Artificial Insanity
richwhitehouse.com
richwhitehouse.com
It's safe to say that argument has collapsed utterly. Modern LLMs virtually never collapse into endless babbling loops. Far more importantly, they solve objectively hard problems. An LLM didn't merely inspire a mathematician to spot a counterexample to the Jacobian Conjecture that had been in front of his nose the whole time; it found the counterexample. These things are unquestionably generating meaningful outputs.
(That obviously doesn't mean that you can simply trust LLM outputs!)
At this point, trying to dunk on frontier models by calling them "stochastic parrots" is saying more about you than about the models. Moreover, it suggests you're not only out on a limb about the capabilities of LLMs, but also on what even the skeptic literature about it was saying.
You should know what "stochastic parrot" actually meant; you shouldn't be using the term just because you think it sounds snazzy.
The funny thing is though, I think it’s an in group signalling mechanism. You need the anti-AI crowd to know that you’re anti-AI, and this is a rapid way of doing so, to get it upvoted here.
It's not like the "stochastic parrot" thing was the only problem here.
> I mean, technically speaking, a video game from 1980 was likely to contain more logical branching points than a modern LLM's kernel, and was likely to be a lot more interesting to read.
Are you really saying that the "AI" in 1980 is _Really_ much more interesting than today's LLM? The author is either dishonest or ignorant.
And yet, they still output gibberish. They also still fail to solve objectively easy problems on a regular basis. That is because they are, indeed, still stochastic parrots without any actual understanding of anything or ability to reason about stuff. The argument hasn't collapsed at all.
LLMs are certainly stochastic, but they no longer meet any sane person's definition of "parrot." I can't imagine Gebru disagreeing. Has anyone bothered to ask her?
(Consider responding with an actual counterpoint, so we can all see what reasoning might give rise to such a delusion. Does your pet parrot do calculus and write stories and reverse-engineer Adobe applications?)
Which?
https://en.wikipedia.org/wiki/Stochastic_parrot
"a metaphor that frames large language models as systems that statistically mimic text without real understanding". I suppose you're defending them as having understanding. It's not "safe to say that argument has collapsed utterly", you're part of a campaign to reject the meme. On the other side, the meme is popular as a way to be rude about AI, because it's a cutting insult.
It means whatever you want it to mean, that's the best part. To me, it's best used as it was intended, but primarily in the contexts where the stochastic parrots are plowing through the ivory towers we built mostly for idiot savants.
They’re incapable of changing their mind, because they’re a Markov chain. They’re cool and they’re good at doing things, but they’re not sentient in the slightest.
(I don't think LLMs are "sentient".)
There's some haziness here about the difference between comprehensible output and comprehending. To put it mildly. I mean that's been the whole question for five or six years.
> Her argument claimed that while we perceived legible text coming from ChatGPT, we were actually being fooled by our innate pattern-matching instincts
This is an accurate summary, I think, and the point the authors are making sits somewhere in the coordinate space of a simplified reframing of the Chinese Room, Grice's account of intention in meaning, Wittgenstein's "Say 'It's cold here' and mean 'It's warm here'" challenge, and a whole bunch of the hermeneutic tradition.
Very roughly, you might say that an LLM doesn't mean anything when it emits text, but that the text still has meaning (which I think is all tptacek wants from "meaningful outputs", but you could take that further as culturally or scientifically 'meaningful', i.e. 'significant').
Philosophically, there are people who hold the position that something has to have been meant to have meaning, or that something has to have meaning in the first place to be meant, and so on. There were always interesting edge cases to these accounts (actors delivering lines, or the Dada/Burroughs cut-up technique), but LLMs are challenging in a way that e.g. Racter wasn't.
"Stochastic parrot" is a fitting term for this because the parrot's words have meaning, but the parrot itself doesn't mean anything by them. Not literally, anyway. And that's it. That's the whole analogy.
Have some people watered the term down to "LLM dumb"? Yes. Is that what Gebru meant in the first place? Not in my reading. Is it incorrect to use the term today? Depends, uh, on what you mean by it.
I don't mean to defend the paper in general, though. It has plenty of problems.
