And because reading and writing are thinking we must not delegate it to AI models as a matter of habit. In particular, during students' formative time, they need
to learn how to think in reading and writing mode - reflecting, note-taking etc.
Compare it with the use of a pocket calculator: once you
have a solid grounding, it's fine to use electronic calculators, but first one ought to learn how to calculate
mentally and using pen and paper. If for no other reason, to check whether we made a typo when entering our calculation, e.g. when the result is off by 100 because we did not press the decimal point firmly enough.
I am very concerned that young people delegate to LLMs before reaching that stage.
> To [Thamus] came Thoth and showed his inventions, desiring that the other Egyptians might be allowed to have the benefit of them; he enumerated them, and Thamus enquired about their several uses, and praised some of them and censured others, as he approved or disapproved of them. It would take a long time to repeat all that Thamus said to Thoth in praise or blame of the various arts. But when they came to letters, this, said Thoth, will make the Egyptians wiser and give them better memories; it is a specific both for the memory and for the wit. Thamus replied: O most ingenious Thoth, the parent or inventor of an art is not always the best judge of the utility or inutility of his own inventions to the users of them. And in this instance, you who are the father of letters, from a paternal love of your own children have been led to attribute to them a quality which they cannot have; for this discovery of yours will create forgetfulness in the learners' souls, because they will not use their memories; they will trust to the external written characters and not remember of themselves.
-- Plato, Phaedrus
We've been having this same conversation for over 2,000 years now. And while I actually think Thamus is probably correct, it doesn't change the reality that we are now using reading and writing for everything.
IMO it's not the recording of ideas that is thinking, but rather the act of putting thoughts into language. To me there isn't a big cognitive difference between conversing about a topic (during which you put thoughts into words) and writing about it.
When you speak or write instead of just think, you create something that did not previously exist: new words and sentences. When you write instead of speak, you aren't exactly creating something new — you're often just recording words that just as well could have been spoken. Using an LLM is much closer to the first case. It's creating something that didn't previously exist (an expanded thesis on a brief thought provided by you), and therefore seems to possibly risk the user's ability to think atrophying.
I think this exposes a pattern, but not necessarily on the subject or antithetical to OP's point. I interpret the above passage to implicate that we lose abilities as we adopt tools that can do it for us, but writing specifically stunts our ability to memorize facts. I would argue that this enabled us to spend less mental energy on memorization but on processing information instead, able to do more complex calculations. This doesn't negate OP's point that by using LLM's we give up another kind of ability to a tool, in the case reasoning.
Now whether or not this will in the abstract become leverage for another type of skill or multiplier is to be seen.
There is something much deeper going on when you force yourself to actually write things down. This is especially relevant in engineering.
That is why "RFCs" are so prevalent in many tech companies. They are often just as useful to the writer as they are to the reviewers.
Reading is thinking someone else's thoughts => That is true if you are strictly reading passively. Typically what happens is that reading opens many doors that leads to your own thinking. Of course depends on the type of material you are reading as well. But often reading broadens your thinking relative to just putting your own on paper.
Definitely a good point. I live in a college town and know many people that read all the time, but don't actually do anything active with what they've read. They just consume it continuously and think the understand many topics. Except when you talk to them, it comes out quickly that they didn't actually understand what they read on a deep level, they just went along for the "thinking ride".
And, as you point out, if you push yourself to read actively, it helps a lot!
I really think the effects of LLMs on thinking is the exact same as a calculator. It shortcuts some forms of thinking to open up other forms of thinking.
My thinking has increased with the use of LLMs, not decreased, most likely because LLMs take the edge off of grind work like reading a lot of noise to capture the 1% signal, formulating accurate statements for abstract ideas, and bringing together various domains that are beyond your area of expertise.
Now will you make mistakes? Sure, but you would have made the same mistakes at a slower pace without LLMs anyways. Or more accurately, you just wouldn’t do the research or apply domains not in your area of expertise, and your thinking would be a lot more narrow.
The strawman is thinking that banning LLMs will induce rigorous thinking. Just like banning calculators does not make everyone good at math.
But allowing calculators WILL make those who like math reach much deeper into the field than without.
> Back of the envelope guesstimating is bread and butter.
And that is not done with calculators, that is done quickly in your head by having practiced a lot of calculations manually. This is why engineer students still practice manual calculation in college in most places.
Google search has worsened so badly. That right know it's impossible to resist using one of those free for a taste LLM service.
And the feeling is similar to how using Google on the 2004-2014 web was.
It used to be Google would return a huge list or relevant links. Loading all of them was quick. Skimming the content was quick.
Now every search is a massive ad. Every site is slow to load full of ads and useless slop. Slop which was written manually at first, then accelerated with Markov chains, now at light speed with LLMs.
So an LLM is required to filter through the LLM slop to find the tiny bit of real content.
It's possible to not use Google for search. I switched to Perplexity many months ago. Almost never come back to Google. No one is forcing me to use a worse option, when a better one exists. Just use something else - easy.
it's like those kids will live in the future, where there's advanced AI
I think we should trust children enought that they'll also figure out a crazy changing technological world.
on the other hand, internet millenial ideals are fast dying. the digital dream of cultural and mediatic abundance is turning into a nightmare of redundant content as information wars saturate the figurative airwaves
Of course I trust my kids to make the most of the environment they are given, and given that their environment will differ from the one I am adapted for, they will likely surpass me in being well adapted to the environment of the future; it's still my responsibility to prepare them as best as I can for it.
You might put a baby in a pool so it can learn to swim, but you make sure their environment is such that drowning is an impossibility. A child destined to be an Olympian swimmer still requires guidance, even if their natural ability and inclinations outpace both their peers and their elders.
Yeah, I think there are environmental things for which our culture cannot prepare us, yet I also think many of our inherited behaviors and beliefs will help us because the environment may not change that much.
Nicolas Carr addresses this issue directly in his book "The Shallows" in which he brilliantly recounted how media has reshaped how humans think and communicate, especially how the word streams of other people increasingly reshaped our collective focus and our ability to focus, which alas, has NOT freed us to think more deeply.
Humans always have and always will use tech as a crutch -- to reduce time and effort (and energy expended). The 'physical enshittification' (PE) that has ensued from using mechanical crutches has made us lazy, fat, and sick. And now _mental_ crutches have arrived, which promise to replace our very thought processes, freeing us from all the annoying cognitive heavy lifting once done by our brains.
IMO, there's every reason to believe that the next step in human evolution will be driven by the continued misuse of tech as crutches, likely leading to widespread _mental_ enshittification (ME) -- doing to our minds what misuse of tech has already done to our culture and to our bodies.
Perhaps mankind can avoid this fate. But only if we insist on _thinking_ for ourselves.
I've heard that some philosophers like Schopenhauer argue that reading can become a passive process, where we simply follow another person's thoughts without engaging our own critical thinking. It's interesting to consider that it's not just LLMs but we too would become like stochastic parrots under certain circumstances.
To quote Paul Graham: "Writing is thinking. To write well you have to think clearly, and thinking clearly is hard. In fact there's a kind of thinking that can only be done by writing. You can't make this point better than Leslie Lamport did: If you're thinking without writing, you only think you're thinking. So a world divided into writes and write-nots is more dangerous than it sounds. It will be a world of thinks and think-nots."
I am quite puzzled how an LLM could even start "write" a scientific paper.
Say you start with a set of findings, for example, western blots, data from a transgenic mouse engineered for the relevant gene, and some single cell sequencing data. Your manuscript describes the identification of a novel protein, editing the gene in a mouse and showing what pathways are affected in the mouse.
What material would you give the LLM? How would the LLM "know" which of these novel findings were in any way meaningful? As far as I'm aware, it is unlikely that the LLM would be able to do anything other that paraphrase what you instruct it to write. It would be a return to the days before word processing became common, and researchers would either dictate their manuscripts to a typist, or hand the typist a stack of hand-written paper.
The actually hard part of writing scientific papers is not putting the words "down on paper" so to speak, but deciding what to say.
When we go to grad school, we’re taught how to write a research paper. Each field has a more or less standard format, where different types of data go in specific sections. So if an LLM is trained on enough papers in that field, it can learn to plug in the information you provide according to those conventions.
