It started out as an interesting read, until I got to this paragraph:
> The value didn't come from what was said at that moment. It came from what had been built across many such moments: a pattern of mutual recognition, a shared context, a baseline of trust that made the later exchange possible. The relationship was the infrastructure. The conversation was where it had been built, one cup of coffee at a time.
And that made me immediately question the worth of the entire piece. I get it, LLMS can rewrite an entire blog post in a minute, which can be quite tempting for people that don't enjoy writing itself, but it just takes so much variety away. I think people should stick to grammatical corrections only, and not rephrase entire paragraphs (never mind letting an LLM write everything in the first place).
Usually the smell is when it sounds profound or insightful but then you realize it's word salad and not really saying anything. LLMs also try to make the most mundane thing sound revolutionary or write about it in an alarmist way like 'increasing x by 1% is a ground breaking innovation' or something.
There's no concept of what truly is important with the writing itself since there's no actual thinking going on. The opinion of the supposed writer, which colors how they structure the writing and the language they use, is often all over the place since an LLM has no real opinions.
It's the cadence and the density of cliches. I've seen Opus produce very similar output. That's not to say it's 100% that this was written by an LLM. They had to learn this style from human writing in the training data. It's just a very condensed form of bad writing.
"Density of cliches" is a great way of putting it. It's exhausting to write that way and unpleasant to read back in the edit process which is why that kind of thing doesn't happen (often) in human-led pieces. LLMs don't get exhausted.
I think whenever a blog post begins to sound like a marketing campaign from versace - it's always a great sign that it may be LLM generated.
This isn't just a comment. It's a an experience. A shared cognition between parties -- To collaborate & exchange ideas. It's not procrastination, it's culture. Culture is what builds civilization.
This shit gets me irrationally mad. Well, the existence of such content doesn't, but the fact that other people aren't disdained by it (this dogshit has almost 300 upvotes on _HN_ right now) is what gets me.
The topic brings an interesting melange of people together. That makes up for it and made me upvote.
I too noticed a shallowing of the piece somewhere in the middle. It started with an interesting observation but ended with triviality (like you'd expect from a synthetic mind without a body).
As someone who doesn't follow the specifics of how LLMs write, how can you be sure it's an LLM?
I'm asking this genuinely because I've been using stuff like these separators before LLMS - now someone can just say "ahh dismiss the content they used style X it's LLM, it's dogshit".
I am not sure if I would manage to correctly identify every typical phrase, but the one I quoted I am absolutely certain of.
You could try asking an LLM to write some prose about whatever you fancy. Maybe two or three different pieces, and ask it to be extra profound. You will probably see that exact sentence structure above a number of times:
> It’s not [X], it’s [Y]. A bit of [A], the [random adjective] of [B], all weaving together to form [C]. A short sentence. Maybe two. A final, culminating closer, gradually transcending into space.
The words are quite interchangeable, but the you will know it when you see it.
I don't think the out loud or someone listening / reacting matters at all here. Suspect it's entirely this:
>The thought that was comfortable as a vague impression has to become a sentence, and sentences have structure.
It's not unlike what people like PG say about writing improving thinking...it's the being forced to go from fuzzy directional notions to something you can put on paper in that will stand up to critique.
Same with rubber duck debugging. The verbal part means you need to articulate it clearly but it's not the speaking that helps. Same with writing a detailed spec/prompt for an LLM - I know if its too fuzzy ("set an appropriate timeout") the LLM will spin it's wheels so it forces clarity.
Also suspect that a big part of who we consider intelligent is linked to this. Maybe their internal monologue is just more crisp - closer to what they'd tell a rubber duck.
- genuine dialog between 2 peers with overlapping knowledge on the topic, both being kind, patient, open-minded yet critical, able to point flaws in reasoning or provenance without being hurtful
- genuine dialog between 2 peers with overlapping knowledge on the topic
- genuine dialog between 2 peers
- genuine dialog between 2 persons
- thinking out loud in front of someone listening
- thinking out loud alone thinking of someone listening
- thinking out loud alone
The "trick" here, and that's why I wanted to write it down, is that having an actual person, real or imaginary, does force one to actually explain. A typical fallacy is the "proof left out to the student" which actually happens to be terribly complex or even infeasible.
I believe trying to explain to someone forces you to identify the gap into their knowledge and your point. It's precisely mapping where knowledge does NOT overlap. So instead of making a lot of assumptions they have to be laid out, at the very least made in part explicit. Doing so forces you to stop skimming parts that are challenging for you. The part that you might not feel so sure.
Obviously the whole problem is that this peer, who is both knowledgeable and kind is good in their field and popular, thus busy. Consequently their time is very VERY precious. Hopefully you yourself are that peer. That means that often we have to go down the list and find the next best person.
So... I do think duck debugging is better if you genuinely trust in the kindness of the duck, not see it as a piece of plastic ;)
I don't think the out loud or someone listening / reacting matters at all here. Suspect it's entirely this:
>The thought that was comfortable as a vague impression has to become a sentence, and sentences have structure.
I often construct full sentences in my head. And have conversations with my mental model of some other person. In full sentences
I was reading the parent and thinking the same thing - I can't do that. There's no "voice" in my head like some people describe. If I want to model speaking like a mock conversation, I have to write it or ... speak it, even silently, to convert from the visuals that are my thought.
As an aside, I end up having to organize certain types of thoughts through speaking, which requires some adaptations and limiting functions :-)
I'm with you there. I have to either hear the full sentences narrated by my internal voice or see the words flashing in my mind in order to "think". This is great for building and maintaining deep mental models but it is also highly susceptible to "bit rot" (such as forgetting the rationale or evidence for a specific assertion or position) days/weeks/years later. I have a friend who simply can't understand how inter-linked note systems (like the kind Obsidian enables) are helpful. It's just a bewildering mess to them and they think more linearly.
Thus, writing things down is a necessity for me: it's not for a need for structure but rather that my "context window" gets filled too quickly. I can counter my own arguments but it's more fun, and often quicker, to do with someone else. Besides, there is such a diversity of thinking out there it would be foolish to not take advantage!
There has been many a time where the process of thinking through in my head how I was going to describe the problem to someone else on the way over to them, that a testable idea pops up.
Could probably have gone for a walk around the block instead, but the formulation of the problem to be able to explain it to someone else seems to be key.
Same here. When I was younger I liked imagining having conversations with people from the past like Plato or something, thinking about their reactions to The Future. (Though I guess I only really imagined my own side of the conversation.)
Haha, I found this genuinely funny. It’s called an internal monologue. Google tells me that it’s the majority position with 30 to 50 percent not having one.
Personally, I find it very difficult to understand how people could not think in words, like you were speaking to another person. Obviously you also have mental imagery and sound etc, so not everything is just words. Internal speech is one channel of thought, but for anything complex I would have thought it was mandatory.
I personally don't since I suffer from aphantasia. Perhaps I also suffer from not having an internal monologue?! I do think in words but not in complete sentences, it kind of happens much faster.
It might be correlated with neurodivergence. But it’s also just correlated with high verbal ability in general, presumably. I find it quite natural to think in complete structured sentences, and it’s often perplexed me why other people seem to find this concept so alien. And no I’m not schizophrenic.
