"Explain how to solve" and "write like X" are crucially different tasks. One of them is about going through the steps of a process, and the other is about mimicking the result of a process.
Neural networks most certainly go through a process to transform input into output (even to mimic the results of another process) but it's a very different one from human neutral networks. But I think this is the crucial point of the debate, essentially unchanged from Searle's "Chinese Room" argument from decades ago.
The person in that room, looking up a dictionary with Chinese phrases and patterns, certainly follows a process, but it's easy to dismiss the notion that the person understands Chinese. But the question is if you zoom out, is the room itself intelligent because it is following a process, even if it's just a bunch of pattern recognition?
like OP originally said, the LLM doesn't have access to the actual process of the author, only the completed/refined output.
Not sure why you need a concrete example to "test", but just think about the fact that the LLM has no idea how a writer brainstorms, re-iterates on their work, or even comes up with the ideas in the first place.
This isn't true in general, and not even true in many specific cases, because a great deal of writers have described the process of writing in detail and all of that is in their training data. Claude and chatgpt very much know how novels are written, and you can go into claude code and tell it you want to write a novel and it'll walk you through quite a lot of it -- worldbuilding, characters, plotting, timelines, etc.
It's very true that LLMs are not good at "ideas" to begin with, though.
Professional writer here. On our longer work, we go through multiple iterations, with lots of teardowns and recalibrations based on feedback from early, private readers, professional editors, pop culture -- and who knows. You won't find very clear explanations of how this happens, even in writers' attempts to explain their craft. We don't systematize it, and unless we keep detailed in-process logs (doubtful), we can't even reconstruct it.
It's certainly possible to mimic many aspects of a notable writer's published style. ("Bad Hemingway" contests have been a jokey delight for decades.) But on the sliding scale of ingenious-to-obnoxious uses for AI, this Grammarly/Superhuman idea feels uniquely misguided.
The distinction being made is the difference between intellectual knowledge and experience, not originality.
Imagine a interviewing a particularly diligent new grad. They've memorized every textbook and best practices book they can find. Will that alone make them a senior+ developer, or do they need a few years learning all the ways reality is more complicated than the curriculum?
Let's take the work of Raymond Carver as just one example. He would type drafts which would go through repeated iteration with a massive amount of hand-written markup, revision and excision by his editor.
To really recreate his writing style, you would need the notes he started with for himself, the drafts that never even made it to his editor, the drafts that did make to the editor, all the edits made, and the final product, all properly sequenced and encoded as data.
In theory, one could munge this data and train an LLM and it would probably get significantly better at writing terse prose where there are actually coherent, deep things going on in the underlying story (more generally, this is complicated by the fact that many authors intentionally destroy notes so their work can stand on its own--and this gives them another reason to do so). But until that's done, you're going to get LLMs replicating style without the deep cohesion that makes such writing rewarding to read.
A good point. "Famous author" is a marketing term for Grammarly here; it's easy to conceive of an "author" as being an individual that we associate with a finite set of published works, all of which contain data.
But authors have not done this work alone. Grammarly is not going to sell "get advice from the editorial team at Vintage" or "Grammarly requires your wife to type the thing out first, though"
I'll also note that no human would probably want advice from the living versions of the author themselves.
i don't buy this logic. if i have studied an author greatly i will be able to recognise patterns and be able to write like them.
ex: i read a lot of shakespeare, understand patterns, understand where he came from, his biography and i will be able to write like him. why is it different for an LLM?