In that sense, you’d give the LLM the purpose of the paper, the field you’re writing in, and the relevant data from your lab notebook. Personally, I never enjoyed writing manuscripts — most of the time goes into citing every claim and formatting everything correctly, which often feels more like clerical work than communicating discovery.
I don’t mind if LLMs help write these papers. I don’t think learning to mimic this stylistic form necessarily adds to the process of discovery. Scientists should absolutely be rigorous and clear, but I’d welcome offloading the unnecessary tedium of stylized writing to automation.
I am experienced in writing scientific papers, so I know what it takes.
I remain to be convinced that the tasks you propose an LLM could do contribute any more to the process of writing a paper than dictating to a typist could do in the 1950's. It's impressive for a machine, but not particularly productivity-boosting. Tedious tasks such as correctly formatting references belong to the copy-editing stage (i.e. very last stage of writing a paper), where indeed I have seen journals adopt "AI" approaches. But these processes are not a bottleneck in the scientist's workflow.
I certainly don't think the performance of LLMs that I'm familiar with would be any use at all in compiling the original data into scientifically accurate figures and text, and providing meaningful interpretations. Most likely they would simply throw out random "hallucinations" in grammatically correct prose.
But correctly formatting references is pretty much a solved task through reference managers, possibly plus bibtex. It's a well-defined task, after all, and well suited to traditional software techniques. [1] If someone used an LLM to format the references, you would still have to go back through them.
If there is any use for LLMs in paper writing, I would think that it is for tedious but not well-defined tasks. For example, asking if an already written paper conforms to a journal's guidelines and style. I don't know about you, but I spend a meaningful amount of time [2] getting my papers into journal page limits. That involves rephrasing to trim overhangs, etc. "Rephrase the following paragraph to reduce the number of words by at least 2" is the kind of thing that LLMs really do seem to be able to do reliably.
1: As usual, the input data can be wrong, but that would be a problem for LLMs too.
2: I don't actually know how much time. It probably isn't all that long, but it's tedious and sure does feel like a long time while I'm doing it.
I must be a bad researcher then because every paper I've written starts as a very vague "here are the overarching implications and important results". But the detailed order of results and the nuts and bolts of how to argue out the conclusions gets decided in drafting. Only the simplest of results I've had is essentially pre-written.
Even speaking is thinking. This is why free speech is the very First Amendment. Whoever denies your right to speak freely is controlling your thinking.
Writing is thinking with a superpower. It's like using the "Pensive" from Harry Potter, depicted in the scene where Harry and Dumbledore pull memory whisps out of their temple to rewatch in a mirror pool. Writing enables you to apply your attention to an idea at multiple levels of analysis with significantly less effort than doing the same while also preserving the idea in your head manually.
> Even speaking is thinking. This is why free speech is the very First Amendment. Whoever denies your right to speak freely is controlling your thinking.
There's an excellent podcast (Radiolab, possibly) about how this conception of what the first amendment means is rather recent (1910s-1920s) and that the ideas of what "free speech" meant before that are really radically different.
I see a lot of people say "writing is so important," and I think what they mean is "I feel really smart/good when I write." And I think what they are experiencing is that they've been assembling ideas in their heads for weeks, and only when it's all come together are they ready synthesize that information at a higher level, and they mistake this synthesis for the writing itself (rather than the writing being a symptom OF the synthesis -- if they had tried to write a week prior they would have found it unproductive).
Let's say, I am making something concrete by putting ideas, thoughts, knowledge into paper. While doing it, I am finding gaps and mistakes and finding opportunities to correct them. But it is not limited to 'correction', it also opens newer dimensions and perspectives- ones that previously didn't exist in my conscious mind.
I consider writing as a tool of thinking. Another tool is brainstorming with a group, or any group discussion in general. These amend to your thoughts, make the existing ones more solid, and opens new direction, and unravels connections previously not accessible.
Read this essay by Paul Graham: Putting Ideas into Words [0]. And also refer to his other essays on writing.
There is also a great book by Paul Zissner: Writing to Learn. I suggest this book to people.
Writing, when done while learning works akin to teaching- one of the most crucial steps in so-called Feynman Technique of learning.
Yeah I think I've read multiple PG essays on the importance of writing, but they always struck me as no different than Katy Perry telling you to sing, or some Grandmaster saying "go play chess." That is -- a personal anecdote that doesn't necessarily generalize.
I'm not saying that writing can't be a useful tool to organize ideas, definitely it can. But I think I've found two things:
- Now the best way to "iterate" my thoughts is to rubberduck with ChatGPT; it's really amazing how much faster I can learn when I admit how little I know, even on something like global warming or an advanced math topic.
- By and large, "organizing my thoughts" isn't really a high-return activity in my life. Having an intelligently written blog that I've put hundreds of hours into has never done anything for my career or led to any personal connections, and honestly who's to say my time wouldn't have been better-spent just coming with some jokes to network better rather than having some cohesive theory of everything that nobody asked for?
Just one personal anecdote: I definitely find contradictions or gaps in my thinking/knowledge when I write. Finding and resolving those deficiencies is what I point to when I say "writing is thinking".
Getting stuff written down makes me realise the weak points and the errors in my thinking which I know from experience I don't find if I don't write it down.
This is a really cynical take. People work differently and get value from different things. It’s probably safe to assume most aren’t virtue signalling about writing.
Sometimes I like to test whether I can actually construct or assemble a finished something in my mind from an inkling of a thought.
For example a few days ago I realized that I found it hard to reverse a word in my mind, even a simple one. Try for yourself, think of a word and then reverse it in your head with your eyes closed.
Some people might struggle with the above, some may find it doable in their heads, but most can agree that it's absurdly easy if you can externalize it to paper or a text editor at least.
I think you're right. I'll add on: there's a lot of thinking that does not need writing, and there's a lot of writing that needs no thinking. Deng Xiaoping and other greats wrote pretty minimally for their own thinking, if at all. Whereas many of us not-so-greats seem to knee-jerk comment without a single thought.
It makes sense for our age. Amid a thousand distractions, typing on the keyboard gives the illusion of getting a grip. Note-taking on my computer gives the illusion of a second brain. Ululating on the internet gives the illusion of sharing thoughts.
Instead of "writing is thinking", I prefer "thought precedes speech" https://inframethodology.cbs.dk/?p=1127; it fits the small human mind better though I've yet to learn it properly.
You need to add into consideration that laying things out in visual display provides cognitive support, reducing the effort to reason about more things. So writing out ideas really does allow you to reach a greater scope of synthesis.
If they tried writing a week prior that would have realized sooner the gaps in the ideas they were assembling, leading to them closing those gaps faster.
I disagree. Not sure how common this workflow is, but I write by putting all the different unsynthesized ideas down and rearrange them as the latent structure "reveals itself." At the end you have something synthesized.
Sure some type latent structure was there all along (thus why I put them down), but it wasn't necessarily visible to me, nor optimal, nor did it include/exclude all the right points. The need for iteration itself proves that the act of writing is actually doing the synthesis.
While I agree with the underlying message, "writing is thinking" is only circumstantially correct. It wasn’t always like this.
We learned to think by writing only after writing became cheap. Yes, we’ve trained our brains to develop ideas by editing raw thoughts on paper, but it is just one of the possible methods.
I have read a lot of late 18th, 19th and early 20th century books and diaries, and it is plainly clear that writers such as Tolstói, Zweig, Goethe and others developed full books in their mind first, then wrote them from cover to cover in 20-30 days.
Thinking used to be detached from writing. That is a fact. We just lost that ability in the modern era thanks to cheap writing technology: pen and paper, then computers. I'm not saying the current approach is wrong, but don't assume that the only way to think is to write.
Socrates argued that writing would destroy people's memory. He wasn't 100% wrong, yet here we are. The criticism towards the use of LLMs is so deliciously ironic. The analogy with writing... writes itself. Kids that grow up with LLMs will just think differently.
You seem to be responding to the reverse, “thinking is writing”, which I agree is not true, you can think without writing.
They are making the point that writing is more than dumping a completed thought. The act of doing that helps you to critique your dumped thoughts, to have more thoughts about your thoughts, to simplify them or expand them.
Before paper became cheap, wax or wooden tablets were used for ephemeral writing.
> I have read a lot of late 18th, 19th and early 20th century books and diaries, and it is plainly clear that writers such as Tolstói, Zweig, Goethe and others developed full books in their mind first, then wrote them from cover to cover in 20-30 days.
