Writing is thinking

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.

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.

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.

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

No , reading is filling up your mind's LLM with the context given by someone else. Your thinking is what happens after reading.

> 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.

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.

> 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.

The LLM can make a plan or outline first, which is also writing.

> 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.

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

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".

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.

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.

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.

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.

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....

> 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?

> 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

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).

> 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.

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.

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...

Meanwhile, more and more posts on discussion forums etc. is clearly "copyedited" by these tools and the result is quite grating for the regulars.

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.

What makes you believe this was written by an LLM?

"Writing is thinking" does not entail "writing requires thought".

Which has a lot to do with how people intuit when text is LLM-generated.

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.