I think I've got this from growing up with a narcissist. Thoughts are constructed and seemingly endlessly whittled to try and create a sentence that can avoid getting belittled.
It's hard to talk in groups, because you have to have a sentence mentally critiqued by 3/4 people in turn, so the topic has usually changed before you can say your piece.
What? It's never occurred to me that this isn't entirely normal, I've done it all my life. I thought people without an inner monologue were the unusual ones.
I'm not wading into the schizophrenia part, but inner monologue doesn't necessarily imply constructing a fully formed sentence you then repeat to yourself.
Created a throwaway for this, but is this not what people do?
Not so much imagining a conversation in your head, but playing through a conversation with "the other side" of the coin (be it an idea, plan, problem, solution etc) which are both yourself if that makes sense.
Is this not what the inner monologue to yourself is? Your inner is conversing with your outer monologue. Hard to explain.
Would consider myself to be both physically and mentally stable, no conditions etc.
I can only speak for myself... but my inner monologue is just a runaway train of thoughts that I vocalize in my head as I work through them. There is no other person in there, abstract or otherwise. Have you ever seen the Death Note anime? Similar to that.
I do what the person you're replying to and for me its just one voice or a faint imagination of them speaking. Its me roleplaying as the other person, not like there is a different entity in my brain.
I think there are tradeoffs though, and this has been a thorn in my side during technical interviews where you are expected to think out loud because:
1. Sometimes you have a vague sense of the shape of the solution, and ime it can be helpful to sit with it for a while before trying to shape it into words.
2. Talking out loud forces structure but it also rate-limits how quickly you can iterate through ideas to find one that plausibly solves the problem at hand
"When you are a Bear of Very Little Brain, and you Think of Things, you find sometimes that a Thing which seemed very Thingish inside you is quite different when it gets out into the open and has other people looking at it."
This was my first thought too. Thought, I will say that even though I think in speech for basically everything conceptual, and find it difficult to imagine even being able to not do this, I do often have these… ill defined ideas. Notions of something or other. I have a few words to describe them and they just sort of point in a direction rather than being fleshed out. It’s this sort of thing, like a thought-stub, that benefits the most from inspection and being forced into a better defined form. These are also often the more difficult things, and the reason they’re fuzzy is that it takes a lot of effort to make them sharp.
It seems like you saying that people who can do alone what most people do alone, have an advantage. Sure but seems more like an agreement with the article than otherwise.
I find this whole concept very interesting, because unless I'm translating it for external reception, my thinking is largely non-verbal - generally more spatial-visual, proprioceptive, or purely tacit than auditory, and I find words to be, at least by that standard, frustratingly limited.
I think that’s part of the point. Execution at its core is mastering the art of projection onto a lower dimensional space.
Taking anything from up in the mind out into the real world, be it a physical action, a speech, an artwork, requires you to make decisions about what’s load bearing and what’s not.
So the more practice you get translating your thoughts into words (or another medium), the better you get at executing.
The "another medium" is doing a lot there, I think the translation between my understanding and expressing it in words is the weakest aspect of software development at the moment, in addition to the entire physical world. Dancing about architecture, if you like.
From my own experience chatting with LLMs: reading the responses definitely does help with thinking, even when you can see obvious flaws or hallucinations. It gives you something to think about, which a rubber duck can't do.
The post you're replying to didn't say that, but regardless, I would say 'thinking aloud' is a pretty good descriptor for the word-by-word narration kind of thinking where you sort of both say and hear it in your head. I have heard that some people who who experience 'inner monologue' experience their thoughts nearly exclusively like this.
In 2017 LLMs weren't powerful enough to generate working code on their own, but my goal was to at least create a chatbot that could help you rubber-duck-debug your way to a solution. Unfortunately the tech wasn't quite strong enough for that, and not enough engineers even knew what rubber-duck-debugging was. RIP Duckly.
Trying to train an LLM on two 1080ti's on the StackOverflow corpus in my living room was a vibe though. Good times.
Duckly deserved to actually work. There’s a small irony here: the closest study I found to this, robots specifically built to simulate attentive listening, found they performed no better than an actual inanimate rubber duck for adult engineers. The mechanical signal of listening doesn’t seem to be the active ingredient. Makes me wonder if Duckly would have needed real disagreement to close a gap a duck can’t, not just better natural language.
You're probably on to something with the value of disagreement. I think it's one reason why chatting with current models doesn't create the same stimulation as rubber-ducking used to bring. The models are typically too quick to agree and amplify what you think rather than truly break it down and push back.
And thanks for saying it should have worked, I agree. My chagrin has increased over the years as I have realized the magnitude of my ill-timing.
Has anyone seen a good set of prompts for that disagreement? For the "skeptical eyebrow-raise" or "confused/doubtful head tilt" aspect of rubber ducks?
Agentic uses adversarial expert, steel-man opponent, risk-mitigation and failure-mode analysis. But what about almost brainstorming, but with thought-provoking nudge questions? Or on the other hand, arm-waving fight-club style discussion? Or... It's a big design space. I used to go to lots of research talks at MIT, in assorted departments. The post-talk Q&A question cultures varied a lot. Like encompassing both "leaves the speaker in tears", and "nudge so subtle, you won't quickly get it if you've not already spotted the fatal flaw in the work".
So aside from dialing down the "transformative insight!" silliness, there seems a rich multi-agent multi-persona space to explore.
I think agreement has value here too. An LLM that's starting to get a bit sycophantic will rephrase your ideas in a few different ways, and seeing the different presentations is helpful for reconsideration.
I wonder how much is actually needed to create an automated rubber duck. How well would ELIZA work? (https://en.wikipedia.org/wiki/ELIZA) (might need some adjustment to not talk like a therapist, but you get the idea)
2017 is a bit early to refer to them as LLMs. I'm not sure when exactly we started to refer to LMs as 'large', but I don't think it was before GPT2 (2019). That said, from the NLP work I've done, it was much more interesting working on small specialized models.
OMG - strong vibes to Einstein crediting Michele Besso, his colleague at the Swiss Patent Office, with helping him discussing some concepts in the special relativity paper: see at the end of the paper https://www.fourmilab.ch/etexts/einstein/specrel/specrel.pdf
I was wondering about the question of whether people who made very deep discoveries (Einstein and Godel come to mind) had others to talk things through with beforehand.
I know Andrew Wiles kept all of his work on Fermat's Last Theorem secret and by that I assume he never talked it through with anyone.
I saw a Facebook copypasta piece that claimed that Einstein's first wife came up with many or most of his ideas, and never got credit because of sexism. No proof whatsoever, other than she was a mathematician and physicist.
But "it could have happened!" is more important than even a microshred of evidence for highly emotional, online topics.
This anecdote nicely pokes a hole in that conspiracy theory: he was thoughtful enough to share credit with a layman work associate, but (supposedly) not the most important woman in his life - that seems even less likely.
Tangentially, Regarding pair programming (as a special case of thinking together):
Programming is serializing ideas into the computer language. Communicating them with someone else first serializes them into human language, which is already much less abstract compared to the thought cloud in your head.
In the case of an effective pair programming collaboration, you also get to debate approaches, discuss details, alternate between coding and watching.
It also helps that the presence of someone else helps avoid many common distractions. Reading non-urgent private messages and checking out HN (I'm no longer so addicted to any other platform to check it out at work).