I seriously doubt that it was ever common for writers to compose a whole book in their head and then write it down. Maybe some writers with exceptional memories did this. But there's a whole book about how War and Peace was written based on textual evidence that wouldn't exist if it had simply popped out of Tolstoy's head fully formed: https://www.amazon.com/Tolstoy-Genesis-Peace-Kathryn-Feuer/d....
Not war and peace, which was episodic, but smaller novels were thought out in Tolstoy’s mind before being written wholly. He mentions this in his diaries. Zweig mentions the same, too, but of course his novels are generally much shorter than the two Tolstoy’s masterpieces.
AFAIK the dominant theory is that they weren't memorized as a whole text, but composed on the fly with the help of a memorized set of stock formulas. [0]
> writers...developed full books in their mind first
When reading long, closely reasoned passages of medieval philosophy, I've wondered about their development process, when there was no such thing as scratch paper.
> Kids that grow up with LLMs will just think differently.
People are just glibly saying this sort of thing, but what specifically is coming? I'm now wrestling with the problem of dealing with university students who don't hesitate to lean on LLMs. I'm trying to not be dismissive, but it feels like they are just thinking less, not differently.
As a similar but distinct theory, you might find Larry McEnerney‘s work interesting. Writing has two classes: a writing for thought and a writing for communication. Larry uses horizontal and vertical spatial metaphors here. Writing for thought still pre-dates cheap paper (and Socrates), but is mostly a private act. Writing for communication is a broad enough brush to span fiction and journalism. For his part, Larry teaches classes aimed at thesis writers who struggle to bridge the divide of using writing to think about a problem to conveying their answer in a paper.
Yeah, LLMs are entirely different from "writing" because they're creative agents. So, writing allows me to give my thoughts several passes, to edit over time. It's like I can have several of me to think, write and edit, spaced over time.
LLMs are like I have someone else to do some or all of the thinking and writing and editing. So I do less thinking.
A bicycle lets my own energy go father. Writing.
A car lets me use an entirely different energy source. LLMs.
Which one is better for my physical fitness?
Btw the idea about Tolstoy and others keeping those massive books in their head and cranking them out over a month is fascinating. Any evidence or others who imagine the same?
In Tolstoy's case, he was a count and surely had the funds, no?
I’ve read Tolstoy’s diaries and he mentions the thought process he uses to write small novels. First he thinks about what should happen, then he writes (or dictates) the text. Thinking takes a few weeks, sometimes a month, then writing is pretty quick. There is some editing, but nothing like we do nowadays.
Bigger novels such as war and peace were written episodically.
> Thinking used to be detached from writing. That is a fact. We just lost that ability in the modern era thanks to cheap writing technology: pen and paper, then computers. I'm not saying the current approach is wrong, but don't assume that the only way to think is to write.
I have a better way to frame this:
Learning your own language and culture is a lifelong process.
A big phase, the adult phase, of learning is learning to write in your language (I'm implying there's more to writing than chosing words; specially in this context of language as thinking)
indeed, a lot of modern people never make it out of this big phase of learning your language. they never go beyond writing = thinking. but some people do learn the next phase
which involves distinguishing language itself from thoughts and ideas (is some idea known? understood? perceived?? but the idea is "the self" or some other complex notion)
so the only quality of the modern era I admit, is that there's a lot of people that only learn rudimentary thinking-writting, and too few people that learn 'advanced' languange-thinking where writing becomes secondary to thinking.
finally, I learned this idea from reading around the meaningness blog/book
> We learned to think by writing only after writing became cheap. Yes, we’ve trained our brains to develop ideas by editing raw thoughts on paper, but it is just one of the possible methods.
I think you have some misconceptions here. First, the article does not claim that thinking is writing, and especially not that there is no thinking without writing. They only explain that writing is supporting and driving a higher quality of thinking.
Second, paper isn't the only medium to write. And writing isn't the only persistent form of communication to support and improve thinking.
I'm not sure if writers developed the entire book in their head first, but: it was indeed very, very common for people to dictate novels, journal entries, and other "written" works to a secretary, typist, or tape recorder.
Nowadays that seems to be rare, but my impression from reading my journals is that it was often more common to dictate than to physically hand write things.
Novels were dictated, that is absolutely correct, but on top of it, the whole plot was developed with a high level of detail before dictation started. There was some editing, of course, but nothing like we do today, where writing books is basically an iterative process. We lean on the written word too much for our thinking (not being critical, just that’s how we are taught)
That's true, but I would phrase it from a different perspective:
It's seems clear that abstract thinking in particular is greatly aided by writing, because the written text acts like a thought cache. A bit like an LLM context window which you can fill with lots of compact, compressed "tokens" (words).
Abstract thoughts are "abstract" because they can't be visualized in our mind, so they don't benefit from our intuitive imagination ability (Kant's "Anschauung"). So it is hard to juggle many abstract thoughts in our working memory.
We can also think of the working memory as the CPU registers, which are limited to a very small number, while the content of the CPU cache or RAM corresponds to the stuff we write down.
Our "anschauung" (visual imagination) is perhaps something like a fixed function hardware on a GPU, which is very good at processing complex audiovisual content, i.e. concrete thoughts, but useless for anything else (abstract thoughts).
Many people here are taking away from this that substituting our own writing with the writing from an LLM is a danger for human development.
I see things more optimistically. If good writing leads to good thinking, then anything I do to improve my ability to write well transitively helps me to think well.
In that sense, I actually see a huge benefit to LLMs in improving my writing and therefore improving my thinking. Not only can I ask for detailed and powerful feedback, I can also ask for more details on background context or related topics that I wouldn't be aware of.
I believe judicious use of LLMs can make us better than we could be without them.
I subscribe to this view. LLMs can be tools for thinking, not substitutes to thinking. In fact, I feel that with time, the way we think may fundamentally change, just like how reading and writing changed thinking. An LLM can be an ideal foil to test out ideas, and the process of thinking could be an iterative process with the LLM as an active participant.
Whenever I need to do some hard thinking and things are not clear, I fire up my sublime text and write down the context in the simplest terms and short lines (only few words per line). While doing this, I will be absolutely rude to myself asking extremely basic and direct questions to bring out the real context, real goals and real path. It's like answering an under-world boss. No bullshit, no pretense, no regard to any norms, no impressing someone. Then the whole thing falls into a meaningful structure. I leave it when it produces some immediate action items.
If writing is thinking, then perhaps we should think more about what language would produce the best thinking, and does it exist? How do we go about creating it?
And if thinking is dependent on language, maybe we should create a new language for artificial intelligence rather than feeding it human languages.
There's been a lot of work in this area but no definitive answer, and few constructed languages speakers [1]. I imagine the wins may be marginal, similar to the wins in dvorak vs qwerty typing.
My experience with writing is that it's often a matter of simply noting down what comes pouring from my subconscious, with the most taxing task being keeping ahold of some given connection, sentiment, or wording (as that subconscious seems to generate more than my working memory can keep track of).
Sometimes I struggle to fit those sentiments and connections to wording that I imagine will make sense, to someone else or even to myself. I guess that would be the, "Writing is thinking," part, but it seems more like, "Effective and coscientious (self-)communication is thinking."
This intuitively makes sense (like deep-frying a JPEG), but it doesn't seem to happen in practice, as modern models are frequently trained on text both output from other models, and curated from other models.
Realistically, going forward model training will just need to incorporate a step to remove data below some quality threshold, LLM-generated or otherwise.
New HN metric: how many comments do you have to scroll down on a LLM-edited/written submission before someone points it out? Triple bonus if the submission is about LLM writing and the need to think for yourself without a LLM, and still no one has pointed this out. Current page score: 333 points.
Because it has many of the typical 4o stylistic tics like 'it's not X, it's Y' or enumeration or the em dashes, or the twist ending.
It's not 100% unedited ChatGPT and far from the most blatant instance that has caught my eye (they've started showing up in the New York Times and New Yorker as well, have you noticed that?), but certainly sounds like that was used: "Writing compels us to think — not in the chaotic, non-linear way our minds typically wander, but in a structured, intentional manner." "This is not merely a philosophical observation; it is backed by scientific evidence." "Importantly, if writing is thinking, are we not then reading the ‘thoughts’ of the LLM rather than those of the researchers behind the paper?" "overcoming writer’s block, provide alternative explanations for findings or identify connections between seemingly unrelated subjects."
