The article says that LLMs don't summarize, only shorten, because...
"A true summary, the kind a human makes, requires outside context and reference points. Shortening just reworks the information already in the text."
Then later says...
"LLMs operate in a similar way, trading what we would call intelligence for a vast memory of nearly everything humans have ever written. It’s nearly impossible to grasp how much context this gives them to play with"
So, they can't summarize, because they lack context... but they also have an almost ungraspably large amount of context?
Good article, it's been told before but it bears repeating.
Also I got caught on this one kind of irrelevant point regarding the characterization of the Matrix: I would say Matrix is not just diguised as a story about escaping systems of control, it's quite clearly about oppressive systems in society, with specific reference to gender expression. Lilly Wachowski has explicitly stated that it was supposed to be an allegory for gender transition.
Regarding Timmy, the Companion Cube from the game Portal is the greatest example of induced anthropomorphism that I've ever experienced. If you know, you know, and if you don't, you should really play the game, since it's brilliant.
LLMs give a very strong appearance of intelligence, because humans are super receptive to information provided via our native language. We often have to deal with imperfect speakers and writers, and we must infer context and missing information on our own. We do this so well that we don't know we're doing it. LLMs have perfect grammar and we subtly feel that they are extremely smart because subconsciously we recognize that we don't have to think about anything that's said, it is all syntactically perfect.
So, LLMs sort of trick us into masking their true limitations and believing that they are truly thinking; there are even models that call themselves thinking models, but they don't think, they just predict what the user is going to complain about and say that to themselves as an additional, dynamic prompt on top of the one you actually enter.
LLMs are very good at fooling us into the idea that they know anything at all; they don't. And humans are very bad at being discriminate about the source of the information presented to them if it is presented in a friendly way. The combination of those things is what has resulted in the insanely huge AI hype cycle that we are currently living in the middle of. Nearly everyone is overreacting to what LLMs actually are, and the few of us that believe that we sort of see what's actually happening are ignored for being naysayers, buzz-kills, and luddites. Shunned for not drinking the Kool-Aid.
> LLMs mimic intelligence, but they aren’t intelligent.
I see statements like this a lot, and I find them unpersuasive because any meaningful definition of "intelligence" is not offered. What, exactly, is the property that humans (allegedly) have and LLMs (allegedly) lack, that allows one to be deemed "intelligent" and the other not?
I see two possibilities:
1. We define "intelligence" as definitionally unique to humans. For example, maybe intelligence depends on the existence of a human soul, or specific to the physical structure of the human brain. In this case, a machine (perhaps an LLM) could achieve "quacks like a duck" behavioral equality to a human mind, and yet would still be excluded from the definition of "intelligent." This definition is therefore not useful if we're interested in the ability of the machine, which it seems to me we are. LLMs are often dismissed as not "intelligent" because they work by inferring output based on learned input, but that alone cannot be a distinguishing characteristic, because that's how humans work as well.
2. We define "intelligence" in a results-oriented way. This means there must be some specific test or behavioral standard that a machine must meet in order to become intelligent. This has been the default definition for a long time, but the goal posts have shifted. Nevertheless, if you're going to disparage LLMs by calling them unintelligent, you should be able to cite a specific results-oriented failure that distinguishes them from "intelligent" humans. Note that this argument cannot refer to the LLMs' implementation or learning model.
I think it has to include some measure of Agency. You can load up the most impressive LLM out there and if you don't give it any instructions, IT WON'T DO ANYTHING.
I don’t really think one needs to define intelligence to be able to acknowledge that inability to distinguish fact from fiction, or even just basic cognition and awareness of when it’s uncertain, telling the truth, or lying — is a glaring flaw in claiming intelligence. Real intelligence doesn’t have an effective stroke from hearing a username (token training errors); this is when you are peeling back the curtain of the underlying implementation and seeing its flaws.
If we measure intelligence as results oriented, then my calculator is intelligent because it can do math better than me; but that’s what it’s programmed/wired to do. A text predictor is intelligent at predicting text, but it doesn’t mean it’s general intelligence. It lacks any real comprehension of the model or world around it. It just know words, and
Is this shocking? We don't have a rigorous definition of intelligence so doesn't it make sense? The question isn't about such a goal post moving so much about how it is moving. It is perfectly acceptable for it to be refined while it wouldn't be to rewrite the definition in a way that isn't similar to the previous one.