It's simpler than this. Explaining a question/issue to someone involves going back to basics and covering all the foundational info that a third party would not have. When thinking alone, you gloss over this, and may not realize that a foundational assumption is incorrect. When you are forced to explain every step explicitly, these errors or gaps can become apparent without any intervention by the listener.
I have my younger kid explain each math problem to me before she submits it on Khan Academy. My older kid thinks in her head how she would explain a problem before turning in a test. It's a good habit to form.
As much as people make fun of "new math", it is really neat to see my 3rd grader working through a math problem step by step in a way that makes sense to them instead of the rote memorization I had to do as a kid. While they don't like showing their work, it helps them to work through each step to make sure their assumptions are correct, like you said.
Yes, it is nice that there are more ways that are taught now. It can turn into too much, however. Some parents I know dislike that their kid is taught 8 different ways to do long division, at the expense of having sufficient time to practice any of the methods. They end up taking the kids to math tutoring so that they can actually learn how to do the mechanical steps.
But it is for sure good that there are multiple ways to do various operations, and there are plenty of youtube videos that students can turn to if what they have been taught does not resonate for them.
In computer science, we have classes on memorizing/implementing various algorithms so that we may understand their behavior. Could anyone name the reason why grade school students memorize and execute the long division algorithm? I certainly hadn't used long division since grade school.
Grade school is years long, so a skill you're supposed to know in 3rd grade is one that you need to be able to faithfully execute for the next 3 years. The fact that you'll get a calculator in middle school doesn't help you if you're struggling through 4th and 5th grade.
In the nineties I was a junior copywriter at BMP DDB in London. The creative department was made up of loads of small offices, each with a two-person creative team working out loud all day. The best ideas rarely came from one person thinking alone. Someone would say something half-formed about a brief and their partner would catch it, often bouncing ideas back and forth. Reckon the whole building ran on that. Well, apart from the late and brilliant John Webster, who had his own office at the top of the corridor. We were lucky enough to have ours opposite his.
This proves nothing but, in my earlier days, I'd come home with something on my mind from work and tell my wife about how I couldn't get something to work the way I wanted. She had no clue what I was talking about but she'd offer up clues and suggestions to which I would try to explain to her how things actually worked.
Somewhere in that process it would lead to a solution that I would bring to work the next day!
I started my web dev career in 1999 so my main code references were a combination of O’reilly and “for dummies” books. As a wet behind the ears engineer I’d find myself regularly walking over to my more senior friend Dan’s cubicle for help.
Half the time on the walk over, trying to frame the question in my mind I’d figure out the answer or at least next step. It got to the point where Dan would see me heading towards him and suddenly turn around and he’d as “Figure it out?” And I’d throw him a thumbs up on the way back to my desk.
Thank you for sharing this. Growing up Asian American many teachers disciplined me and made assumptions about my intelligence for not being as vocal as the other kids. Culture shapes cognition and vice versa.
The move from thinking to semantic conversion is important for investigation/introspection.
Arguing with yourself also seems to engage your brains "theory of mind" centers, so different pathways get activated to examine the problem space.
The problem with Ai is the fact that it hallucinates and if you're doing anything truly novel in an integration or framing sense it bottoms out very quickly and can't engage. A human operator can decompose the problem and get accuracy checks for known areas in the training data of course.
Now to be I'm not saying Ai can't produce novel work on the edge but in my experience it is antagonistic towards those goals.
Case in point, CRDTs, many don't use tombstones but they are the minority, and if you try iterate a new CRDT off of one that doesn't use tombstones, let's say diamond-types, it will keep pulling you back to tombstones.
The problem is that the number of humans who understand dynamic investigation and the push pull of exploring an idea you don't hold with someone has always been very small, and now with reflexive internet argument culture driving how we view "debate" and "discussion".
I don't know if we've reduced the leisure to think or what but things are not great for finding speculative thinking partners.
Great write up, thanks for posting. I think I could have avoided so many mistakes in my life if I had just spoken to someone about my plans first.
And not necessarily just based on their feedback, but just hearing the words come out of my mount and predicting their reactions would have helped a ton.
Really interesting someone brings this up here. It's kind of an obscure little piece of text. But I always thought, yeah, this is more or less how we produce thought, and how LLMs operate, too.
Something funny i discover: i had thoughts or tried to explain something decades ago and made a drawing. Some of these are just a circle with some arrows. Turns out it is really hard to make a drawing bad enough that you dont know what it is. It sometimes works better than even a written description.
So you can just doodle in a notebook and tell yourself you've written it down. If you look at it regularly im sure recall will be even more perfect.
If you dont write anything down you wont be alle to access an orderly array of thoughs.
I draw a lot to figure out engineering details. I used to make beautiful drawings in diagramming software (draw.io and the lil). But man, it took a lot of time. Now I scribble on printer paper unless it’s something I plan to share.
Fellows devs at my company think I'm strange, but I always have a cheap coil-notebook. It has diagrams, notes, doodles, minutes, checklists, mostly not even in order. I try to remember to write the date at the top of each page.
I just counted, 11 notebooks (b5, 60 pages) in two years, so just about a new notebook every 2 months.
Funny thing is, coworkers will now ask to use my pen and paper to flesh-out something together, as I'm certain it's the only paper and writing utensil in the building.
And yeah, draw.io or even an ipad with a whiteboard app aren't good replacements for me.
You are compressing thoughts/ideas to symbols. Via a visual feedback loop your mind can operate on more complex ideas as your working memory doesn't need to hold everything at once. The term "second brain" really applies here. While you're "doodling forth" you become proficient in your on-the-fly invented symbol language.
Key is also the slowdown effect of being careful to draw properly. It's like a harness for your thought process. Fixating your thoughts on paper is a way to be safe not to go astray; like a climber who puts hooks in the wall.
It's really more complex than it seems from the outside. A picture of a group of people drawing lines in the sand comes to mind. And one wonders - once again - how language came into the world.
A different funny that should be archived here with the above.
Assuming you can type well.
Take a sentence, put your hands on the keyboard with the intend to type it but dont do it.
Some time later dont try to remember the sentence but just put the hands on the keys and let them type. Its really funny seeing the text come out of your keyboard buffer muscle memory.
Some japanese guy said he played with the concept and reported that he could fit a whole paragraph or a list of things big enough that he couldnt recall it without using his hands. He could do an invisible abacus to recall it.
You can't think of a sentence and type it later without actively remembering it. You either forgot the sentence by then or you're able to recall it - just typing from memory without actively remembering is just not a thing that is possible. Muscle memory isn't built from thinking one time about what you'd do.
One time early in the morning i couldnt remember the 6 digit accesscode for a warehouse. It seemed completely pointless to go there. Approaching the door i thought and now what?? When close enough my hand went up and typed the code. The door opened and i thought, wait, what was it? The memory was only available if my hand was on the keypad. It then seemed pointless to conciously remember it.
Sometimes you might sit and stare at the same text, grappling with it in silence.
For technical problems, I find the act of writing out a request for help, even just into a text editor, is often sufficient for me to solve the problem at hand. Writing things out is a way of organizing and structuring your thinking, and is itself a powerful troubleshooting tool. Things will become obvious that, unwritten, you might not even notice. I think this is very similar to thinking out loud; when you listen to what you say, or read what you have written, your mind is somehow keyed to react in a useful way.