(Note that this is particularly ironic because as the op-ed notes, if they did use it, they are required by Nature to disclose this... https://www.nature.com/articles/d41586-023-00191-1 But of course, how would anyone ever prove they did so? You know how difficult it is to get Nature to retract even blatant fraud.)
Writing has always been how I organize my thoughts. A lot of ideas only become clear as I work through them on the page. AI can save time, but it also makes it easy to skip the slower, more reflective parts of thinking. For me, that’s often where the real value of writing comes from.
We can't really talk about this topic without mentioning the book "How to Take Smart Notes" by Soenke Ahrens
One key takeaway is if you want to learn/remember something better, always rewrite things in your own words as both the act of writing AND paraphrasing makes it more sticky
Part of me can’t help but think that scientific journals as billion dollar industries need to keep the status quo of how articles are written, where they get submitted and who reviews them. Even though per review today is failing
Writing what someone else wrote is thinking what someone else thought. My favorite learning technique is reading, listening or viewing something and then typing it into libreoffice. Specially useful when it is something that is transcribed. Works really well for code, too.
Give it a try!
While the article’s point seems intuitively true, it cites only two papers related to the benefits of handwriting. But it’s argument is stronger than that. Is there peer-reviewed evidence to support the stronger claim of the benefits of typewriting?
I love this! Especially the part about if writing is thinking, are we not then reading the ‘thoughts’ of the LLM rather than those of the researchers behind the paper?
I would also compare reading to being reprogrammed like EEPROM. Although the process is slower, the changes feel more permanent when learning: you need to create connections yourself from examples compared to someone demonstrating it on the video.
It seems obvious to me. Just like drawing and sketching helps thinking about design, coding helps thinking about programming.
It's one of the reasons for the "one to throw away"-idea of writing shitty code first just to get it to work, and then remake it after you have thought through the problem by coding it.
This morning I asked ChatGPT a question about how Quickbooks handles charts of accounts compared to NetSuite. It answered my question better than anything else would have.
Also, I'm currently using Claude Code to fix some bugs -- it's handling the heavy lifting while I think about what needs to happen.
I'm in favor of human writing as an underrated tool of culture-making...but the scope of what counts as "thinking" is expanding.
Writing needs a conceptual split analogous to the split between math and calculating.
Just as a calculating can be implemented on a computer which has low cognitive abilities but high algorithmic and procedural abilities, we need to extract out the word-smithing capabilities from writing separate from the thinking portion. Our lack of distinction in terms reflects a muddled conceptual framework.
LLMs are excellent wordsmiths completely divorced from the concept of thinking. They break the correlative assumption - that excellent writing is corresponds with excellent thinking. Until now, we've been able to discern poor idea because they have a certain aesthetic, think conspiracy rants in docx saying something about a theory of everything based on vibrations. But that no longer holds. We have decent enough word-smithing coupled with a deficit of thinking. Unfortunately this breaks our heuristics with consequences ranging from polluting our online commons to folks end up believing nonsense like ChatGPT named itself Nova and they are a torchbearer for spiritual gobbledygook.
My point is that we're in the process of untangling these two and as a result, we're likely to see confusion and maybe even persistent misunderstanding until this distinction becomes a more common part of how we talk about and evaluate written work. They're living in an AGI-world and we're just..not.
We think by association. We can sometimes tighten up the process when there's a formal logical framework that applies, but it's not as natural or automatic.
What writing changes is that in words, you have to make it explicit how one thing leads to another. Partly, that's just due to the imposition of sentence structure.
Ironically, this is precisely the crazy thing about Trumpspeech: it's just associations - vibe-chaining if you will.
To write something you must form a "theory of the system", which is orders of magnitude more difficult than feeding on fodder. It's the difference between being fluent in a language and sort of understanding something.
It's quite similar in hard sciences as it's in natural languages. For instance I don't understand Hungarian at all. Few words "igen", "jó napot kívánok" doesn't a knowledge of the language form.
Then German. I had to learn it in school so I have orders of magnitude better grasp at it because I can actually say a few statements that form in my mind: "Nein, ich brauch nicht ein anderes stück Steak". Might not be 100% correct gramatically and vocabulary wise but it conveys the message and also transmits that I understand the context.
And then come English which I speak since 33 years. I actually THINK in English a lot of times and there are concepts I can't easily express in my native Romanian language without resorting to a painfully long and sometimes unsuccessful software-driven (as opposed to FPGA-encoded) translation process.
> For example, LLMs can aid in improving readability and grammar, which might be particularly useful to those for which English is not their first language.
I don't know whether this has been empirically confirmed, but I have the strong belief that a manuscript with poor grammar, by a non-native English speaker, has a much higher probability of being rejected than the same manuscript but copyedited by something like Grammerly or a SOTA LLM.
Ideally writing style should matter much less than the quality of the research, but reviewers are not just influenced by objective criteria but, unconsciously, also by vibes, which includes things like writing style and even formatting.
Broken English still has its charm and brings the structure of the writer's native language to the fore, which makes it relatively easier to parse and glean intentions from than polished LLM-speak.
That might be true, but I think it's false. Or more precisely, I think manuscripts with broken English have statistically a higher probability of being rejected than ones that are copyedited with AI.
the act of writing takes raw experience, puts it in front of the eyes, and then filters it back through your critical faculties so you can refine and reason about it. the iteration makes it higher quality thought.
however when I encounter people with low written or verbal acuity, they have to survive somehow, so it's wise to observe what tools of cunning they tend to reach for.
To add: Reading is also thinking (ideally).
And because reading and writing are thinking we must not delegate it to AI models as a matter of habit. In particular, during students' formative time, they need to learn how to think in reading and writing mode - reflecting, note-taking etc.
Compare it with the use of a pocket calculator: once you have a solid grounding, it's fine to use electronic calculators, but first one ought to learn how to calculate mentally and using pen and paper. If for no other reason, to check whether we made a typo when entering our calculation, e.g. when the result is off by 100 because we did not press the decimal point firmly enough.
I am very concerned that young people delegate to LLMs before reaching that stage.
Compare it with the invention of writing:
> To [Thamus] came Thoth and showed his inventions, desiring that the other Egyptians might be allowed to have the benefit of them; he enumerated them, and Thamus enquired about their several uses, and praised some of them and censured others, as he approved or disapproved of them. It would take a long time to repeat all that Thamus said to Thoth in praise or blame of the various arts. But when they came to letters, this, said Thoth, will make the Egyptians wiser and give them better memories; it is a specific both for the memory and for the wit. Thamus replied: O most ingenious Thoth, the parent or inventor of an art is not always the best judge of the utility or inutility of his own inventions to the users of them. And in this instance, you who are the father of letters, from a paternal love of your own children have been led to attribute to them a quality which they cannot have; for this discovery of yours will create forgetfulness in the learners' souls, because they will not use their memories; they will trust to the external written characters and not remember of themselves.
-- Plato, Phaedrus
We've been having this same conversation for over 2,000 years now. And while I actually think Thamus is probably correct, it doesn't change the reality that we are now using reading and writing for everything.
IMO it's not the recording of ideas that is thinking, but rather the act of putting thoughts into language. To me there isn't a big cognitive difference between conversing about a topic (during which you put thoughts into words) and writing about it.
When you speak or write instead of just think, you create something that did not previously exist: new words and sentences. When you write instead of speak, you aren't exactly creating something new — you're often just recording words that just as well could have been spoken. Using an LLM is much closer to the first case. It's creating something that didn't previously exist (an expanded thesis on a brief thought provided by you), and therefore seems to possibly risk the user's ability to think atrophying.
> IMO it's not the recording of ideas that is thinking, but rather the act of putting thoughts into language.
I agree with you but that article itself says, "for example, handwriting can lead to widespread brain connectivity."
It doesn’t say anywhere that conversing doesn’t.
I think this exposes a pattern, but not necessarily on the subject or antithetical to OP's point. I interpret the above passage to implicate that we lose abilities as we adopt tools that can do it for us, but writing specifically stunts our ability to memorize facts. I would argue that this enabled us to spend less mental energy on memorization but on processing information instead, able to do more complex calculations. This doesn't negate OP's point that by using LLM's we give up another kind of ability to a tool, in the case reasoning.
Now whether or not this will in the abstract become leverage for another type of skill or multiplier is to be seen.