So I think there are a lot more than your two possibilities. I mean psychologists and neuroscientists have been saying for decades that tests aren't a precise way to measure knowledge or intelligence, but that it is still a useful proxy.
> "quacks like a duck" behavioral
I see this phrase used weirdly frequently. The duck test is
| If it looks like a duck, swims like a duck, and quacks like a duck, then it ***probably*** is a duck.
I emphasize probably because the duck test doesn't allow you to distinguish a duck from a highly sophisticated animatronic. It's a good test, don't get me wrong, but that "probably" is a pretty important distinction.
I think if we all want to be honest, the reality is "we don't know". There's arguments to be made in both directions and with varying definitions of intelligence with different nuances involved. I think these arguments are fine as they make us refine our definitions but I think they can also turn to be entirely dismissive and that doesn't help us refine and get closer to the truth. We all are going to have opinions on this stuff but frankly, the confidence of our opinions needs to be proportional to the amount of time and effort spent studying the topic. I mean the lack of a formal definition means nuances dominate the topic. Even if things are simple once you understand them that doesn't mean they aren't wildly complex before that. I mean I used to think Calculus was confusing and now I don't. Same process but not on an individual scale.
> I emphasize probably because the duck test doesn't allow you to distinguish a duck from a highly sophisticated animatronic. It's a good test, don't get me wrong, but that "probably" is a pretty important distinction.
Why is it an important distinction? The relevance of the duck test is that if you can't tell a duck from a non-duck, then the non-duck is sufficiently duck-like for the difference to not matter.
I agree with your basic argument: intelligence is ill-defined and human/LLM intelligence being indistinguishable IS the basis for the power of these models.
But the point of the article is a distinct claim: personification of a model, expecting human or even human-like responses is a bad idea. These models can be held responsible for their answers independently because they are tools. They should be used as tools until they are powerful enough to be responsible for their actions and interactions legally.
But we're not there. These are tools. With tool limitations.
What if the problem is not that we overestimate LLMs, but that we overestimate intelligence? Or to express the same idea for a more philosophically inclined audience, what if the real mistake isn’t in overestimating LLMs, but in overestimating intelligence itself by imagining it as something more than a web of patterns learned from past experiences and echoed back into the world?
LLM's can shorten and maybe tend to if you just say "summarize this" but you can trivially ask them to do more. I asked for a summary of Jenson's post and then offer a reflection, GPT-5 said, "It's similar to the Plato’s Cave analogy: humans see shadows (the input text) and infer deeper reality (context, intent), while LLMs either just recite shadows (shorten) or imagine creatures behind them that aren’t there (hallucinate). The “hallucination” behavior is like adding “ghosts”—false constructs that feel real but aren’t grounded.
That ain't shortening because none of that was in his post.
Who are you going to lodge your complaint to that the set of systems and machines that just took your job isn’t “intelligent?”
Humans seem to get wrapped around these concepts like intelligence consciousness etc. because they seem to be the only thing differentiating us from every other animal when in fact it’s all a mirage.
Seems like this is close to the Uncanny Valley effect.
LLM intelligence is in the spot where it is simultaneously genius-level but also just misses the mark a tiny bit, which really sticks out for those who have been around humans their whole lives.
I feel that, just like more modern CGI, this will slowly fade with certain techniques and you just won't notice it when talking to or interacting with AI.
Just like in his post during the whole Matrix discussion.
> "When I asked for examples, it suggested the Matrix and even gave me the “Summary” and “Shortening” text, which I then used here word for word. "
He switches in AI-written text and I bet you were reading along just the same until he pointed it out.
You can compare the current state of LLMs to the days of chess machines when they first approached grandmaster level play. The machine approach was very brute force, and there was a lot of work done to improve the sheer amount of look ahead that was required to complete at the grandmaster level.
As opposed to what grandmasters actually did, which was less look ahead and more pattern matching to strengthen the position.
