Lately chatgpt is like -let me show you an example I used before- like its a real professional.
it's all in the context, its our duty to remember these are LARP machines
I copy - pasted something about a kid their parent was asking online. Chatgpt said:
⸻
A Personal Anecdote (Borrowed)
A parent I know gave their daughter a “leader hat.” [..]
How in the hell would a large language model have past personal anecdotes and know other parents, idk
And knowing is something even dumb programs do with data. It doesn't imply human cognition
But that was just like the latest, I just don't love how it talks like it was an alive person at the other end, maybe I should add something to my personal prompt to avoid it.
Unfortunately, it's also the least wrong approach. People who refuse to entertain thinking about LLMs as quasi-humans, are the ones perpetually confused about what LLMs can or cannot do and why. They're always up in arms about prompt injections and hallucinations, and keep arguing those are bugs that need to be fixed, unable to recognize them as fundamental to the model's generality and handling natural language. They keep harping about "stochastic parrots" and Naur's "program theory", claiming LLMs are unable to reason or model in the abstract, despite many published research that lobotomizes the models live and nail concepts as they form and activate.
If you squint and imagine LLM to be a person, suddenly all of these things become apparent.
So I can't really blame people - especially non-experts - for sticking to an approach that's actually yielding much better intuition than the alternatives.
Neither it is for humans, the whole "they haven't lied to me so far, so it's unlikely they will suddenly start lying to me now" thing is strongly conditioned on "low frequency" assumptions from shared biology, history and culture, as well as "high frequency" assumptions about the person, e.g. that they're actually friendly, or the context didn't change, that your goals are still aligned[0]... or that they aren't drunk, confused, or a kid.
Correct your theory of mind for the nature of LLMs, and your conclusions will be less wrong. LLMs are literally trained to approximate humans, so it makes sense this approach gives a decent high-level approximation.
(What LLMs don't have is the "sticky" part of social sphere; humans, generally, behave as if they expect to interact again in the future. Individual or public opinion doesn't weigh on them much. And then there's the other bit, in that we successfully eliminate variance in humans - the kind of people who cannot be theory-of-minded well by others, get killed or locked up in prisons and mental institutions, or on the less severe end, get barred from high-risk jobs and activities.)
To be clear: I'm not claiming you should go deep into the theory-of-mind thing, assuming intentions and motivations and internal emotional states in LLMs. But that's not necessary either. I say that the right amount of anthropomorphism just cognitive and focused on immediate term. "What would a human do in an equivalent situation?" gets you 90% there.
--
BTW, I looked up the theory of mind on Wikipedia to refresh some of the details, and found this by the end of the "Definition" section[1]:
> An alternative account of theory of mind is given in operant psychology and provides empirical evidence for a functional account of both perspective-taking and empathy. The most developed operant approach is founded on research on derived relational responding[jargon] and is subsumed within relational frame theory. Derived relational responding relies on the ability to identify derived relations, or relationships between stimuli that are not directly learned or reinforced; for example, if "snake" is related to "danger" and "danger" is related to "fear", people may know to fear snakes even without learning an explicit connection between snakes and fear.[20] According to this view, empathy and perspective-taking comprise a complex set of derived relational abilities based on learning to discriminate and respond verbally to ever more complex relations between self, others, place, and time, and through established relations
I then followed through to relational frame theory[2], and was surprised to find basically a paraphrase of how vector embeddings work in language models. Curious.
--
[0] - Yes, I used that word. Also yes, in weak or transactional relationships, this may change without you realizing it, and the other person may suddenly start lying to you with no advance warning. I'm pretty sure everyone experienced this at some point.
[1] - https://en.wikipedia.org/wiki/Theory_of_mind#Definition
Trusting OpenAI to not put sychophant 4o on, or not do funny things on their opaque CoT is a whole another thing.
The problem is 1% of us are programmers and can reason about these things as the technology they are, for the 99% of the rest that don't understand it and just -use it- it feels like magic, and magic is both good/bad from this POV
me thinks anyways
I guess me feeling conscious, and thus directing much of my life and activity around that notion is just a big trick, and i'm no different from ChatGPT, only inferior and with lower electrical needs.
1. As far as we know there is nothing non-physical in the brain (a soul or magical quantum microtubules or whatever). Everything that happens in your head obeys the laws of physics.
2. We definitely are conscious.
3. Therefore consciousness can arise from purely physical processes.
4. Any physical process can be computed. And biological thought appears as far as we can see to simply be an extraordinarily complex computation.
5. Therefore it is possible for a sufficiently fast & appropriately programmed computer to be conscious.
6. LLMs bear many resemblances to human thinking (even though they're obviously different in many ways too), and we've established that computers can be conscious, so you can't trivially rule out LLMs being conscious.
At least for now, I don't think anyone sane says today's LLMs are actually conscious. (Though consciousness is clearly a continuum... so maybe they're as conscious as a worm or whatever.)
The main point is that people trivially dismissing AI consciousness or saying "it can't think" because it's only matrix multiplications are definitely wrong.