That response might not be what you want or need for problems that need to be wrestled with, chewed, and pondered on deeply, though.
I don't really buy any explanation here. The reason why thinking out loud with someone beats thinking alone is because when I'm alone I'll loop a lot more.
My reactions to thoughts that I myself have had are stereotyped and repetitive (every reaction is typical of me.) Reactions to me that come from another person are going to be different than those, and in turn stimulate me differently. Thinking tends to be an enumeration (then elimination) of possibilities, and that random input helps with the enumeration part by keeping me out of local minima.
I always found explaining the problem to one of my cats was a good way of getting to the solution, sadly I had to put the last one down on Monday and now my office feels empty and lifeless.
I'd wait a bit and get a kitten but having lost both cats inside a year frankly it hurts too much and they don't live long enough.
I find it very useful to talk to my dog. When I have a problem I'm banging my head against, just explaining it to him from first principles makes my head "click". When that works, his dog treats are billed as "consulting".
Everyone inherently knows this, that's why they reach out to others when stuck on a bug (see also rubber-ducking). But why is it so hard to convince individuals and organizations of the benefits of pair programming?
I wonder if there's a difference between people who have a strong inner voice and those with no inner voice (Anendophasia). I.e. do people w Anendophasia do better on their own than those with a strong inner voice?
> The Enigma of Reason (Hugo Mercier and Dan Sperber, 2017): their argumentative theory holds that reasoning evolved for social rather than individual epistemic purposes, to produce and evaluate arguments in group contexts.
Yes! I love that someone wrote this down!
This seems so obvious to me now. I often ask LLMs to cite their sources (they do hallucinate from time to time), and they often give me sources that don't say what is claimed. "How would the LLM know not to give this to me?" I wonder. They're trained to explain but not to convince, so they don't know what's convincing, and they should.
I think humans hallucinate at least as much as LLMs—arguments of any complexity are impossible to formulate without leaping at least a bit—but other humans ground us. That's why when people become socially isolated, they join cults or adopt conspiracy theories or the like.
Conversely, "this is convincing to an expert" converges on “this is true" as our collective expertise grows over time. This is the foundation of the scientific method, of progress in all engineering disciplines, etc.
This makes sense when you think about the divided state of the USA today. People were isolated during COVID and had no feedback on opinions formed then. When they started creating bubbles after lockdown, they wanted validation for the opinions they formed without that feedback. Now, people constantly argue online because that feedback they needed while cementing those ideas wasn't there and we aren't willing to take that communication feedback to tweak our opinions based on common sense because we all believe we are right. This article makes a lot of sense.
Speaking out loud makes you translate your weird personal brain logic into structured instruction meant to inform others. And taking that other person's view, in order to explain what they don't know, gives you an alternative perspective.
I feel like trying to explain it to yourself as if you were ignorant of the problem may give similar insights.
Harvard's cs50 had a "Rubber Duck debugger" which was an LLM trained on all their educational content with specific guardrails in place to not give any answers. It would chat back with you planting little ideas to help push your understanding just a little bit.
I loved it so much, because I was learning programming completely by myself usually after midnight while taking care of a new-born. I had nobody to pair-program with, no lead dev nor teacher to speak with.
It was years ago, but such an amazing implementation of an LLM for education.
It was a tough time, indeed. But I feel that had I used an "all unknowing and answering" LLM, my retention and understanding of the class material would have been significantly lower. But I didn't have any choice as to having anyone to actually chat with back and forth. So it worked for me.
It started out as an interesting read, until I got to this paragraph:
> The value didn't come from what was said at that moment. It came from what had been built across many such moments: a pattern of mutual recognition, a shared context, a baseline of trust that made the later exchange possible. The relationship was the infrastructure. The conversation was where it had been built, one cup of coffee at a time.
And that made me immediately question the worth of the entire piece. I get it, LLMS can rewrite an entire blog post in a minute, which can be quite tempting for people that don't enjoy writing itself, but it just takes so much variety away. I think people should stick to grammatical corrections only, and not rephrase entire paragraphs (never mind letting an LLM write everything in the first place).
Similar: as soon as I see that LLM style I bail out, you can't trust the content.
If they can't be arsed to write it, why TF should I be arsed to read it.
If I smell AI writing, I ask my AI to condense it to a few bullet points
fwiw Pangram says 100% of the original article was written by AI
Pangram is a cool idea. It can be wildly wrong though.
seems the author removed that paragraph from the article after reading this comment
how can you tell it's llm that wrote this paragraph
Usually the smell is when it sounds profound or insightful but then you realize it's word salad and not really saying anything. LLMs also try to make the most mundane thing sound revolutionary or write about it in an alarmist way like 'increasing x by 1% is a ground breaking innovation' or something.
There's no concept of what truly is important with the writing itself since there's no actual thinking going on. The opinion of the supposed writer, which colors how they structure the writing and the language they use, is often all over the place since an LLM has no real opinions.
https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
[dead]
It's the cadence and the density of cliches. I've seen Opus produce very similar output. That's not to say it's 100% that this was written by an LLM. They had to learn this style from human writing in the training data. It's just a very condensed form of bad writing.
"Density of cliches" is a great way of putting it. It's exhausting to write that way and unpleasant to read back in the edit process which is why that kind of thing doesn't happen (often) in human-led pieces. LLMs don't get exhausted.
I think whenever a blog post begins to sound like a marketing campaign from versace - it's always a great sign that it may be LLM generated.
This isn't just a comment. It's a an experience. A shared cognition between parties -- To collaborate & exchange ideas. It's not procrastination, it's culture. Culture is what builds civilization.
fwiw Pangram says 100% of the original article was written by AI
This shit gets me irrationally mad. Well, the existence of such content doesn't, but the fact that other people aren't disdained by it (this dogshit has almost 300 upvotes on _HN_ right now) is what gets me.
The topic brings an interesting melange of people together. That makes up for it and made me upvote.
I too noticed a shallowing of the piece somewhere in the middle. It started with an interesting observation but ended with triviality (like you'd expect from a synthetic mind without a body).
As someone who doesn't follow the specifics of how LLMs write, how can you be sure it's an LLM?
I'm asking this genuinely because I've been using stuff like these separators before LLMS - now someone can just say "ahh dismiss the content they used style X it's LLM, it's dogshit".
I am not sure if I would manage to correctly identify every typical phrase, but the one I quoted I am absolutely certain of.
You could try asking an LLM to write some prose about whatever you fancy. Maybe two or three different pieces, and ask it to be extra profound. You will probably see that exact sentence structure above a number of times:
> It’s not [X], it’s [Y]. A bit of [A], the [random adjective] of [B], all weaving together to form [C]. A short sentence. Maybe two. A final, culminating closer, gradually transcending into space.
The words are quite interchangeable, but the you will know it when you see it.
Look for triads and "it's not this, but that". These are dead giveaways. You can also pass it through Pangram which is quite reliable these days.
That's the worst part, I'm never 100% sure, and that uncertainty is so ... disempowering?
fwiw Pangram says 100% of the original article was written by AI
lol Pangram says 100% of this was written by AI
I don't think the out loud or someone listening / reacting matters at all here. Suspect it's entirely this:
>The thought that was comfortable as a vague impression has to become a sentence, and sentences have structure.