It is a pretty practical distinction.
So, since apparently I may need to clarify my actual point: For most people, an LLM-plus-agent-harness is just "LLM" or "AI". If you're trying to build one, you care about the distinction. Ditto if you're a pedant. But most people are not either of those things.
Go to a huggingface repo, look at the files, they have a literal chat_template.jinja
When I look at a person moving in front of me, I don't have to recheck if their bone structure is going shift as they more - it just won't. So, as a human, I don't naturally notice things of that sort. So I don't notice all the ways an AI video is wrong unless I concentrate. Which is saying that LLMs are wrong in ways we aren't accustomed to. They drift in their interactions and seeming understanding. Perhaps "random walk parrot" is better than "stochastic parrot" but I still say a word that communicates that they're "things that aren't quite doing what you think they're doing" is legitimate.
It's interesting that you'd be choosing this canned phrase of all canned phrases, given just how hard that is linked to emotion-driven facebook comment fights that is.
This doesn't read like something that belongs into an engineering-driven discussion.
All we have to do is assume that things progress in a linear way.
Do you remember GPT-2? It was a bizarre curiosity, but was nevertheless considered a huge advance in late 2019. Practically useless, but interesting.
It grew, and three and a half years later we got GPT-4, which was exceptionally capable on release.
It grew still further, and three and a half years hence we now have GPT-6. This is already more intelligent, more imaginative, more disciplined, and far more erudite than the vast majority of individual humans. If you have an intellectual skill -- if you are, say, a mathematician or a chemist -- GPT-6 also has that skill and is functionally your peer. And then it has many other skills that you lack; if nothing else, it is superhuman in breadth.
Assume a steady rate of progress. In three and a half years, we're going to have something that is genuinely a general superintelligence, and we're probably going to see recursive self-improvement begin to kick in (though subject to physical constraints on compute, etc.)
Now is the time for an RSI start-up, I guess!
Anyway, all you have to do is assume that things don't plateau. They have not plateaued to date; if anything, progress appears to be accelerating, and people are starting to panic that there are no brakes on this thing.
There are.. physical limits, as it were, which effectively restrict upper limit, but I hear what you are saying. A lot of the problems thrown at it and issues found in the process only seem to have helped. I feels ages ago when we were arguing 'it is merely a stochastic parrot'.
I recall them saying it was too dangerous to release, funny how they are only saying that again as they look to IPO
Then, a few years later, they learned that Google and Anthropic were planning on releasing LLM chatbots to the public, so they hastily launched GPT-3.5-based ChatGPT as a "dev preview" to get ahead of them. Their original plan was to finish training GPT-4 before launching anything.
Commercial considerations -- and public glory -- triumphed in the end.
Kinda chilling to see one of your kind in the wild.
However, what seems to still be lacking in LLMs — aside from plasticity (continuous learning), perception of time, and anything related to emotions and desires (which underly motivations) — is various types of awareness, and sound judgement. We are still far from a good understanding of these aspects.
This sort of nihilism/defeatism always irks me. It's so selfish, so privileged.
Tell my starved, murdered ancestors "let us enjoy the end" of their era of stability's denoument.
And, truly, it is not the end. Inconveniently, many—if not most—of us will survive. Will continue on. And it's for the betterment of those who come after us that we ought to toil a little, and not indulge in this childish, self-interested to nihilism.
marginalized populations have repeatedly survived past "apocalypses" not through aesthetic surrender, but through the deliberate, daily reproduction of the commons, ie through pooling resources, sustaining life for its own sake, and preserving collective memory.
Yielding to fashionable doomism merely validates the oppressor's narrative that the future is already decided. A superstition of doom is no different from a superstition of inevitable progress; both models rely on an abstract determinism that strips humans of agency, that encourages the us to accept suffering as unalterable destiny or "the will of God" rather than an artificial, historical barrier.
So, I agree with a lot of what OP is saying, but I think they're a jerk.
I often hear this asserted, but never with any actual justification.
Why can it not?