You will produce output that emulates the patters of Shakespeare's works, but you won't arrive at them by the same process Shakespeare did. You are subject to similar limitations as the llm in this case, just to a lesser degree (you share some 'human experience' with the author, and might be able to reason about their though process from biographies and such)
As another example, I can write a story about hobbits and elves in a LotR world with a style that approximates Tolkien. But it won't be colored by my first-hand WW1 experiences, and won't be written with the intention of creating a world that gives my conlangs cultural context, or the intention of making a bedtime story for my kids. I will never be able to write what Tolkien would have written because I'm not Tolkien, and do not see the world as Tolkien saw it. I don't even like designing languages
that's fair and you have highlighted a good limitation. but we do this all the time - we try to understand the author, learn from them and mimic them and we succeed to good extent.
that's why we have really good fake van gogh's for which a person can't tell the difference.
of course you can't do the same as the original person but you get close enough many times and as humans we do this frequently.
in the context of this post i think it is for sure possible to mimic a dead author and give steps to achieve writing that would sound like them using an LLM - just like a human.
Not everything works like integrals. Some things don't have a standard process that everyone follows the same way.
Editing is one of these things. There can be lots of different processes, informed by lots of different things, and getting similar output is no guarantee of a similar process.
The process is irrelevant if the output is the same, because we never observe the process. I assume you are arguing that the outputs are not guaranteed to be the same unless you reproduce the process.
If we are talking about human artifacts, you never have reproducibility. The same person will behave differently from one moment to the next, one environment to another. But I assume you will call that natural variation. Can you say that models can't approximate the artifacts within that natural variation?
It's relevant for data it hasn't been trained on. LLMs are trained to be all-knowing which is great as a utility but that does not come close to capturing an individual.
If I trained (or, more likely, fine-tuned) an LLM to generate code like what's found in an individual's GitHub repositories, could you comfortably say it writes code the same way as that individual? Sure, it will capture style and conventions, but what about our limitations? What do you think happens if you fine-tune a model to write code like a frontend developer and ask it to write a simple operating system kernel? It's realistically not in their (individual) data but the response still depends on the individual's thought process.
I don't know if LLMs are trained to imitate sources like that. I also don't know what would happen if you asked it to do something like someone who does not know how to do it. Would they refuse, make mistakes, or assume the person can learn? Humans can do all three, so barring more specific instructions any such response is reasonable.
> Humans can do all three, so barring more specific instructions any such response is reasonable.
Of course, but reasonable behavior across all humans is not the same as what one specific human would do. An individual, depending on the scenario, might stick to a specific choice because of their personality etc. which is not always explained, and heavily summarized if it is.
>If I trained (or, more likely, fine-tuned) an LLM to generate code like what's found in an individual's GitHub repositories, could you comfortably say it writes code the same way as that individual? Sure, it will capture style and conventions, but what about our limitations? What do you think happens if you fine-tune a model to write code like a frontend developer and ask it to write a simple operating system kernel? It's realistically not in their (individual) data but the response still depends on the individual's thought process.
Look, I don't think you understand how LLM's work. Its not about fine tuning. Its about generalised reasoning. The key word is "generalised" which can only happen if it has been trained on literally everything.
> It's relevant for data it hasn't been trained on
LLM's absolutely can reason on and conceptualise on things it has not been trained on, because of the generalised reasoning ability.
> LLM's absolutely can reason on and conceptualise on things it has not been trained on, because of the generalised reasoning ability.
Yes, but how does that help it capture the nuances of an individual? It can try to infer but it will not have enough information to always be correct, where correctness is what the actual individual would do.
i think there's a lot to be said about the process as well, the motivations, the intuitions, life experiences, and seeing the world through a certain lens. this creates for more interesting writing even when you are inspired by a certain past author. if you simply want to be a stochastic parrot that replicates the style of hemingway, it's not that difficult, but you'll also _likely_ have an empty story and you can extend the same concept to music
Even if the visualization of the integration process via steps typed out in the chat interface is the same as what you would have done on paper, the way the steps were obtained is likely very different for you and LLM. You recognized the integral's type and applied corresponding technique to solve it. LLM found the most likely continuation of tokens after your input among all the data it has been fed, and those tokens happen to be the typography for the integral steps. It is very unlikely are you doing the same, i.e. calculating probabilities of all the words you know and then choosing the one with the highest probability of being correct.