It's not unlike what people like PG say about writing improving thinking...it's the being forced to go from fuzzy directional notions to something you can put on paper in that will stand up to critique.
Same with rubber duck debugging. The verbal part means you need to articulate it clearly but it's not the speaking that helps. Same with writing a detailed spec/prompt for an LLM - I know if its too fuzzy ("set an appropriate timeout") the LLM will spin it's wheels so it forces clarity.
Also suspect that a big part of who we consider intelligent is linked to this. Maybe their internal monologue is just more crisp - closer to what they'd tell a rubber duck.
My own efficiency ranking :
- genuine dialog between 2 peers with overlapping knowledge on the topic, both being kind, patient, open-minded yet critical, able to point flaws in reasoning or provenance without being hurtful
- genuine dialog between 2 peers with overlapping knowledge on the topic
- genuine dialog between 2 peers
- genuine dialog between 2 persons
- thinking out loud in front of someone listening
- thinking out loud alone thinking of someone listening
- thinking out loud alone
The "trick" here, and that's why I wanted to write it down, is that having an actual person, real or imaginary, does force one to actually explain. A typical fallacy is the "proof left out to the student" which actually happens to be terribly complex or even infeasible.
I believe trying to explain to someone forces you to identify the gap into their knowledge and your point. It's precisely mapping where knowledge does NOT overlap. So instead of making a lot of assumptions they have to be laid out, at the very least made in part explicit. Doing so forces you to stop skimming parts that are challenging for you. The part that you might not feel so sure.
Obviously the whole problem is that this peer, who is both knowledgeable and kind is good in their field and popular, thus busy. Consequently their time is very VERY precious. Hopefully you yourself are that peer. That means that often we have to go down the list and find the next best person.
So... I do think duck debugging is better if you genuinely trust in the kindness of the duck, not see it as a piece of plastic ;)
Different people’s minds work very differently. More differently than most people imagine.
I was reading the parent and thinking the same thing - I can't do that. There's no "voice" in my head like some people describe. If I want to model speaking like a mock conversation, I have to write it or ... speak it, even silently, to convert from the visuals that are my thought.
As an aside, I end up having to organize certain types of thoughts through speaking, which requires some adaptations and limiting functions :-)
Agreed. I have to write stuff down, with a pen, on paper, for this to work. Typing into a machine works sort of, but not on the same level.
I'm with you there. I have to either hear the full sentences narrated by my internal voice or see the words flashing in my mind in order to "think". This is great for building and maintaining deep mental models but it is also highly susceptible to "bit rot" (such as forgetting the rationale or evidence for a specific assertion or position) days/weeks/years later. I have a friend who simply can't understand how inter-linked note systems (like the kind Obsidian enables) are helpful. It's just a bewildering mess to them and they think more linearly.
Thus, writing things down is a necessity for me: it's not for a need for structure but rather that my "context window" gets filled too quickly. I can counter my own arguments but it's more fun, and often quicker, to do with someone else. Besides, there is such a diversity of thinking out there it would be foolish to not take advantage!
There has been many a time where the process of thinking through in my head how I was going to describe the problem to someone else on the way over to them, that a testable idea pops up.
Could probably have gone for a walk around the block instead, but the formulation of the problem to be able to explain it to someone else seems to be key.
This is what I miss from (the early) Stack Overflow: often the process of writing up a well-framed question ended up solving the problem.
I sometimes put out the question anyway and added an answer. Never quite figured out if this was considered bad style.
Jeff Attwood in his Coding Horror post about Rubber Duck Debugging mentioned that as well. I would think he wouldn't have considered it bad style.
https://blog.codinghorror.com/rubber-duck-problem-solving/
Same here. When I was younger I liked imagining having conversations with people from the past like Plato or something, thinking about their reactions to The Future. (Though I guess I only really imagined my own side of the conversation.)
I assume your anecdotal experience is rare.
> I often construct full sentences in my head
I've only ever heard that associated with schizophrenia, but I don't even know if that's true or not.
Haha, I found this genuinely funny. It’s called an internal monologue. Google tells me that it’s the majority position with 30 to 50 percent not having one.
Personally, I find it very difficult to understand how people could not think in words, like you were speaking to another person. Obviously you also have mental imagery and sound etc, so not everything is just words. Internal speech is one channel of thought, but for anything complex I would have thought it was mandatory.
> Obviously you also have mental imagery
I personally don't since I suffer from aphantasia. Perhaps I also suffer from not having an internal monologue?! I do think in words but not in complete sentences, it kind of happens much faster.
Look into "worded thinking", "unworded speech", "partially worded speech": https://hurlburt.faculty.unlv.edu/codebook.html
See also papers from Hurlburt and others on "unsymbolized thinking" and surrounding topics
"There is a depth of thought untouched by words, and deeper still a depth of formless feeling untouched by thought." - Rilke
I have done this for my entire life (no diagnosis) ... but I also find it easy to write stories and books.
Your comment made me pause in the "wait, other people don't do that?" way.
It might be correlated with neurodivergence. But it’s also just correlated with high verbal ability in general, presumably. I find it quite natural to think in complete structured sentences, and it’s often perplexed me why other people seem to find this concept so alien. And no I’m not schizophrenic.
Then again, different people’s minds work very differently. My verbal SAT was low 790’s, yet some of my best thinking is purely kinesthetic.
> And no I’m not schizophrenic.
Which of you? :)
I think I've got this from growing up with a narcissist. Thoughts are constructed and seemingly endlessly whittled to try and create a sentence that can avoid getting belittled.
It's hard to talk in groups, because you have to have a sentence mentally critiqued by 3/4 people in turn, so the topic has usually changed before you can say your piece.
What? It's never occurred to me that this isn't entirely normal, I've done it all my life. I thought people without an inner monologue were the unusual ones.
I'm not wading into the schizophrenia part, but inner monologue doesn't necessarily imply constructing a fully formed sentence you then repeat to yourself.
I think there is a difference between having an inner monologue, and imagining a conversation with another person in your head.
Created a throwaway for this, but is this not what people do?
Not so much imagining a conversation in your head, but playing through a conversation with "the other side" of the coin (be it an idea, plan, problem, solution etc) which are both yourself if that makes sense.
Is this not what the inner monologue to yourself is? Your inner is conversing with your outer monologue. Hard to explain.
Would consider myself to be both physically and mentally stable, no conditions etc.
> Created a throwaway for this, but is this not what people do?
I think it's okay, unless you converse online with your throwaways.
No, I've hacked other peoples account to do that.
I can only speak for myself... but my inner monologue is just a runaway train of thoughts that I vocalize in my head as I work through them. There is no other person in there, abstract or otherwise. Have you ever seen the Death Note anime? Similar to that.
I do what the person you're replying to and for me its just one voice or a faint imagination of them speaking. Its me roleplaying as the other person, not like there is a different entity in my brain.
For people who think in words maybe? I don’t have an internal monologue, so it’s quite foreign to me.
[dead]
I think there are tradeoffs though, and this has been a thorn in my side during technical interviews where you are expected to think out loud because:
1. Sometimes you have a vague sense of the shape of the solution, and ime it can be helpful to sit with it for a while before trying to shape it into words.