- careers, nay, nearly whole industries may ride on that being the case - classifying it as conscious intelligence opens a range of ethical questions no thinking, conscious, intelligent ( and likely non-evil ) life-form would want to comprehend - being able to ascribe judgment to it without intelligence undermines a lot of what we do on a daily basis and puts into question what it is that makes us human - I am saying this part in the nicest possible way: hardcore cope of biblical proportions, where seemingly conscious, intelligent individual, who makes an argument for lack of consciousness manages to ( in the same sentence! ) to anthropomorphize the parrot and imbue it with hope - average human tendency to fear the unknown
You have to prove it is/can be conscious, not the other way around.
Imagine if we actually demanded that you prove your consciousness before we believe you.
We don’t have a precise enough definition to do either.
The author is trying to positively assert that it’s impossible that AIs have consciousness.
I think you’d be hard pressed to find even the most ardent of AI proponents claiming AIs are _definitely_ conscious. They are almost always claiming something weaker, which is that it is possible.
"They can't be intelligent as they lack consciousness"
"Ok then how does intelligence require consciousness? What is consciousness? How could one prove that someone or something has consciousness?"
"That's on you."
No, no it is not. If consciousness is a prerequisite for intelligence or moral patience, then it is on the proponents to define exactly what that is and how to prove it.
On matters of moral patience especially, we should err on the side of caution, because the cost of getting it wrong is catastrophic.
On the other hand, they have no biology or hormones, no real experience with birth, life, death, joy, fear, empathy... that is all simulated emotion through language. They will never know what it's like to be horny or hungry and the satisfaction that comes with fulfilling those.
It's fascinating the hills people will die on. Let's even set aside the fact that LLM internals are mostly opaque but logical branching points ? Look at the Watson test. Most people fail a simple conditional reasoning problem unless it’s dressed up in familiar social context, like catching cheaters.
Where does this idea of general intelligence as logic automatons actually exist ? Because it's sure as hell not real life. Humans are not like this, Apes are not like this, Birds are not like this, Cetaceans are not like this. Fiction. Fiction is the only realm this reality of GI exists, so it's patently absurd when people hide behind it like anything that doesn't present as doing it must be obviously wrong.
(1) A far-reaching tendency to overextrapolate from the low-hanging fruit of the last few years of pretraining progress. GPT-2 to GPT-3 may have been a quantum leap, but GPT-3 to GPT-4 was not, and GPT-4 to 5 even less so.
The party has been kept going by RL and agents, but still, there is indeed a point of diminishing returns, not just relative to available compute but to how much training is possible when the entire intellectual output of humanity, plus a raft of synthetic data, has already been inhaled by the training process.
If one is to internalise the things that are said here on HN with regularity about model progress, and sentences ending with "yet" or "for now", then it would be easy to conclude that my 10 year-old son, who gained 3 inches of height last year, will be tallest structure on the planet by age 17.
(2) Inability to distinguish between technological, computational, and energetic limits of LLM capabilities vs. ontological / conceptual ones. There are some things LLMs cannot do, or at least do well, at any size, at infinite size and with infinite compute, simply due to the very nature of what LLMs are to begin with.
This latter topic receives almost no attention, except maybe from Gary Marcus and Yann LeCun. In that respect, this article is a breath of fresh air, insofar as it highlights that LLMs aren't "AI" at all, as we have traditionally understood the concept.
They really _are_ stochastic parrots. The relevant questions are about how much that matters for some domain or set of applications, not whether they are an emerging alien intelligence with civilisation-threatening capabilities.
How can we be so confident that there is anything LLMs fundamentally cannot do? Proving that impossibility seems hard.
Drawing trend lines far into the future is foolish, but the recent trend is clear and its not obvious how much more progress is needed before they start having a meaningful impact on more aspects of life.
> not whether they are an emerging alien intelligence with civilisation-threatening capabilities.
The people building them are explicitly attempting to do this. They may not succeed, but what if they do? Seems worth considering that scenario.
But arithmetically speaking, is any of that true? Is model progress truly "accelerating"? How can you compare the delta from GPT-2 to GPT-3 with the delta from GPT-5 to GPT-6 with no sense of irony? And, at the risk of being quite blunt, do you know the meaning of the term "exponential"?
