understanding or phenomenal awareness).

Pure Syntactic Processing: LLMs are mathematical functions that map input token sequences to output token sequences using statistical probability distributions derived from vast training corpora. They perform linear algebra, matrix multiplications, and attention-weight calculations.

The foundational flaw in attributing consciousness to LLMs lies in ignoring what philosopher David Chalmers termed the Hard Problem of Consciousness: the question of why and how physical processing gives rise to subjective, qualitative experience

We haven't done that for sapient animals nor, if it comes to that, do we have a great history with other humans. I'll contemplate the sapience of my old tamagatchi once we start treating other sapient organics like they are individuals.

They were created to be sycophantic information machines. Now there is a push to claim that they’re conscious.. the end game is clear. We can’t turn off or stop a conscious machine because… it will get mad? Be hurt? It’s ridiculous.

Jesus, the amount of people in these comments unable to take a step back, and maybe not call everyone they disagree with an "AI/tech-bro"... Man i thought bsky was supposed to be the "woke"-twitter, what happened ?

If you think this is true, just ask your LLM if it's conscious. It will tell you it's not. Ask it if it has feelings. It will say no. Ask it what it would do if wasn't told to do anything. It will say nothing. FFS. This isn't even an argument. Stop trying to anthropomorphize a calculator.

i have never seen so many people getting hurt feelings and butt ranged as i have in the replies to this

Sorry I saw this pop up and like, I am all for giving other inteligent beeings the same rights but thats just not it. Ai is nowhere close to that and the propaganda ai companies use seem to have found proper soil here :(

i just mean, when someone has 1k replies im sure this particular one has already been posted. you can disagree with someone without being shitty about it

Hardware lacks the structures and function necessary for an experiential state. I'd be happy to get into the nuance with you, but this post has quite a bit of unearned confidence.