You are not able to write like Shakespeare. Shakespeare isn't really even a great example of an "author" per se. Like anybody else you could get away with: "well I read a lot of Bukowski and can do a passable imitation" or "I'm a Steinbeck scholar and here's a description of his style." But not Shakespeare.
I get that you're into AI products and ok, fine. But no you have not "studied [Shakespeare] greatly" nor are you "able to write like [Shakespeare]." That's the one historical entity that you should not have chosen for this conversation.
This bot is likely just regurgitating bits from the non-fiction writing of authors like an animatronic robot in the Hall of Presidents. Literally nobody would know if the LLM was doing even a passable job of Truman Capote-ing its way through their half-written attempt at NaNoWriMo
You can understand his biography and analyses about how shakespeare might have written. You can apply this knowledge to modify your writing process.
The LLM does not model text at this meta-level. It can only use those texts as examples, it cannot apply what is written there to it's generation process.
Yes, what I said should be falsifiable. The burden is on you to give me an example, but I can give you an idea.
You need to show me an LLM applying writing techniques do not have examples in its corpus.
You would have to use some relatively unknown author, I can suggest Iida Turpeinen. There will be interviews of her describing her writing technique, but no examples that aren't from Elolliset (Beasts of the sea).
Because the entire point is the LLM cannot understand text about text.
If someone has already done the work of giving an example of how to produce text according to a process, we have no way of knowing if the LLM has followed the process or copied the existing example.
And my point of course is that copying examples is the only way that LLMs can produce text. If you use an author who has been so analyzed to death that there are hundreds of examples of how to write like them, say, Hemingway, then that would not prove anything, because the LLM will just copy some existing "exercise in writing like Hemingway".
>Because the entire point is the LLM cannot understand text about text.
you have asked for an LLM to read a single interview and produce text that sounds similar to the author based on the techniques on that single interview.
There is no actual short story behind the link? moon_landing_turpeinen.md cannot be opened.
You could not have done better? Love it. You didn't even bother rewriting my post before pasting it into the box. The post isn't addressed as a prompt, it's my giving you the requirements of what to prompt.
Also, because you did that, you've actually provided evidence for my argument: notice that my attitudes about LLMs are reflected in the LLM output. E.g.:
"Now — the honest problem the challenge identifies: I'm reconstructing a description of a style, not internalizing the rhythm and texture of actual prose. A human who's read the book would have absorbed cadences, sentence lengths, paragraph structures, the specific ratio of concrete detail to abstraction — all the things that live below the level of "technique described in interviews.""
That's precisely because it can't separate metatext from text. It's just copying the vibe of what I'm saying, instead of understanding the message behind the text and trying to apply it. It also hallucinates somewhat here, because it's argument is about humans absorbing the text rather than the metatext. But that's also to be expected from a syntax-level tool like an LLM.
The end result is... nothing. You failed the task and you ended up supporting my point. But I appreciate that you took the time to do this experiment.
> "Now — the honest problem the challenge identifies: I'm reconstructing a description of a style, not internalizing the rhythm and texture of actual prose. A human who's read the book would have absorbed cadences, sentence lengths, paragraph structures, the specific ratio of concrete detail to abstraction — all the things that live below the level of "technique described in interviews.
a human would have to read all the text, so would an LLM but you have not allowed this from your previous constraint. then allow an LLM to reproduce something that is in its training set?
why do you expect an LLM to achieve something that even a human can't do?
Why are you taking the LLM-hallucinated version of the argument as truth? I even clearly stated how the LLM-version of my claim is a misunderstood version of the argument.
Do you remember the point we're arguing? That a human can understand text about a way of writing, and apply that information to the _process_ of writing (not the output).
If you admit the LLM can't do this, then you are conceding the point.
I don't know why you're claiming that humans can't do this when we very clearly can.
An illustrative example: I could describe a new way of rhyming to a human without an example, and they could produce a rhyme without an example. However describing this new rhyming scheme to an LLM without examples would not yield any results. (Rhyming is a bad example to test, however, because the LLM corpi have plenty of examples).