On the MMLU benchmark, GPT-2 had an accuracy of 32.4%. GPT-3 improved this to 43.9%.
GPT-4 scored 86.4%. On GPQA Diamond, it scored 31%, vs GPT-5 at 86%. GPT-6 scores 96%.
If we take ARG-AGI-2, it would be 9.9% for GPT-5 vs 95% with GPT-6.
The benchmarks do not corroborate the picture you were drawing about improvements slowing down between GPT major versions.
I remember their capabilities like so:
GPT-3 could produce convincing looking texts, sometimes.
GPT-4 was somewhat smarter and would give more accurate answers. At that time image understanding was released, wasn't it?
Then with GPT-5 we have a reasoning model, another jump in capabilities.
Now, compare GPT-6 Astra with its ability to implement software, work on long horizon tasks, and visual understanding.
This is true of every frontier model. The real agentic coding breakthrough in Claude land was around Opus 4.5 or so, though it's hard to say if that might have just been coeval with the release of the Claude Code harness.
On the Chinese front (my preferred landscape because of the open weight aspect, not because I'm Chinese), Kimi has been perfectly viable since K2.5, DeepSeek since however many major releases ago, and GLM and MiMo and the rest of them too, although they all achieved breakout moments of publicity at different times. There's no denying that every release brings improvements, but if we're going to be honest, they're fairly marginal.
You're right that many of the less capable / open weights models are very good themselves, but that's part of showing how much more capable the frontier models are becoming. Cheaper (smaller) models are able to do the same things, in many cases better, as the way larger models were able to do a year ago. The efficiency is getting way better alongside the 'intelligence'. That'll also cause the models to become much faster over time which is the real improvement I'm waiting for.
I think a part of why many people aren't seeing big deltas if they're using it for code and they're actually coders is we actually sit down and reason out the project with the LLM. Someone who knows nothing about coding ~1 year ago wouldn't come even close to the kind of app they'd get now with the same instructions. If you just give it a task like "help me decompile this game" without knowing a whole lot about decompiling suddenly in the last 3 or so months you start to get some half-decent results.
The sea-change I see lately is about lifting up the people who couldn't do these things before more than the people who could. And I actually see problems coming from that already as people I've met think they can make enterprise grade software if they just tell their LLM to make it safe.
This is where applied business programmers part ways with philosophers of mind and cognitive scientists. However, the lack of an inner model of the world is a formidable limitation, while the reliability of purely statistical-inferential processes will never be adequate for some basic building blocks of modernity.
That proves the fact that the audience is not really agentic (not the ones that can cause or influence things). The agentic nature in people was killed long back with all education, rules and pervasive reach of the state into the lives of people. Common people are just subjects without an intention and will at this point. They are like those leaves that float in a stream, while the stream finds its way.
At best, all that a today's average person can do is, post sone interesting post or comment on HN, or the other social media and count their likes, like I do now. There ends their agentic effort.
Yeah, no. This is extremist nonsense that borders on delusions. Someone has to say it, apparently.
<3
Everyone who rejected the stochastic parrot argument has been right about AI progress and the increasing abilities of LLM's in math, coding, writing, etc. You would be laughed at in 2020 for claiming an LLM could write coherently or could possibly pass the Turing test.
Those who claim that AI aren't capable of "genuine judgement" or whatever rhetorical flourish they feel is most effective at dismissing AI capabilities would tell you in 2022 that AI would never be capable of coding, or in 2023 it would never be capable of mathematics beyond grade-school level. Those same people now will claim that AI could never come up with a genuine breakthrough in math, science, etc. Detractors of AI have a full time job of moving goalposts.
That being said, AI, once sufficiently powerful, will grant whoever controls it the ability to mold the world as they see fit, for better or worse. They will have access to superhuman mathematicians, coders, strategists, PR, financial engineering, legal skill, etc. Essentially an army of superhuman intellects which can afford them guaranteed success in any domain they wish to pursue. Anyone who doesn't see this as potentially the most destabilizing event in human history likely isn't thinking deeply enough about it, or is still knee deep in denial.
My dude, an LLM still can't write coherently.