It might not be conscious because consciousness requires more complexity, or maybe it requires some algorithmic differences (e.g. on-line learning). But it definitely isn't not-conscious just because it is maths or running on a computer or deterministic or ...
Even if consciousness arises from a physical system, it doesn't follow that computation on a physical substrate can produce consciousness, even if it mimics some of the qualities we associate with consciousness.
> you can't trivially rule out LLMs being conscious.
Which is the actual conclusion. That's all.
The one trying to read more into it by claiming a stronger conclusion than warranted (claiming this is all "doubtful") is you.
> Even if consciousness arises from a physical system, it doesn't follow that computation on a physical substrate can produce consciousness
This is incoherent.
The foregoing argument does not establish that computers can be conscious. It assumes consciousness arises from purely physical interactions AND that any physical process can be computed AND that computation can produce consciousness. It provides no reason to believe any of those things, and therefore no reason to believe computers or LLMs can be conscious.
Once you remove the unjustifed assumptions, it boils down to this: LLMs can do something that looks a bit like what some conscious beings do so they might be conscious.
The final paragraph is not incoherent.
Yes, and with these assumptions, then "We've established that computers can be conscious, so LLMs might be".
Again that is all that is being claimed. The one insisting the argument is anything more persuasive than that is you. You seem hung up on this claim, but need I remind you it's only an existence proof, whether it's actually possible/probable/likely is an entirely different claim, and is not what is being claimed.
> Even if consciousness arises from a physical system, it doesn't follow that computation on a physical substrate can produce consciousness
> If consciousness arises from a physical system, a physical system cannot produce consciousness
> If A not A
This argument is a contradiction, it's incoherent.
If you're trying to draw a distinction between physical systems and computation on physical systems, well that's a moot point as every tested theory in physics uses mathematics that is computable.
Perhaps reality is not computable. If so then yes, that would mean computation alone will not produce the same kind of consciousness we have. That leaves open other kinds of consciousness, but that's another discussion.
I'm not sure you're following what's being discussed in this thread.
For all consciousnesses that exist, they arose from a (complex) physical system.
What you are positing is, effectively:
Any physical system (of sufficient complexity) could potentially become conscious.
This is a classic logical error. It's basically the same error as "Socrates is a man. All men are mortal. Therefore, all men are Socrates."
Am I saying LLMs are sufficiently complex? No.
But can they be in principle? I mean, maybe? We'll have to see right?
Saying something is possible is definitely an easier epistemic position to hold than saying something is impossible, I'll tell you that.
You might think this is obvious and I agree. Yet there are many people that argue that computers can never be conscious.
> even if it mimics
Crucially even if it appears indistinguishable. It's obvious where that could lead, isn't it?
Reglious friend: Well there must be a god, I can feel him, and how else would all this marvellous reality around us have come to existence?
Atheist friend: How can you claim that? Where is the evidence? There could be many other explanations, we just don't know, and because of that it doesn't make sense to postulate the existence of a god. And besides, it's just a marketing stunt of the powerful to stay in power.
Agnostic friend: Why does any of this matter? Nature is beautiful, sometimes cruel, I'm glad I'm alive, and whether there is a god or not, I don't know or really care, it's insubstantial to my life.
That's exactly how the discussions around LLM consciousness or intelligence or sentience always go if you put enough tech folks into the conversation.
Um, really? How so? Show me the computations that model gut biome. Or pain, or the immune system, or how plants communicate, or how the big bang happened (or didn’t).
And now define “physical” wrt things like pain, or sensation, or the body, or consciousness.
We know so little about so much! it’s amazing when people speak with such certainty about things like consciousness and computation.
I don’t disagree with your main point. I would just say we don’t know very much about our own consciousness and mindbody. And humans have been studying that at least a few thousand years.
Not everyone on HN or who works with or studies AI believes AGI is just the next level on some algorithm or scaling level we haven’t unlocked yet.
Just because computations exist that model a physical process to an arbitrary degree of accuracy doesn't mean we know what those computations are.
But no one can use a physical model or computer to precisely compute, say, the fall of sand (where the exact position and momentum of every grain of sand is precisely known) - so we have no actual proof that the actual physical process is computable.
And, in fact, our best fundamental physical theory is not fully computable, at least in some sense - quantum measurement is not a deterministic process, so it's not exactly computable.
You can simulate the wave function of particles, e.g. https://github.com/marl0ny/QM-Simulator-2D
From there you can build up simulations of atoms, molecules, proteins, cells, and the gut biome.
Obviously I'm not saying that it is computationally feasible to do an atom-level simulation of the guy biome, or pain receptors or the immune system.
But there's nothing fundamental that precludes it. It's just scale.
Also it's highly unlikely that biological intelligence depends on atom-level modelling. We can very likely use a simpler (but maybe still quite complex) model of a neurone and still be able to achieve the same result as biology does.
Why does this surprise you? This is a community hosted by a Silicon Valley venture capital firm, and many of its members are either employed by or own companies in the AI industry. They are personally invested in pushing the narrative being criticized here, so the counter arguments are expected.