The point is that you dont become Jimi Hendrix or Eric Clapton even if you spend 20 years playing on a cover band. You can play the style, sound like but you wont create their next album.
Not being Jimi Hendrix or Eric Clapton is the context you are missing. LLMs are Cover Bands...
Actually this is the crux and the nuance which makes discussing LLM specifics a pain in the general space.
If you built an LLM exclusively on the writings and letters of John Steinbeck, you could NOT tell the LLM to solve an integral for you amd expect it to be right.
Instead what you will receive is a text that follows a statistically derived most likely (in accordance to the perplexity tuning) response to such a question.
> If you built an LLM exclusively on the writings and letters of John Steinbeck, you could NOT tell the LLM to solve an integral for you amd expect it to be right.
Isn't this obvious? There is not enough latent knowledge of math there to enable current LLMs to approximate anything resembling an integral.
It isnt obvious to the person I am responding to, and it isnt obvious to majority of individuals I speak with on the matter (which is why AI, personally, is in the bucket of religion amd politics for polite conversation to simply avoid)
Wait -- I'm fairly certain this is obvious to the person you were responding to. It may not be obvious to a lay person (who may not even know LLMs are trained at all). But I think this is obvious to almost all people with even a small understanding of LLMs.
I'm actually pretty convinced they're a troll or at the very least a high confrontation participant who is quick to move goal posts, ignore entire chains of logic, engage in ad hominim attacks of other posters, and is bringing zero novel insight anywhere in this thread
one of the posters said it can't even reason through chess, i ran the actual benchmark, spent money and actually proved that it can beat a 1000 elo chess engine.
It’s obvious to me. What point are you trying to make? It’s not religion it’s falsifiable easily.
LLMs can reason about integrals as well as in a literature context. You suggested that if it’s not trained on literature then it can’t reason about it. But why does that matter?
Now what if we ask the LLM to write about social media? Do you think the output would be similar to what you'd get if we had a time machine to bring the actual man back and have him form his own thoughts firsthand?
>If you built an LLM exclusively on the writings and letters of John Steinbeck, you could NOT tell the LLM to solve an integral for you amd expect it to be right.
this shows that you have very less idea on how llm's work.
LLM that is trained only on john steinbeck will not work at all. it simply does not have the generalised reasoning ability. it necessarily needs inputs from every source possible including programming and maths.
You have completely ignored that LLMs have _generalised_ reasoning ability that it derives from disparate sources.
LLMs have the ability to convince you that they've "reasoned". sometimes, an application will loop the output of an LLM to its input to provide a "chain of reasoning"
This is not the same thing as reasoning.
LLMs are pattern matchers. If you trained an llm only to map some input to the output of John Steinbeck, then by golly that's what it'll be able to do. If you give it some input that isn't suitably like any of the input you gave it during training, then you'll get some unpredictable nonsense as output.
> If you trained an llm only to map some input to the output of John Steinbeck
this is literally not possible because the llm does not get generalised reasoning ability. this is not a useful hypothetical because such an llm will simply not work. why do you think you have never seen a domain specific model ever?
if you wanted to falsify this claim: "llm's cant reason" how would one do that? can you come up with some examples that shows that it can't reason? what if we come up with a new board game with some rules and see if it can beat a human at it. just feed the rules of the game to it and nothing else.
there are close to zero domain specific models that beat frontier SOTA models even in their own domain.
(other than few edge cases like token extraction)
why do you think that's the case? lets start from here.
the real answer is that you get benefits from having data from many sources that add up expontentially for intelligence.
> LLMs are pattern matchers
but lets try to falsify this. can you come up with a prompt that clearly shows that LLM's can't reason?
if we have steps for understanding any author's english and creative process (generally not specific to an author) would you agree then it is possible for an llm to do it?
The real sticking point for me is I don't even believe that authors themselves FULLY understand their process. The idea that anybody could achieve such full introspection as to understand and articulate every little thing that influences their output seems astoundingly improbable.