Now LLMs successfully leverage pattern matching, but interestingly it is still a kind of brute force pattern matching, requiring the statistical absorption of all available texts, far more than a human absorbs in a lifetime.
This enables the LLM to interpolate an answer from the structure of the absorbed texts with reasonable statistical relevance. This is still not quite “what humans do” as it still requires brute force statistical analysis of vast amounts of text to achieve pretty good results. For example training on all available Python sources in github and elsewhere (curated to avoid bad examples) yields pretty good results, not how a human would do it, but statistically likely to be pertinent and correct.
I feel this article should be paired with this other one [1] that was on the frontpage a few days ago.
My impression is, there is currently one tendency to "over-anthropomorphize" LLMs and treat them like conscious or even superhuman entities (encouraged by AI tech leaders and AGI/Singularity folks) and another to oversimplify them and view them as literal Markov chains that just got lots of training data.
Maybe those articles could help guarding against both extremes.
Even stronger than our need to anthropomorphize seems to be our innate desire to believe our species is special, and that “real intelligence” couldn’t ever be replicated.
If you keep redefining real intelligence as the set of things machines can’t do, then it’s always going to be true.
I might be mixing the concepts of intelligence and conscience etc, but the human mind is more than language and data; it's also experience. LLMs have all the data and can express anything around that context, but will never experience anything, which is singular for each of us, and it's part of what makes what we call intelligence (?). So they will never replicate the human mind; they can just mimic it.
I heard from Miguel Nicolelis that language is a filter for the human mind, so you can never build a mind from language. I interpreted this like trying to build an orange from its juice.
The author's argument is built on fallacies that always pop up in these kinds of critiques.
The "summary vs shortening" distinction is moving the goalposts. They makes the empirical claim that LLMs fail at summarizing novel PDFs without any actual evidence. For a model trained on a huge chunk of the internet, the line between "reworking existing text" and "drawing on external context" is so blurry it's practically meaningless.
Similarly, can we please retire the ELIZA and Deep Blue analogies? Comparing a modern transformer to a 1960s if-then script or a brute-force chess engine is a category error. It's a rhetorical trick to make LLMs seem less novel than they actually are.
And blaming everything on anthropomorphism is an easy out. It lets you dismiss the model's genuinely surprising capabilities by framing it as a simple flaw in human psychology. The interesting question isn't that we anthropomorphize, but why this specific technology is so effective at triggering that response from humans.
The whole piece basically boils down to: "If we define intelligence in a way that is exclusively social and human, then this non-social, non-human thing isn't intelligent." It's a circular argument.
- LLMs don't need to be intelligent to take jobs, bash scripts have replaced people.
- Even if CEOs are completely out of touch and the tool can't do the job you can still get laid off in an ill informed attempt to replace you. Then when the company doesn't fall over because the leftover people, desperate to keep covering rent fill the gaps it just looks like efficiency to the top.
- I don't think our tendency anthropomorphize LLMs is really the problem here.
LLMs mimic intelligence, but they aren’t intelligent.
They aren’t just intelligence mimics, they are people mimics, and they’re getting better at it with every generation.
Whether they are intelligent or not, whether they are people or not, it ultimately does not matter when it comes to what they can actually do, what they can actually automate. If they mimic a particular scenario or human task well enough that the job gets done, they can replace intelligence even if they are “not intelligent”.
If by now someone still isn’t convinced that LLMs can indeed automate some of those intelligence tasks, then I would argue they are not open to being convinced.
> A philosophical exploration of free will and reality disguised as a sci-fi action film about breaking free from systems of control.
How is that a summary? It reads as a one-liner review I would leave on Letterboxed or something I would say, trying to be pretentious and treating the movie as a work of art. It is a work of art, because all movies are art, but that's an awful summary.
37 comments
[ 2.9 ms ] story [ 59.1 ms ] thread"A true summary, the kind a human makes, requires outside context and reference points. Shortening just reworks the information already in the text."
Then later says...
"LLMs operate in a similar way, trading what we would call intelligence for a vast memory of nearly everything humans have ever written. It’s nearly impossible to grasp how much context this gives them to play with"
So, they can't summarize, because they lack context... but they also have an almost ungraspably large amount of context?