Or, from the perspective of memetics, writing has always been using us for everything.
Skullface sends his regards. I kneel Hideo Kojima.
Sounds profoundly anti-humanist.
What you mean by "humanist" does not seem to be what philosophers mean by it.
Sounds like "memetics" gives agency to writing as a thing independent of human beings doing writing, which is an interesting frame but also untrue.
Ideas don't die. https://scp-wiki.wikidot.com/introductory-antimemetics
Yeah, well, that's just how the humanism meme wants to instrumentalize you at this moment.
I recommend Havelock's Preface to Plato and Walter Ong's Orality and Literacy
There is something much deeper going on when you force yourself to actually write things down. This is especially relevant in engineering. That is why "RFCs" are so prevalent in many tech companies. They are often just as useful to the writer as they are to the reviewers.
Reading is thinking someone else's thoughts.
Writing is thinking your own thoughts.
There's a big difference, and is why writing is so painful for so many people. It's also why writing is critically important.
edit: Likewise teaching is really important. Crystallization of thought is incredible valuable and difficult.
Reading is thinking someone else's thoughts => That is true if you are strictly reading passively. Typically what happens is that reading opens many doors that leads to your own thinking. Of course depends on the type of material you are reading as well. But often reading broadens your thinking relative to just putting your own on paper.
Definitely a good point. I live in a college town and know many people that read all the time, but don't actually do anything active with what they've read. They just consume it continuously and think the understand many topics. Except when you talk to them, it comes out quickly that they didn't actually understand what they read on a deep level, they just went along for the "thinking ride".
And, as you point out, if you push yourself to read actively, it helps a lot!
I think the best way to actively read is to write down your own thoughts as commentary.
No , reading is filling up your mind's LLM with the context given by someone else. Your thinking is what happens after reading.
I’d say reading is fine tuning. It changes your weights. That’s the whole point of doing it.
I really think the effects of LLMs on thinking is the exact same as a calculator. It shortcuts some forms of thinking to open up other forms of thinking.
My thinking has increased with the use of LLMs, not decreased, most likely because LLMs take the edge off of grind work like reading a lot of noise to capture the 1% signal, formulating accurate statements for abstract ideas, and bringing together various domains that are beyond your area of expertise.
Now will you make mistakes? Sure, but you would have made the same mistakes at a slower pace without LLMs anyways. Or more accurately, you just wouldn’t do the research or apply domains not in your area of expertise, and your thinking would be a lot more narrow.
The strawman is thinking that banning LLMs will induce rigorous thinking. Just like banning calculators does not make everyone good at math.
But allowing calculators WILL make those who like math reach much deeper into the field than without.
>> But allowing calculators WILL make those who like math reach much deeper into the field than without.
Have you ever run into any mathematician that praised the calculator for his/her career? I’d be really curious to read about that.
Pure math people probably don’t reach for calculators. But engineers do all the time. Back of the envelope guesstimating is bread and butter.
The modern equivalent of a calculator is Excel.
> Back of the envelope guesstimating is bread and butter.
And that is not done with calculators, that is done quickly in your head by having practiced a lot of calculations manually. This is why engineer students still practice manual calculation in college in most places.
Calculators are taken for granted but many mathematicians use computers extesively in their careers.
Google search has worsened so badly. That right know it's impossible to resist using one of those free for a taste LLM service.
And the feeling is similar to how using Google on the 2004-2014 web was.
It used to be Google would return a huge list or relevant links. Loading all of them was quick. Skimming the content was quick.
Now every search is a massive ad. Every site is slow to load full of ads and useless slop. Slop which was written manually at first, then accelerated with Markov chains, now at light speed with LLMs.
So an LLM is required to filter through the LLM slop to find the tiny bit of real content.
It's possible to not use Google for search. I switched to Perplexity many months ago. Almost never come back to Google. No one is forcing me to use a worse option, when a better one exists. Just use something else - easy.
it's like those kids will live in the future, where there's advanced AI
I think we should trust children enought that they'll also figure out a crazy changing technological world.
on the other hand, internet millenial ideals are fast dying. the digital dream of cultural and mediatic abundance is turning into a nightmare of redundant content as information wars saturate the figurative airwaves
Of course I trust my kids to make the most of the environment they are given, and given that their environment will differ from the one I am adapted for, they will likely surpass me in being well adapted to the environment of the future; it's still my responsibility to prepare them as best as I can for it.
You might put a baby in a pool so it can learn to swim, but you make sure their environment is such that drowning is an impossibility. A child destined to be an Olympian swimmer still requires guidance, even if their natural ability and inclinations outpace both their peers and their elders.
Yeah, I think there are environmental things for which our culture cannot prepare us, yet I also think many of our inherited behaviors and beliefs will help us because the environment may not change that much.
Trusting kids to figure out the unfiltered Internet led to a massive mental health crisis.
Nicolas Carr addresses this issue directly in his book "The Shallows" in which he brilliantly recounted how media has reshaped how humans think and communicate, especially how the word streams of other people increasingly reshaped our collective focus and our ability to focus, which alas, has NOT freed us to think more deeply.
Humans always have and always will use tech as a crutch -- to reduce time and effort (and energy expended). The 'physical enshittification' (PE) that has ensued from using mechanical crutches has made us lazy, fat, and sick. And now _mental_ crutches have arrived, which promise to replace our very thought processes, freeing us from all the annoying cognitive heavy lifting once done by our brains.
IMO, there's every reason to believe that the next step in human evolution will be driven by the continued misuse of tech as crutches, likely leading to widespread _mental_ enshittification (ME) -- doing to our minds what misuse of tech has already done to our culture and to our bodies.
Perhaps mankind can avoid this fate. But only if we insist on _thinking_ for ourselves.
> To add: Reading is also thinking (ideally).
I've heard that some philosophers like Schopenhauer argue that reading can become a passive process, where we simply follow another person's thoughts without engaging our own critical thinking. It's interesting to consider that it's not just LLMs but we too would become like stochastic parrots under certain circumstances.
To quote Paul Graham: "Writing is thinking. To write well you have to think clearly, and thinking clearly is hard. In fact there's a kind of thinking that can only be done by writing. You can't make this point better than Leslie Lamport did: If you're thinking without writing, you only think you're thinking. So a world divided into writes and write-nots is more dangerous than it sounds. It will be a world of thinks and think-nots."
https://www.paulgraham.com/writes.html
Thinking and writing are closely linked.
Thinking and using ChatGPT are not. Overview ‹ Your Brain on ChatGPT — MIT Media Lab https://share.google/RYjkIU1y4zdsAUDZt
https://www.media.mit.edu/projects/your-brain-on-chatgpt/ove...
I am quite puzzled how an LLM could even start "write" a scientific paper.
Say you start with a set of findings, for example, western blots, data from a transgenic mouse engineered for the relevant gene, and some single cell sequencing data. Your manuscript describes the identification of a novel protein, editing the gene in a mouse and showing what pathways are affected in the mouse.
What material would you give the LLM? How would the LLM "know" which of these novel findings were in any way meaningful? As far as I'm aware, it is unlikely that the LLM would be able to do anything other that paraphrase what you instruct it to write. It would be a return to the days before word processing became common, and researchers would either dictate their manuscripts to a typist, or hand the typist a stack of hand-written paper.
The actually hard part of writing scientific papers is not putting the words "down on paper" so to speak, but deciding what to say.
When we go to grad school, we’re taught how to write a research paper. Each field has a more or less standard format, where different types of data go in specific sections. So if an LLM is trained on enough papers in that field, it can learn to plug in the information you provide according to those conventions.
In that sense, you’d give the LLM the purpose of the paper, the field you’re writing in, and the relevant data from your lab notebook. Personally, I never enjoyed writing manuscripts — most of the time goes into citing every claim and formatting everything correctly, which often feels more like clerical work than communicating discovery.
I don’t mind if LLMs help write these papers. I don’t think learning to mimic this stylistic form necessarily adds to the process of discovery. Scientists should absolutely be rigorous and clear, but I’d welcome offloading the unnecessary tedium of stylized writing to automation.
I am experienced in writing scientific papers, so I know what it takes.