2. Talking out loud forces structure but it also rate-limits how quickly you can iterate through ideas to find one that plausibly solves the problem at hand
"When you are a Bear of Very Little Brain, and you Think of Things, you find sometimes that a Thing which seemed very Thingish inside you is quite different when it gets out into the open and has other people looking at it."
- The House at Pooh Corner
This was my first thought too. Thought, I will say that even though I think in speech for basically everything conceptual, and find it difficult to imagine even being able to not do this, I do often have these… ill defined ideas. Notions of something or other. I have a few words to describe them and they just sort of point in a direction rather than being fleshed out. It’s this sort of thing, like a thought-stub, that benefits the most from inspection and being forced into a better defined form. These are also often the more difficult things, and the reason they’re fuzzy is that it takes a lot of effort to make them sharp.
It seems like you saying that people who can do alone what most people do alone, have an advantage. Sure but seems more like an agreement with the article than otherwise.
I find this whole concept very interesting, because unless I'm translating it for external reception, my thinking is largely non-verbal - generally more spatial-visual, proprioceptive, or purely tacit than auditory, and I find words to be, at least by that standard, frustratingly limited.
I think that’s part of the point. Execution at its core is mastering the art of projection onto a lower dimensional space.
Taking anything from up in the mind out into the real world, be it a physical action, a speech, an artwork, requires you to make decisions about what’s load bearing and what’s not.
So the more practice you get translating your thoughts into words (or another medium), the better you get at executing.
The "another medium" is doing a lot there, I think the translation between my understanding and expressing it in words is the weakest aspect of software development at the moment, in addition to the entire physical world. Dancing about architecture, if you like.
From my own experience chatting with LLMs: reading the responses definitely does help with thinking, even when you can see obvious flaws or hallucinations. It gives you something to think about, which a rubber duck can't do.
The value of writing an AGENT.md might be more in the writing of project overview than in the subsequent ability to run an LLM on the project.
fwiw Pangram says 100% of the original article was written by AI
[dead]
One can't think aloud.
It's illogical to use one word when another is more concise and succinct like talking and speaking in this case.
Some words are self-explanatory like thinking.
The post you're replying to didn't say that, but regardless, I would say 'thinking aloud' is a pretty good descriptor for the word-by-word narration kind of thinking where you sort of both say and hear it in your head. I have heard that some people who who experience 'inner monologue' experience their thoughts nearly exclusively like this.
In 2017 LLMs weren't powerful enough to generate working code on their own, but my goal was to at least create a chatbot that could help you rubber-duck-debug your way to a solution. Unfortunately the tech wasn't quite strong enough for that, and not enough engineers even knew what rubber-duck-debugging was. RIP Duckly.
Trying to train an LLM on two 1080ti's on the StackOverflow corpus in my living room was a vibe though. Good times.
Duckly deserved to actually work. There’s a small irony here: the closest study I found to this, robots specifically built to simulate attentive listening, found they performed no better than an actual inanimate rubber duck for adult engineers. The mechanical signal of listening doesn’t seem to be the active ingredient. Makes me wonder if Duckly would have needed real disagreement to close a gap a duck can’t, not just better natural language.
You're probably on to something with the value of disagreement. I think it's one reason why chatting with current models doesn't create the same stimulation as rubber-ducking used to bring. The models are typically too quick to agree and amplify what you think rather than truly break it down and push back.
And thanks for saying it should have worked, I agree. My chagrin has increased over the years as I have realized the magnitude of my ill-timing.
Has anyone seen a good set of prompts for that disagreement? For the "skeptical eyebrow-raise" or "confused/doubtful head tilt" aspect of rubber ducks?
Agentic uses adversarial expert, steel-man opponent, risk-mitigation and failure-mode analysis. But what about almost brainstorming, but with thought-provoking nudge questions? Or on the other hand, arm-waving fight-club style discussion? Or... It's a big design space. I used to go to lots of research talks at MIT, in assorted departments. The post-talk Q&A question cultures varied a lot. Like encompassing both "leaves the speaker in tears", and "nudge so subtle, you won't quickly get it if you've not already spotted the fatal flaw in the work".
So aside from dialing down the "transformative insight!" silliness, there seems a rich multi-agent multi-persona space to explore.
I think agreement has value here too. An LLM that's starting to get a bit sycophantic will rephrase your ideas in a few different ways, and seeing the different presentations is helpful for reconsideration.
fwiw Pangram says 100% of the original article was written by AI
I wonder how much is actually needed to create an automated rubber duck. How well would ELIZA work? (https://en.wikipedia.org/wiki/ELIZA) (might need some adjustment to not talk like a therapist, but you get the idea)
2017 is a bit early to refer to them as LLMs. I'm not sure when exactly we started to refer to LMs as 'large', but I don't think it was before GPT2 (2019). That said, from the NLP work I've done, it was much more interesting working on small specialized models.
fwiw Pangram says 100% of the original article was written by AI
perhaps it is time to resurrect Duckly queue Frankenstein music and thunder in background
That would be interesting but the space is so choked now.
I'm also already busy building Forgetmenaut and enabling data deletion at scale: forgetmenaut.com
It's AIive!
AI ive, the thinnest neural network ever.
OMG - strong vibes to Einstein crediting Michele Besso, his colleague at the Swiss Patent Office, with helping him discussing some concepts in the special relativity paper: see at the end of the paper https://www.fourmilab.ch/etexts/einstein/specrel/specrel.pdf
Thanks,
I was wondering about the question of whether people who made very deep discoveries (Einstein and Godel come to mind) had others to talk things through with beforehand.
I know Andrew Wiles kept all of his work on Fermat's Last Theorem secret and by that I assume he never talked it through with anyone.
Interesting.
I saw a Facebook copypasta piece that claimed that Einstein's first wife came up with many or most of his ideas, and never got credit because of sexism. No proof whatsoever, other than she was a mathematician and physicist.
But "it could have happened!" is more important than even a microshred of evidence for highly emotional, online topics.
This anecdote nicely pokes a hole in that conspiracy theory: he was thoughtful enough to share credit with a layman work associate, but (supposedly) not the most important woman in his life - that seems even less likely.
fwiw Pangram says 100% of the original article was written by AI
[dead]
Tangentially, Regarding pair programming (as a special case of thinking together):
Programming is serializing ideas into the computer language. Communicating them with someone else first serializes them into human language, which is already much less abstract compared to the thought cloud in your head.
In the case of an effective pair programming collaboration, you also get to debate approaches, discuss details, alternate between coding and watching.
It also helps that the presence of someone else helps avoid many common distractions. Reading non-urgent private messages and checking out HN (I'm no longer so addicted to any other platform to check it out at work).
Yes, and, this is how experts should be using e.g. Claude Code.
It's simpler than this. Explaining a question/issue to someone involves going back to basics and covering all the foundational info that a third party would not have. When thinking alone, you gloss over this, and may not realize that a foundational assumption is incorrect. When you are forced to explain every step explicitly, these errors or gaps can become apparent without any intervention by the listener.
I have my younger kid explain each math problem to me before she submits it on Khan Academy. My older kid thinks in her head how she would explain a problem before turning in a test. It's a good habit to form.