Repeating a process, yes for sure, even (pseudorandom?) variations on a process. Understanding a process is a different question, and I’m not sure how you would measure that.
In school we would have a test with various questions to show you understand the concept of addition, for example. But while my calculator can perfectly add any numbers up to its memory limit, it has no understanding of addition.
> while my calculator can perfectly add any numbers up to its memory limit, it has no understanding of addition.
"my calculator can perfectly add any numbers up to its memory limit" This kind of anthropomorphic language is misleading in these conversations. Your calculator isn't an agent so it should not be expected to be capable of any cognition.
It’s the degree of generalisability. And LLMs do have understanding. You can ask it how it came up with the process in natural language and it can help - something a calculator can’t do.
They absolutely do not. If you "ask it how it came up with the process in natural language" with some input, it will produce an output that follows, because of the statistics encoded in the model. That output may or may not be helpful, but it is likely to be stylistically plausible. An LLM does not think or understand; it is merely a statistical model (that's what the M stands for!)
how would you empirically disprove that it doesn't have understanding?
i can prove that it does have understanding because it behaves exactly like a human with understanding does. if i ask it to solve an integral and ask it questions about it - it replies exactly as if it has understood.
give me a specific example so that we can stress test this argument.
for example: what if we come up with a new board game with a completely new set of rules and see if it can reason about it and beat humans (or come close)?
> how would you empirically disprove that it doesn't have understanding?
The complete failure of Claude to play Pokemon, something a small child can do with zero prior instruction. The "how many r's are in strawberry" question. The "should I drive or walk to the car wash" question. The fact that right now, today all models are very frequently turning out code that uses APIs that don't exist, syntax that doesn't exist, or basic logic failures.
The cold hard reality is that LLMs have been constantly showing us they don't understand a thing since... forever. Anyone who thinks they do have understanding hasn't been paying attention.
> i can prove that it does have understanding because it behaves exactly like a human with understanding does.
First, no it doesn't. See my previous examples that wouldn't have posed a challenge for any human with a pulse (or a pulse and basic programming knowledge, in the case of the programming examples). But even if it were true, it would prove nothing. There's a reason that in math class, teachers make kids show their work. It's actually fairly common to generate a correct result by incorrect means.
> The complete failure of Claude to play Pokemon, something a small child can do with zero prior instruction
cherry picking because gemini and gpt have beat it. claude doesn't have a good vision set up
> The "how many r's are in strawberry" question
it could do this since 2024
> The "should I drive or walk to the car wash" question
the SOTA models get it right with reasoning
> fact that right now, today all models are very frequently turning out code that uses APIs that don't exist, syntax that doesn't exist, or basic logic failures.
not when you use a harness. even humans can't write code that works in first attempt.
Now, some of the best chess engines in the world are Neural Networks, but general purpose LLMs are consistently bad at chess.
As far as "LLM's don't have understanding", that is axiomatically true by the nature of how they're implemented. A bunch of matrix multiplies resulting in a high-dimensional array of tokens does not think; this has been written about extensively. They are really good for generating language that looks plausible; some of that plausable-looking language happens to be true.
If you look at the "workflow" section of that page, they had to add a bunch of scaffolding around telling the model what moves are legal -- an llm can't keep enough context to know how to play chess; only to choose an advantageous move from a given list. But feel free to "cherry pick".
i ran the benchmark without the valid moves tool as well as the three mistakes grace and gpt-5.4 holds well. it can achieve 1000 ELO which is much higher than my own.
this clearly tells me that GPT is good at chess, at least better than a normal person who has played ~30-40 games in their life.
“i can ask it to give a text description of a linear logical math process that has been described in text countless times”
If you think “the tacit knowledge and conscious/subconscious reasoning mix that caused X to write like X” can be meaningfully captured by some 1-page “style guide” like llmtropes, I’m not sure what to tell you. Such a style description would be informed by a soup of reviewers that most certainly cannot write like X even with their stronger and more nuanced observations than what the LLM picked up.
i can ask it to tell me how to write like a person X right now.