I remain to be convinced that the tasks you propose an LLM could do contribute any more to the process of writing a paper than dictating to a typist could do in the 1950's. It's impressive for a machine, but not particularly productivity-boosting. Tedious tasks such as correctly formatting references belong to the copy-editing stage (i.e. very last stage of writing a paper), where indeed I have seen journals adopt "AI" approaches. But these processes are not a bottleneck in the scientist's workflow.
I certainly don't think the performance of LLMs that I'm familiar with would be any use at all in compiling the original data into scientifically accurate figures and text, and providing meaningful interpretations. Most likely they would simply throw out random "hallucinations" in grammatically correct prose.
But correctly formatting references is pretty much a solved task through reference managers, possibly plus bibtex. It's a well-defined task, after all, and well suited to traditional software techniques. [1] If someone used an LLM to format the references, you would still have to go back through them.
If there is any use for LLMs in paper writing, I would think that it is for tedious but not well-defined tasks. For example, asking if an already written paper conforms to a journal's guidelines and style. I don't know about you, but I spend a meaningful amount of time [2] getting my papers into journal page limits. That involves rephrasing to trim overhangs, etc. "Rephrase the following paragraph to reduce the number of words by at least 2" is the kind of thing that LLMs really do seem to be able to do reliably.
1: As usual, the input data can be wrong, but that would be a problem for LLMs too. 2: I don't actually know how much time. It probably isn't all that long, but it's tedious and sure does feel like a long time while I'm doing it.
Re-phrasing to fit within word or character limits is certainly something I would pay for!
I have often spent more time doing this than writing the original draft, especially for grant applications...
The LLM can make a plan or outline first, which is also writing.
Any researcher already has this in their head long before any writing takes place.
I must be a bad researcher then because every paper I've written starts as a very vague "here are the overarching implications and important results". But the detailed order of results and the nuts and bolts of how to argue out the conclusions gets decided in drafting. Only the simplest of results I've had is essentially pre-written.
>"here are the overarching implications and important results".
That's the outline.
I doubt an LLM would help much in deciding how best to present the finer details, as they will be very specific to your particular manuscript.
> How would the LLM "know" which of these novel findings were in any way meaningful
Given that they are trained on all of arXiv, ..., it's much more likely they are aware of all public relevant papers than your average researcher.
A researcher on any particular topic is not supposed to be an "average" researcher, but already deeply familiar with their subject.
Even speaking is thinking. This is why free speech is the very First Amendment. Whoever denies your right to speak freely is controlling your thinking.
Writing is thinking with a superpower. It's like using the "Pensive" from Harry Potter, depicted in the scene where Harry and Dumbledore pull memory whisps out of their temple to rewatch in a mirror pool. Writing enables you to apply your attention to an idea at multiple levels of analysis with significantly less effort than doing the same while also preserving the idea in your head manually.
> Even speaking is thinking. This is why free speech is the very First Amendment. Whoever denies your right to speak freely is controlling your thinking.
There's an excellent podcast (Radiolab, possibly) about how this conception of what the first amendment means is rather recent (1910s-1920s) and that the ideas of what "free speech" meant before that are really radically different.
https://voicebraindump.com
I see a lot of people say "writing is so important," and I think what they mean is "I feel really smart/good when I write." And I think what they are experiencing is that they've been assembling ideas in their heads for weeks, and only when it's all come together are they ready synthesize that information at a higher level, and they mistake this synthesis for the writing itself (rather than the writing being a symptom OF the synthesis -- if they had tried to write a week prior they would have found it unproductive).
This is not true for at least me.
Let's say, I am making something concrete by putting ideas, thoughts, knowledge into paper. While doing it, I am finding gaps and mistakes and finding opportunities to correct them. But it is not limited to 'correction', it also opens newer dimensions and perspectives- ones that previously didn't exist in my conscious mind.
I consider writing as a tool of thinking. Another tool is brainstorming with a group, or any group discussion in general. These amend to your thoughts, make the existing ones more solid, and opens new direction, and unravels connections previously not accessible.
Read this essay by Paul Graham: Putting Ideas into Words [0]. And also refer to his other essays on writing.
There is also a great book by Paul Zissner: Writing to Learn. I suggest this book to people.
Writing, when done while learning works akin to teaching- one of the most crucial steps in so-called Feynman Technique of learning.
[0]: https://paulgraham.com/words.html
Anything by William Zinsser is worth reading.
https://www.goodreads.com/author/show/7881675.William_Zinsse...
Yeah I think I've read multiple PG essays on the importance of writing, but they always struck me as no different than Katy Perry telling you to sing, or some Grandmaster saying "go play chess." That is -- a personal anecdote that doesn't necessarily generalize.
I'm not saying that writing can't be a useful tool to organize ideas, definitely it can. But I think I've found two things:
- Now the best way to "iterate" my thoughts is to rubberduck with ChatGPT; it's really amazing how much faster I can learn when I admit how little I know, even on something like global warming or an advanced math topic.
- By and large, "organizing my thoughts" isn't really a high-return activity in my life. Having an intelligently written blog that I've put hundreds of hours into has never done anything for my career or led to any personal connections, and honestly who's to say my time wouldn't have been better-spent just coming with some jokes to network better rather than having some cohesive theory of everything that nobody asked for?
I just noticed that I wrote William Zissner's name wrong. Please read 'Paul Zissner' as William Zissner.
Just one personal anecdote: I definitely find contradictions or gaps in my thinking/knowledge when I write. Finding and resolving those deficiencies is what I point to when I say "writing is thinking".
Getting stuff written down makes me realise the weak points and the errors in my thinking which I know from experience I don't find if I don't write it down.
This is a really cynical take. People work differently and get value from different things. It’s probably safe to assume most aren’t virtue signalling about writing.
Sometimes I like to test whether I can actually construct or assemble a finished something in my mind from an inkling of a thought.
For example a few days ago I realized that I found it hard to reverse a word in my mind, even a simple one. Try for yourself, think of a word and then reverse it in your head with your eyes closed.
Some people might struggle with the above, some may find it doable in their heads, but most can agree that it's absurdly easy if you can externalize it to paper or a text editor at least.
I think you're right. I'll add on: there's a lot of thinking that does not need writing, and there's a lot of writing that needs no thinking. Deng Xiaoping and other greats wrote pretty minimally for their own thinking, if at all. Whereas many of us not-so-greats seem to knee-jerk comment without a single thought.
It makes sense for our age. Amid a thousand distractions, typing on the keyboard gives the illusion of getting a grip. Note-taking on my computer gives the illusion of a second brain. Ululating on the internet gives the illusion of sharing thoughts.
Instead of "writing is thinking", I prefer "thought precedes speech" https://inframethodology.cbs.dk/?p=1127; it fits the small human mind better though I've yet to learn it properly.
You need to add into consideration that laying things out in visual display provides cognitive support, reducing the effort to reason about more things. So writing out ideas really does allow you to reach a greater scope of synthesis.
it depends how you write.
often when I write an idea down, my "inner critic" process gets more activated upon seeing the textual representation.
thus I find gaps and flaws more easily.
not true for all domains, but many.
If they tried writing a week prior that would have realized sooner the gaps in the ideas they were assembling, leading to them closing those gaps faster.
I disagree. Not sure how common this workflow is, but I write by putting all the different unsynthesized ideas down and rearrange them as the latent structure "reveals itself." At the end you have something synthesized.
Sure some type latent structure was there all along (thus why I put them down), but it wasn't necessarily visible to me, nor optimal, nor did it include/exclude all the right points. The need for iteration itself proves that the act of writing is actually doing the synthesis.
[dead]
While I agree with the underlying message, "writing is thinking" is only circumstantially correct. It wasn’t always like this.
We learned to think by writing only after writing became cheap. Yes, we’ve trained our brains to develop ideas by editing raw thoughts on paper, but it is just one of the possible methods.
I have read a lot of late 18th, 19th and early 20th century books and diaries, and it is plainly clear that writers such as Tolstói, Zweig, Goethe and others developed full books in their mind first, then wrote them from cover to cover in 20-30 days.
Thinking used to be detached from writing. That is a fact. We just lost that ability in the modern era thanks to cheap writing technology: pen and paper, then computers. I'm not saying the current approach is wrong, but don't assume that the only way to think is to write.
Socrates argued that writing would destroy people's memory. He wasn't 100% wrong, yet here we are. The criticism towards the use of LLMs is so deliciously ironic. The analogy with writing... writes itself. Kids that grow up with LLMs will just think differently.