As much as people make fun of "new math", it is really neat to see my 3rd grader working through a math problem step by step in a way that makes sense to them instead of the rote memorization I had to do as a kid. While they don't like showing their work, it helps them to work through each step to make sure their assumptions are correct, like you said.
Yes, it is nice that there are more ways that are taught now. It can turn into too much, however. Some parents I know dislike that their kid is taught 8 different ways to do long division, at the expense of having sufficient time to practice any of the methods. They end up taking the kids to math tutoring so that they can actually learn how to do the mechanical steps.
But it is for sure good that there are multiple ways to do various operations, and there are plenty of youtube videos that students can turn to if what they have been taught does not resonate for them.
In computer science, we have classes on memorizing/implementing various algorithms so that we may understand their behavior. Could anyone name the reason why grade school students memorize and execute the long division algorithm? I certainly hadn't used long division since grade school.
Grade school is years long, so a skill you're supposed to know in 3rd grade is one that you need to be able to faithfully execute for the next 3 years. The fact that you'll get a calculator in middle school doesn't help you if you're struggling through 4th and 5th grade.
In the nineties I was a junior copywriter at BMP DDB in London. The creative department was made up of loads of small offices, each with a two-person creative team working out loud all day. The best ideas rarely came from one person thinking alone. Someone would say something half-formed about a brief and their partner would catch it, often bouncing ideas back and forth. Reckon the whole building ran on that. Well, apart from the late and brilliant John Webster, who had his own office at the top of the corridor. We were lucky enough to have ours opposite his.
btw Pangram says 100% of the original article was written by AI
This proves nothing but, in my earlier days, I'd come home with something on my mind from work and tell my wife about how I couldn't get something to work the way I wanted. She had no clue what I was talking about but she'd offer up clues and suggestions to which I would try to explain to her how things actually worked.
Somewhere in that process it would lead to a solution that I would bring to work the next day!
Why are you making your wife work for free :(
Strange perspective. They are partners.
I started my web dev career in 1999 so my main code references were a combination of O’reilly and “for dummies” books. As a wet behind the ears engineer I’d find myself regularly walking over to my more senior friend Dan’s cubicle for help.
Half the time on the walk over, trying to frame the question in my mind I’d figure out the answer or at least next step. It got to the point where Dan would see me heading towards him and suddenly turn around and he’d as “Figure it out?” And I’d throw him a thumbs up on the way back to my desk.
[dead]
Communicating ideas helps, but thinking out loud may not work better to some people.
Thinking silently fits Asian Americans better than Euro Americans*.
https://www.psychologytoday.com/us/blog/sex-murder-and-the-m...
Thank you for sharing this. Growing up Asian American many teachers disciplined me and made assumptions about my intelligence for not being as vocal as the other kids. Culture shapes cognition and vice versa.
[flagged]
A collection of thoughts on this.
Pierces Firstness is exactly what drives this.
The move from thinking to semantic conversion is important for investigation/introspection.
Arguing with yourself also seems to engage your brains "theory of mind" centers, so different pathways get activated to examine the problem space.
The problem with Ai is the fact that it hallucinates and if you're doing anything truly novel in an integration or framing sense it bottoms out very quickly and can't engage. A human operator can decompose the problem and get accuracy checks for known areas in the training data of course.
Now to be I'm not saying Ai can't produce novel work on the edge but in my experience it is antagonistic towards those goals.
Case in point, CRDTs, many don't use tombstones but they are the minority, and if you try iterate a new CRDT off of one that doesn't use tombstones, let's say diamond-types, it will keep pulling you back to tombstones.
The problem is that the number of humans who understand dynamic investigation and the push pull of exploring an idea you don't hold with someone has always been very small, and now with reflexive internet argument culture driving how we view "debate" and "discussion".
I don't know if we've reduced the leisure to think or what but things are not great for finding speculative thinking partners.
[dead]
Great write up, thanks for posting. I think I could have avoided so many mistakes in my life if I had just spoken to someone about my plans first.
And not necessarily just based on their feedback, but just hearing the words come out of my mount and predicting their reactions would have helped a ton.
Heinrich von Kleist – On the gradual production of thoughts during speech
is a classic german text from 1805 on this subject that I have always valued deeply
https://franxfiction.com/on-the-gradual-fabrication-of-thoug...
Really interesting someone brings this up here. It's kind of an obscure little piece of text. But I always thought, yeah, this is more or less how we produce thought, and how LLMs operate, too.
This is so true!
1. Talking or writing requires thoughts to be sequenced so they come out in a way someone can follow
Thinking in your head won’t organize your thought.
2. Talking or writing to someone invites feedback and forces you to make sense, fit in socially etc.
Chatting with an AI or writing in your diary won’t refine or improve your thoughts.
Something funny i discover: i had thoughts or tried to explain something decades ago and made a drawing. Some of these are just a circle with some arrows. Turns out it is really hard to make a drawing bad enough that you dont know what it is. It sometimes works better than even a written description.
So you can just doodle in a notebook and tell yourself you've written it down. If you look at it regularly im sure recall will be even more perfect.
If you dont write anything down you wont be alle to access an orderly array of thoughs.
I draw a lot to figure out engineering details. I used to make beautiful drawings in diagramming software (draw.io and the lil). But man, it took a lot of time. Now I scribble on printer paper unless it’s something I plan to share.
Fellows devs at my company think I'm strange, but I always have a cheap coil-notebook. It has diagrams, notes, doodles, minutes, checklists, mostly not even in order. I try to remember to write the date at the top of each page.
I just counted, 11 notebooks (b5, 60 pages) in two years, so just about a new notebook every 2 months.
Funny thing is, coworkers will now ask to use my pen and paper to flesh-out something together, as I'm certain it's the only paper and writing utensil in the building.
And yeah, draw.io or even an ipad with a whiteboard app aren't good replacements for me.
I didnt have one for a while and accumulated some kind of panic when it was needed.
You remind me of a joke that with good employees you need only one pen and they can all write.
> even an ipad with a whiteboard app aren't good replacements for me.
Curious. My first thought would have been "... because the line is physically displaced from the pen, of course"; but that doesn't apply there.
I can't really explain it; it's simply a preference. It may not actually be optimal efficiency. But I like the pen and paper.
my remarkable comes close enough for my similar purposes
It takes me ages to draw my diagrams (I use excalidraw) but I find it so enjoyable because it forces me to slow down and check what I think I know
Interesting observation indeed!
You are compressing thoughts/ideas to symbols. Via a visual feedback loop your mind can operate on more complex ideas as your working memory doesn't need to hold everything at once. The term "second brain" really applies here. While you're "doodling forth" you become proficient in your on-the-fly invented symbol language.
Key is also the slowdown effect of being careful to draw properly. It's like a harness for your thought process. Fixating your thoughts on paper is a way to be safe not to go astray; like a climber who puts hooks in the wall.
It's really more complex than it seems from the outside. A picture of a group of people drawing lines in the sand comes to mind. And one wonders - once again - how language came into the world.
A different funny that should be archived here with the above.
Assuming you can type well.
Take a sentence, put your hands on the keyboard with the intend to type it but dont do it.
Some time later dont try to remember the sentence but just put the hands on the keys and let them type. Its really funny seeing the text come out of your keyboard buffer muscle memory.
Some japanese guy said he played with the concept and reported that he could fit a whole paragraph or a list of things big enough that he couldnt recall it without using his hands. He could do an invisible abacus to recall it.