Also I got caught on this one kind of irrelevant point regarding the characterization of the Matrix: I would say Matrix is not just diguised as a story about escaping systems of control, it's quite clearly about oppressive systems in society, with specific reference to gender expression. Lilly Wachowski has explicitly stated that it was supposed to be an allegory for gender transition.
Scott Jenson is one of my favorite authors.
He's really big on integrating an understanding of basic human nature, into design.
How do you know LLMs aren't intelligent, if you can't define what that means?
LLMs give a very strong appearance of intelligence, because humans are super receptive to information provided via our native language. We often have to deal with imperfect speakers and writers, and we must infer context and missing information on our own. We do this so well that we don't know we're doing it. LLMs have perfect grammar and we subtly feel that they are extremely smart because subconsciously we recognize that we don't have to think about anything that's said, it is all syntactically perfect.
So, LLMs sort of trick us into masking their true limitations and believing that they are truly thinking; there are even models that call themselves thinking models, but they don't think, they just predict what the user is going to complain about and say that to themselves as an additional, dynamic prompt on top of the one you actually enter.
LLMs are very good at fooling us into the idea that they know anything at all; they don't. And humans are very bad at being discriminate about the source of the information presented to them if it is presented in a friendly way. The combination of those things is what has resulted in the insanely huge AI hype cycle that we are currently living in the middle of. Nearly everyone is overreacting to what LLMs actually are, and the few of us that believe that we sort of see what's actually happening are ignored for being naysayers, buzz-kills, and luddites. Shunned for not drinking the Kool-Aid.
I see statements like this a lot, and I find them unpersuasive because any meaningful definition of "intelligence" is not offered. What, exactly, is the property that humans (allegedly) have and LLMs (allegedly) lack, that allows one to be deemed "intelligent" and the other not?
I see two possibilities:
1. We define "intelligence" as definitionally unique to humans. For example, maybe intelligence depends on the existence of a human soul, or specific to the physical structure of the human brain. In this case, a machine (perhaps an LLM) could achieve "quacks like a duck" behavioral equality to a human mind, and yet would still be excluded from the definition of "intelligent." This definition is therefore not useful if we're interested in the ability of the machine, which it seems to me we are. LLMs are often dismissed as not "intelligent" because they work by inferring output based on learned input, but that alone cannot be a distinguishing characteristic, because that's how humans work as well.
2. We define "intelligence" in a results-oriented way. This means there must be some specific test or behavioral standard that a machine must meet in order to become intelligent. This has been the default definition for a long time, but the goal posts have shifted. Nevertheless, if you're going to disparage LLMs by calling them unintelligent, you should be able to cite a specific results-oriented failure that distinguishes them from "intelligent" humans. Note that this argument cannot refer to the LLMs' implementation or learning model.
If we measure intelligence as results oriented, then my calculator is intelligent because it can do math better than me; but that’s what it’s programmed/wired to do. A text predictor is intelligent at predicting text, but it doesn’t mean it’s general intelligence. It lacks any real comprehension of the model or world around it. It just know words, and
So I think there are a lot more than your two possibilities. I mean psychologists and neuroscientists have been saying for decades that tests aren't a precise way to measure knowledge or intelligence, but that it is still a useful proxy.
I see this phrase used weirdly frequently. The duck test is I emphasize probably because the duck test doesn't allow you to distinguish a duck from a highly sophisticated animatronic. It's a good test, don't get me wrong, but that "probably" is a pretty important distinction.I think if we all want to be honest, the reality is "we don't know". There's arguments to be made in both directions and with varying definitions of intelligence with different nuances involved. I think these arguments are fine as they make us refine our definitions but I think they can also turn to be entirely dismissive and that doesn't help us refine and get closer to the truth. We all are going to have opinions on this stuff but frankly, the confidence of our opinions needs to be proportional to the amount of time and effort spent studying the topic. I mean the lack of a formal definition means nuances dominate the topic. Even if things are simple once you understand them that doesn't mean they aren't wildly complex before that. I mean I used to think Calculus was confusing and now I don't. Same process but not on an individual scale.
Why is it an important distinction? The relevance of the duck test is that if you can't tell a duck from a non-duck, then the non-duck is sufficiently duck-like for the difference to not matter.