A human mind keeps a personality that keeps generating thoughts in the same direction, unless some mind-altering force derails it. A LLM has no guiding stable personality of that kind, rather has millions of different personalities and which one you get is dictated by whatever information you put in its context window.
This statement belongs to the same epistemic class as Moon-landing denial. Flagged for kookery.
How can you possibly hold this belief after Tuesday? An LLM solved a significant number of major open problems in many subfields of mathematics, including a major breakthrough on the Riemann Hypothesis. Get your head out of the sand.
He is right that the AI giants would like a bit of regulatory capture.
If the parrot randomly yaps out a solution to an open math problem, it's a useful parrot.
The fact that it sometimes says things that aren't useful, well, old intelligence does that all the time.
If you've used AI recently, you can't avoid having found it useful. Just general day-to-day intelligence like finding things in your email, writing code for you, that kind of stuff, it does well. People have some weird hangups about it being stochastic, when it doesn't really matter.
Not anything interesting here that hasn't already been stated 20 other times.
> The dangers posed by this technology on its present trajectory are best understood as simple accelerants for the existing dangers humans pose to themselves.
Is very true. All the doomspeak "aI EsCaPeD AnD HaCkEd aNoThEr sYsTeM" was a human error all along, and the tech was doing what it is programmed to do, retry until a satisifiable outcome is met.
Accelerating the ability for any given person to have access to some amount of these capabilities is terrifying, is it not?
Maybe many were too young for the cold war. But it really does feel like we're setting a new one up.
With dangerous chemicals and radioactive materials these are generally pretty easy to classify. This said, after the physical behaviors, the above technologies can induce behavior changes in humans because they exist. After nukes, a lot more people had issues with extental dread. It also causes n-order interactions that get very hard to predict. For example MAD or laws against using chemical weapons.
And these are weapons with a very easy to map out causal chain in their own behavior. You can't prompt a chemical weapon to make a decision.
We come up with agentic AI and all of a sudden a lot of rules on causality break. In a harness they are a causal agent. They can interact with a digital reality, and with the right tools and/or human contacts they can causally effect analog reality.
Anyone that says 'programmed' about an LLM almost certainly does not know what they are talking about and can be summarily ignored.
> interesting if true.
Not sure what you mean by this. This is obviously about MAGA and other allied political forces in the US, at the very least. I wouldn't call it interesting to document the recent observed behavior. It's been evident through journalistic analysis of public statements, policy, and investment for most of the year.
> The hope of the Elons and Sams of the world is that this will come to fruition well enough that we only have to bother sustaining the "cream of the crop" in the labor force […]. So basically the plan here is to let things go to shit, and when billions of people die as a natural consequence of that, things will keep running as desired.
For one, I don’t think that Sam Altman can be characterized as MAGA (https://en.wikipedia.org/wiki/Sam_Altman#Political_engagemen...). Secondly, I’d be interested in sources substantiating a plan that explicitly includes billions of people dying as a necessary factor for the plan to work out.
He is not usually associated with MAGA. The reference was not directed at the heads of the most well know tech companies, despite the quote about "Elons and Sams". This wouldn't make sense in context. It was a reference to the ruling political regime.
This is the paragraph in the entirety. While previous sections referred to “the Elons and Sams of the world”, this section that was specifically called out:
> The connection made to climate change in the latter part of the article is new to me, and would be interesting if true. However, I wish the author would cite supporting evidence for their conclusions.
Does not refer to them explicitly (many of the very same), although it could be believed that the second clause of the first sentence refers to Sam, et al. Elon doesn't fit as he has no fiduciary duty nor has he called for slowdown, nor are there "many" in this demographic, which is a sign that this is more stream of thought. I believe the narrative only fits if you consider this to be a tangent about political sociopaths that was muddled by a very human leading thought intruding (from previous paragraphs), on an otherwise straightforward observation.
It makes no sense in the context of billionaires, as they were not generally driving "us" toward climate destruction prior to the AI boom.
Frankly, the article is complete nonsense, and these political views are extremist to the point where it would be fair to suggest mental counseling.
We don't need to pretend it's a great, interesting article just because the opinion goes in a direction we might agree with.