You seem to be responding to the reverse, “thinking is writing”, which I agree is not true, you can think without writing.
They are making the point that writing is more than dumping a completed thought. The act of doing that helps you to critique your dumped thoughts, to have more thoughts about your thoughts, to simplify them or expand them.
It’s easier to go meta once you dump your state.
I think you are right and I understood it the other way around
Kind of ironic, though - I wrote, but my thinking process wasn't so great :)
Thanks for the correction!
Before paper became cheap, wax or wooden tablets were used for ephemeral writing.
> I have read a lot of late 18th, 19th and early 20th century books and diaries, and it is plainly clear that writers such as Tolstói, Zweig, Goethe and others developed full books in their mind first, then wrote them from cover to cover in 20-30 days.
I seriously doubt that it was ever common for writers to compose a whole book in their head and then write it down. Maybe some writers with exceptional memories did this. But there's a whole book about how War and Peace was written based on textual evidence that wouldn't exist if it had simply popped out of Tolstoy's head fully formed: https://www.amazon.com/Tolstoy-Genesis-Peace-Kathryn-Feuer/d....
Not war and peace, which was episodic, but smaller novels were thought out in Tolstoy’s mind before being written wholly. He mentions this in his diaries. Zweig mentions the same, too, but of course his novels are generally much shorter than the two Tolstoy’s masterpieces.
Holding long epic poems in your memory alone was once a celebrated skill.
AFAIK the dominant theory is that they weren't memorized as a whole text, but composed on the fly with the help of a memorized set of stock formulas. [0]
[0]: https://poets.org/glossary/oral-formulaic-method
So, basically the ancient precursor to the skills of a good rapper.
> writers...developed full books in their mind first
When reading long, closely reasoned passages of medieval philosophy, I've wondered about their development process, when there was no such thing as scratch paper.
> Kids that grow up with LLMs will just think differently.
People are just glibly saying this sort of thing, but what specifically is coming? I'm now wrestling with the problem of dealing with university students who don't hesitate to lean on LLMs. I'm trying to not be dismissive, but it feels like they are just thinking less, not differently.
As a similar but distinct theory, you might find Larry McEnerney‘s work interesting. Writing has two classes: a writing for thought and a writing for communication. Larry uses horizontal and vertical spatial metaphors here. Writing for thought still pre-dates cheap paper (and Socrates), but is mostly a private act. Writing for communication is a broad enough brush to span fiction and journalism. For his part, Larry teaches classes aimed at thesis writers who struggle to bridge the divide of using writing to think about a problem to conveying their answer in a paper.
Yeah, LLMs are entirely different from "writing" because they're creative agents. So, writing allows me to give my thoughts several passes, to edit over time. It's like I can have several of me to think, write and edit, spaced over time.
LLMs are like I have someone else to do some or all of the thinking and writing and editing. So I do less thinking.
A bicycle lets my own energy go father. Writing. A car lets me use an entirely different energy source. LLMs. Which one is better for my physical fitness?
Btw the idea about Tolstoy and others keeping those massive books in their head and cranking them out over a month is fascinating. Any evidence or others who imagine the same? In Tolstoy's case, he was a count and surely had the funds, no?
I’ve read Tolstoy’s diaries and he mentions the thought process he uses to write small novels. First he thinks about what should happen, then he writes (or dictates) the text. Thinking takes a few weeks, sometimes a month, then writing is pretty quick. There is some editing, but nothing like we do nowadays.
Bigger novels such as war and peace were written episodically.
> Thinking used to be detached from writing. That is a fact. We just lost that ability in the modern era thanks to cheap writing technology: pen and paper, then computers. I'm not saying the current approach is wrong, but don't assume that the only way to think is to write.
I have a better way to frame this:
Learning your own language and culture is a lifelong process.
A big phase, the adult phase, of learning is learning to write in your language (I'm implying there's more to writing than chosing words; specially in this context of language as thinking)
indeed, a lot of modern people never make it out of this big phase of learning your language. they never go beyond writing = thinking. but some people do learn the next phase
which involves distinguishing language itself from thoughts and ideas (is some idea known? understood? perceived?? but the idea is "the self" or some other complex notion)
so the only quality of the modern era I admit, is that there's a lot of people that only learn rudimentary thinking-writting, and too few people that learn 'advanced' languange-thinking where writing becomes secondary to thinking.
finally, I learned this idea from reading around the meaningness blog/book
[dead]
> We learned to think by writing only after writing became cheap. Yes, we’ve trained our brains to develop ideas by editing raw thoughts on paper, but it is just one of the possible methods.
I think you have some misconceptions here. First, the article does not claim that thinking is writing, and especially not that there is no thinking without writing. They only explain that writing is supporting and driving a higher quality of thinking.
Second, paper isn't the only medium to write. And writing isn't the only persistent form of communication to support and improve thinking.
> Thinking used to be detached from writing.
It still is.
I'm not sure if writers developed the entire book in their head first, but: it was indeed very, very common for people to dictate novels, journal entries, and other "written" works to a secretary, typist, or tape recorder.
Nowadays that seems to be rare, but my impression from reading my journals is that it was often more common to dictate than to physically hand write things.
Novels were dictated, that is absolutely correct, but on top of it, the whole plot was developed with a high level of detail before dictation started. There was some editing, of course, but nothing like we do today, where writing books is basically an iterative process. We lean on the written word too much for our thinking (not being critical, just that’s how we are taught)
That's true, but I would phrase it from a different perspective:
It's seems clear that abstract thinking in particular is greatly aided by writing, because the written text acts like a thought cache. A bit like an LLM context window which you can fill with lots of compact, compressed "tokens" (words).
Abstract thoughts are "abstract" because they can't be visualized in our mind, so they don't benefit from our intuitive imagination ability (Kant's "Anschauung"). So it is hard to juggle many abstract thoughts in our working memory.
We can also think of the working memory as the CPU registers, which are limited to a very small number, while the content of the CPU cache or RAM corresponds to the stuff we write down.
Our "anschauung" (visual imagination) is perhaps something like a fixed function hardware on a GPU, which is very good at processing complex audiovisual content, i.e. concrete thoughts, but useless for anything else (abstract thoughts).
Many people here are taking away from this that substituting our own writing with the writing from an LLM is a danger for human development.
I see things more optimistically. If good writing leads to good thinking, then anything I do to improve my ability to write well transitively helps me to think well.
In that sense, I actually see a huge benefit to LLMs in improving my writing and therefore improving my thinking. Not only can I ask for detailed and powerful feedback, I can also ask for more details on background context or related topics that I wouldn't be aware of.
I believe judicious use of LLMs can make us better than we could be without them.
I subscribe to this view. LLMs can be tools for thinking, not substitutes to thinking. In fact, I feel that with time, the way we think may fundamentally change, just like how reading and writing changed thinking. An LLM can be an ideal foil to test out ideas, and the process of thinking could be an iterative process with the LLM as an active participant.
Whenever I need to do some hard thinking and things are not clear, I fire up my sublime text and write down the context in the simplest terms and short lines (only few words per line). While doing this, I will be absolutely rude to myself asking extremely basic and direct questions to bring out the real context, real goals and real path. It's like answering an under-world boss. No bullshit, no pretense, no regard to any norms, no impressing someone. Then the whole thing falls into a meaningful structure. I leave it when it produces some immediate action items.
A former manager (and editor) used to say writing is discovery which is more or less the same thing. I agree.
If writing is thinking, then perhaps we should think more about what language would produce the best thinking, and does it exist? How do we go about creating it?
And if thinking is dependent on language, maybe we should create a new language for artificial intelligence rather than feeding it human languages.
There's been a lot of work in this area but no definitive answer, and few constructed languages speakers [1]. I imagine the wins may be marginal, similar to the wins in dvorak vs qwerty typing.
--
1: https://en.wikipedia.org/wiki/Linguistic_relativity#Artifici...
With the latest technology we can fix that. Page full, head empty.
My experience with writing is that it's often a matter of simply noting down what comes pouring from my subconscious, with the most taxing task being keeping ahold of some given connection, sentiment, or wording (as that subconscious seems to generate more than my working memory can keep track of).
Sometimes I struggle to fit those sentiments and connections to wording that I imagine will make sense, to someone else or even to myself. I guess that would be the, "Writing is thinking," part, but it seems more like, "Effective and coscientious (self-)communication is thinking."