You can't think of a sentence and type it later without actively remembering it. You either forgot the sentence by then or you're able to recall it - just typing from memory without actively remembering is just not a thing that is possible. Muscle memory isn't built from thinking one time about what you'd do.
One time early in the morning i couldnt remember the 6 digit accesscode for a warehouse. It seemed completely pointless to go there. Approaching the door i thought and now what?? When close enough my hand went up and typed the code. The door opened and i thought, wait, what was it? The memory was only available if my hand was on the keypad. It then seemed pointless to conciously remember it.
Does anyone really sit there thinking and make much progress? You write.
Sometimes you might sit and stare at the same text, grappling with it in silence.
For technical problems, I find the act of writing out a request for help, even just into a text editor, is often sufficient for me to solve the problem at hand. Writing things out is a way of organizing and structuring your thinking, and is itself a powerful troubleshooting tool. Things will become obvious that, unwritten, you might not even notice. I think this is very similar to thinking out loud; when you listen to what you say, or read what you have written, your mind is somehow keyed to react in a useful way.
That response might not be what you want or need for problems that need to be wrestled with, chewed, and pondered on deeply, though.
Writing is as good, or even better https://www.paulgraham.com/words.html
[dead]
You can tell you didn't read the article before jumping into commenting, as this point is addressed.
The string "writ" only occurs once in an article about how to develop ones ideas on non-trivial subjects. Pretty telling.
Only as
> the act of writing out a problem to a model still forces the same sentence-level precision described earlier
(model referring to LLM here)
but not as writing for writing's sake
I don't really buy any explanation here. The reason why thinking out loud with someone beats thinking alone is because when I'm alone I'll loop a lot more.
My reactions to thoughts that I myself have had are stereotyped and repetitive (every reaction is typical of me.) Reactions to me that come from another person are going to be different than those, and in turn stimulate me differently. Thinking tends to be an enumeration (then elimination) of possibilities, and that random input helps with the enumeration part by keeping me out of local minima.
I always found explaining the problem to one of my cats was a good way of getting to the solution, sadly I had to put the last one down on Monday and now my office feels empty and lifeless.
I'd wait a bit and get a kitten but having lost both cats inside a year frankly it hurts too much and they don't live long enough.
Sorry for your loss.
Furthermore, it helps to talk with people who think differently from you.
Maybe they studied the same subject but at a different school, or maybe they specialize in something else entirely.
Maybe their first language is different from yours, since language idioms can affect the way we frame problems.
Maybe they want to get into the field you're working on, and your thinking can also be teaching.
For me, this is a big part of the value of a hackerspace/makerspace. The tools are nice, but the intellectual environment is amazing.
Social engagement circuitry in your brain is in other parts than dedicated internal conversation logic thought streams.
By talking to someone else, you're simply mobilizing a different set of circuitry and thinking approaches.
It's like you switched llm models
I don't think this is particularly a complicated phenomenon
btw Pangram says 100% of the original article was written by AI
Turned this into a short explainer video if someone wants to see:
https://chatoctopus.com/share/94ed2beb-3878-4fbe-aeee-1f86a1...
I find it very useful to talk to my dog. When I have a problem I'm banging my head against, just explaining it to him from first principles makes my head "click". When that works, his dog treats are billed as "consulting".
What field is he a dogtor in?
Greenfield projects
Everyone inherently knows this, that's why they reach out to others when stuck on a bug (see also rubber-ducking). But why is it so hard to convince individuals and organizations of the benefits of pair programming?
Doing pair programming well is a skill you have to learn from someone.
Putting two regular programmers together in front of a computer will typically annoy both and not accomplish much.
I was fortunate to learn from experienced people, and I've had very good experiences with it.
rubber-ducking and pair programming have wildly different investment costs
Are you saying the reason people don't like pair programming is because it's considered more expensive than it's worth?
I wonder if there's a difference between people who have a strong inner voice and those with no inner voice (Anendophasia). I.e. do people w Anendophasia do better on their own than those with a strong inner voice?
> The Enigma of Reason (Hugo Mercier and Dan Sperber, 2017): their argumentative theory holds that reasoning evolved for social rather than individual epistemic purposes, to produce and evaluate arguments in group contexts.
Yes! I love that someone wrote this down!
This seems so obvious to me now. I often ask LLMs to cite their sources (they do hallucinate from time to time), and they often give me sources that don't say what is claimed. "How would the LLM know not to give this to me?" I wonder. They're trained to explain but not to convince, so they don't know what's convincing, and they should.
I think humans hallucinate at least as much as LLMs—arguments of any complexity are impossible to formulate without leaping at least a bit—but other humans ground us. That's why when people become socially isolated, they join cults or adopt conspiracy theories or the like.
Conversely, "this is convincing to an expert" converges on “this is true" as our collective expertise grows over time. This is the foundation of the scientific method, of progress in all engineering disciplines, etc.
[flagged]
This makes sense when you think about the divided state of the USA today. People were isolated during COVID and had no feedback on opinions formed then. When they started creating bubbles after lockdown, they wanted validation for the opinions they formed without that feedback. Now, people constantly argue online because that feedback they needed while cementing those ideas wasn't there and we aren't willing to take that communication feedback to tweak our opinions based on common sense because we all believe we are right. This article makes a lot of sense.
> Now, people constantly argue…
Ah yes, the newfound pastime of arguing online – sure miss those pre-COVID days of online's universal consensus…
Some of the value I get coding with LLMs is like that. Like I am working on some code that I 80% understand and going back and forth I get to 100%.
Externalizing a thought forces you to make it concrete. Most of the “magic” comes from that translation step.
Exhibit A: All episodes of Dr. House
> The thought that was comfortable as a vague impression has to become a sentence, and sentences have structure. They have a subject and a predicate.
I already think in sentences so idk what this guy is on about, sounds like a skill issue.
Speaking out loud makes you translate your weird personal brain logic into structured instruction meant to inform others. And taking that other person's view, in order to explain what they don't know, gives you an alternative perspective.
I feel like trying to explain it to yourself as if you were ignorant of the problem may give similar insights.
Thinking out loud with yourself helps the same way.
Wouldn't writing out loud do it too?
LLMs are really good rubber ducks.
using the audio recorder works as well.
wish I had discovered that trick sooner.
See also https://en.wikipedia.org/wiki/Rubber_duck_debugging
I keep a rubber duck on my desk and have for at least 10 years, because of this. Great conversation starter.
Harvard's cs50 had a "Rubber Duck debugger" which was an LLM trained on all their educational content with specific guardrails in place to not give any answers. It would chat back with you planting little ideas to help push your understanding just a little bit.
I loved it so much, because I was learning programming completely by myself usually after midnight while taking care of a new-born. I had nobody to pair-program with, no lead dev nor teacher to speak with.
It was years ago, but such an amazing implementation of an LLM for education.
That sounds horrible, honestly.
It was a tough time, indeed. But I feel that had I used an "all unknowing and answering" LLM, my retention and understanding of the class material would have been significantly lower. But I didn't have any choice as to having anyone to actually chat with back and forth. So it worked for me.
The Buddha famously thought out loud with a good buddy, while pursuing Enlightenment under the Bodhi tree
[dead]
[dead]
[dead]