But the point of the article is a distinct claim: personification of a model, expecting human or even human-like responses is a bad idea. These models can be held responsible for their answers independently because they are tools. They should be used as tools until they are powerful enough to be responsible for their actions and interactions legally.
But we're not there. These are tools. With tool limitations.
That ain't shortening because none of that was in his post.
Humans seem to get wrapped around these concepts like intelligence consciousness etc. because they seem to be the only thing differentiating us from every other animal when in fact it’s all a mirage.
LLM intelligence is in the spot where it is simultaneously genius-level but also just misses the mark a tiny bit, which really sticks out for those who have been around humans their whole lives.
I feel that, just like more modern CGI, this will slowly fade with certain techniques and you just won't notice it when talking to or interacting with AI.
Just like in his post during the whole Matrix discussion.
> "When I asked for examples, it suggested the Matrix and even gave me the “Summary” and “Shortening” text, which I then used here word for word. "
He switches in AI-written text and I bet you were reading along just the same until he pointed it out.
This is our future now I guess.
As opposed to what grandmasters actually did, which was less look ahead and more pattern matching to strengthen the position.
Now LLMs successfully leverage pattern matching, but interestingly it is still a kind of brute force pattern matching, requiring the statistical absorption of all available texts, far more than a human absorbs in a lifetime.
This enables the LLM to interpolate an answer from the structure of the absorbed texts with reasonable statistical relevance. This is still not quite “what humans do” as it still requires brute force statistical analysis of vast amounts of text to achieve pretty good results. For example training on all available Python sources in github and elsewhere (curated to avoid bad examples) yields pretty good results, not how a human would do it, but statistically likely to be pertinent and correct.
My impression is, there is currently one tendency to "over-anthropomorphize" LLMs and treat them like conscious or even superhuman entities (encouraged by AI tech leaders and AGI/Singularity folks) and another to oversimplify them and view them as literal Markov chains that just got lots of training data.
Maybe those articles could help guarding against both extremes.
[1] https://www.verysane.ai/p/do-we-understand-how-neural-networ...
If you keep redefining real intelligence as the set of things machines can’t do, then it’s always going to be true.
This is right out of Community
I heard from Miguel Nicolelis that language is a filter for the human mind, so you can never build a mind from language. I interpreted this like trying to build an orange from its juice.
The "summary vs shortening" distinction is moving the goalposts. They makes the empirical claim that LLMs fail at summarizing novel PDFs without any actual evidence. For a model trained on a huge chunk of the internet, the line between "reworking existing text" and "drawing on external context" is so blurry it's practically meaningless.
Similarly, can we please retire the ELIZA and Deep Blue analogies? Comparing a modern transformer to a 1960s if-then script or a brute-force chess engine is a category error. It's a rhetorical trick to make LLMs seem less novel than they actually are.
And blaming everything on anthropomorphism is an easy out. It lets you dismiss the model's genuinely surprising capabilities by framing it as a simple flaw in human psychology. The interesting question isn't that we anthropomorphize, but why this specific technology is so effective at triggering that response from humans.
The whole piece basically boils down to: "If we define intelligence in a way that is exclusively social and human, then this non-social, non-human thing isn't intelligent." It's a circular argument.
- Even if CEOs are completely out of touch and the tool can't do the job you can still get laid off in an ill informed attempt to replace you. Then when the company doesn't fall over because the leftover people, desperate to keep covering rent fill the gaps it just looks like efficiency to the top.
- I don't think our tendency anthropomorphize LLMs is really the problem here.
They aren’t just intelligence mimics, they are people mimics, and they’re getting better at it with every generation.
Whether they are intelligent or not, whether they are people or not, it ultimately does not matter when it comes to what they can actually do, what they can actually automate. If they mimic a particular scenario or human task well enough that the job gets done, they can replace intelligence even if they are “not intelligent”.
If by now someone still isn’t convinced that LLMs can indeed automate some of those intelligence tasks, then I would argue they are not open to being convinced.
How is that a summary? It reads as a one-liner review I would leave on Letterboxed or something I would say, trying to be pretentious and treating the movie as a work of art. It is a work of art, because all movies are art, but that's an awful summary.