We may enter a vicious loop where writing is increasingly generated by LLMs. Then, LLMs have to train on their own output leading to model collapse.
Hence, the models depend on human writing.
This intuitively makes sense (like deep-frying a JPEG), but it doesn't seem to happen in practice, as modern models are frequently trained on text both output from other models, and curated from other models.
Realistically, going forward model training will just need to incorporate a step to remove data below some quality threshold, LLM-generated or otherwise.
New HN metric: how many comments do you have to scroll down on a LLM-edited/written submission before someone points it out? Triple bonus if the submission is about LLM writing and the need to think for yourself without a LLM, and still no one has pointed this out. Current page score: 333 points.
What makes you believe this was written by an LLM?
Because it has many of the typical 4o stylistic tics like 'it's not X, it's Y' or enumeration or the em dashes, or the twist ending.
It's not 100% unedited ChatGPT and far from the most blatant instance that has caught my eye (they've started showing up in the New York Times and New Yorker as well, have you noticed that?), but certainly sounds like that was used: "Writing compels us to think — not in the chaotic, non-linear way our minds typically wander, but in a structured, intentional manner." "This is not merely a philosophical observation; it is backed by scientific evidence." "Importantly, if writing is thinking, are we not then reading the ‘thoughts’ of the LLM rather than those of the researchers behind the paper?" "overcoming writer’s block, provide alternative explanations for findings or identify connections between seemingly unrelated subjects."
(Note that this is particularly ironic because as the op-ed notes, if they did use it, they are required by Nature to disclose this... https://www.nature.com/articles/d41586-023-00191-1 But of course, how would anyone ever prove they did so? You know how difficult it is to get Nature to retract even blatant fraud.)
Writing has always been how I organize my thoughts. A lot of ideas only become clear as I work through them on the page. AI can save time, but it also makes it easy to skip the slower, more reflective parts of thinking. For me, that’s often where the real value of writing comes from.
Related (from a few days ago)? https://news.ycombinator.com/item?id=44641669
We can't really talk about this topic without mentioning the book "How to Take Smart Notes" by Soenke Ahrens
One key takeaway is if you want to learn/remember something better, always rewrite things in your own words as both the act of writing AND paraphrasing makes it more sticky
Part of me can’t help but think that scientific journals as billion dollar industries need to keep the status quo of how articles are written, where they get submitted and who reviews them. Even though per review today is failing
Maybe as a coder, I will someday will just calm down and write code without using any LLMs, just for train my thinking capability...
Writing what someone else wrote is thinking what someone else thought. My favorite learning technique is reading, listening or viewing something and then typing it into libreoffice. Specially useful when it is something that is transcribed. Works really well for code, too. Give it a try!
While the article’s point seems intuitively true, it cites only two papers related to the benefits of handwriting. But it’s argument is stronger than that. Is there peer-reviewed evidence to support the stronger claim of the benefits of typewriting?
I love this! Especially the part about if writing is thinking, are we not then reading the ‘thoughts’ of the LLM rather than those of the researchers behind the paper?
I would also compare reading to being reprogrammed like EEPROM. Although the process is slower, the changes feel more permanent when learning: you need to create connections yourself from examples compared to someone demonstrating it on the video.
This is one of the most important reasons why it's so harmful to let an LLM write for you: what they're actually doing is thinking for you.
I wonder how much of this applies to coding. Is coding thinking?
It seems obvious to me. Just like drawing and sketching helps thinking about design, coding helps thinking about programming.
It's one of the reasons for the "one to throw away"-idea of writing shitty code first just to get it to work, and then remake it after you have thought through the problem by coding it.
True. Writing by hand is even better.
Using an LLM is also thinking, when used as such.
This morning I asked ChatGPT a question about how Quickbooks handles charts of accounts compared to NetSuite. It answered my question better than anything else would have.
Also, I'm currently using Claude Code to fix some bugs -- it's handling the heavy lifting while I think about what needs to happen.
I'm in favor of human writing as an underrated tool of culture-making...but the scope of what counts as "thinking" is expanding.
Writing needs a conceptual split analogous to the split between math and calculating.
Just as a calculating can be implemented on a computer which has low cognitive abilities but high algorithmic and procedural abilities, we need to extract out the word-smithing capabilities from writing separate from the thinking portion. Our lack of distinction in terms reflects a muddled conceptual framework.
LLMs are excellent wordsmiths completely divorced from the concept of thinking. They break the correlative assumption - that excellent writing is corresponds with excellent thinking. Until now, we've been able to discern poor idea because they have a certain aesthetic, think conspiracy rants in docx saying something about a theory of everything based on vibrations. But that no longer holds. We have decent enough word-smithing coupled with a deficit of thinking. Unfortunately this breaks our heuristics with consequences ranging from polluting our online commons to folks end up believing nonsense like ChatGPT named itself Nova and they are a torchbearer for spiritual gobbledygook.
My point is that we're in the process of untangling these two and as a result, we're likely to see confusion and maybe even persistent misunderstanding until this distinction becomes a more common part of how we talk about and evaluate written work. They're living in an AGI-world and we're just..not.
We think by association. We can sometimes tighten up the process when there's a formal logical framework that applies, but it's not as natural or automatic.
What writing changes is that in words, you have to make it explicit how one thing leads to another. Partly, that's just due to the imposition of sentence structure.
Ironically, this is precisely the crazy thing about Trumpspeech: it's just associations - vibe-chaining if you will.
> Current LLMs might also be wrong
Current humans might as well :-)
The best way to learn a subject is to teach it.
I've been saying this for years.
Memory is intelligence
It is interesting to hold in one's head the following two beliefs:
Writing requires thought.
LLMs do not think.
"Writing is thinking" does not entail "writing requires thought".
Which has a lot to do with how people intuit when text is LLM-generated.
If this were to be an analogy to AI, would inference discover information that wasn't found during training? Is this where hallucinations come from?
To write something you must form a "theory of the system", which is orders of magnitude more difficult than feeding on fodder. It's the difference between being fluent in a language and sort of understanding something.
It's quite similar in hard sciences as it's in natural languages. For instance I don't understand Hungarian at all. Few words "igen", "jó napot kívánok" doesn't a knowledge of the language form.
Then German. I had to learn it in school so I have orders of magnitude better grasp at it because I can actually say a few statements that form in my mind: "Nein, ich brauch nicht ein anderes stück Steak". Might not be 100% correct gramatically and vocabulary wise but it conveys the message and also transmits that I understand the context.
And then come English which I speak since 33 years. I actually THINK in English a lot of times and there are concepts I can't easily express in my native Romanian language without resorting to a painfully long and sometimes unsuccessful software-driven (as opposed to FPGA-encoded) translation process.
This is an important point:
> For example, LLMs can aid in improving readability and grammar, which might be particularly useful to those for which English is not their first language.
I don't know whether this has been empirically confirmed, but I have the strong belief that a manuscript with poor grammar, by a non-native English speaker, has a much higher probability of being rejected than the same manuscript but copyedited by something like Grammerly or a SOTA LLM.
Ideally writing style should matter much less than the quality of the research, but reviewers are not just influenced by objective criteria but, unconsciously, also by vibes, which includes things like writing style and even formatting.
Meanwhile, more and more posts on discussion forums etc. is clearly "copyedited" by these tools and the result is quite grating for the regulars.
Probably less grating though than broken English. (Copyediting is also different from pure LLM replies which don't involve editing anything.)
Broken English still has its charm and brings the structure of the writer's native language to the fore, which makes it relatively easier to parse and glean intentions from than polished LLM-speak.
That might be true, but I think it's false. Or more precisely, I think manuscripts with broken English have statistically a higher probability of being rejected than ones that are copyedited with AI.
writing is logical original
They are delusional if they think that the only use of LLM for scientific research is to correct grammar.
the act of writing takes raw experience, puts it in front of the eyes, and then filters it back through your critical faculties so you can refine and reason about it. the iteration makes it higher quality thought.
making writing valuable is another skill (see this evergreen lecture from the university of chicago leadership lab: https://www.youtube.com/watch?v=aFwVf5a3pZM)
however when I encounter people with low written or verbal acuity, they have to survive somehow, so it's wise to observe what tools of cunning they tend to reach for.
[dead]
relevant PG essay https://paulgraham.com/writes.html