I don't disagree, but I think it's often not appreciated how much there's other work to writing too.
The biggest one is that you have to communicate non-interactively to an unknown audience. Having to (literally) put it in someone else's assumed terms does help giving different perspectives into the matter, but doesn't necessarily help one's own thinking that much. Instead you have to do some of the reader's thinking for them.
You also have to spend time on textual matters like grammar and style and a lot of "unspoken rules", which aren't really about linearizing your thinking about the contents.
Not all writing is thinking and not all thinking is writing.
That is the best way I’ve seen anyone put into words what I feel about those types of theses
That said, abdicating to an LLM is the worst of all worlds - you’re not thinking and the product is not tailored.
The solution is obvious - write as much detail as you need and allow readers to interrogate the virtual you with an LLM, maybe not even reading what you write.
Hard disagree. Constraint is the driver of creativity. Also rewording sentences to sound better or make sense can make you reconceptualize the whole concept you are expressing
Take for example a non-native writer of the language. I'm sure having to check up words from a dictionary may help to reconceptualize things, but I'm sure also that it's not often very efficient. And I think similar is going on for natives too for many types of writing.
And not all thinking is writing is a clear truism, there’s no point to even stating that.
Writing helps us think about the world, it’s a pivotal intellectual technology.
Much like money decoupled selling and buying to move away from bartering, writing decoupled saying and hearing so they didn't have to happen at the same time. The incredible step that happened was not that people had to think a whole lot, it was that thinking that was already happening had to happen once.
> All writing is thinking when done by a human, you’re literally distilling your thoughts into words. You can’t write without thought.
Of course you can. You can write down exactly what you hear, for dictation.
You can write down a stream of consciousness and put barely any thought into it at all.
I can't help but feel most here are massively over estimating human writing. Human writing is, almost universally, terrible. We have entire jobs that are hard to fill just to make things sort of ok. Good writing is a small subset of human output.
Because one can copy a text and write it down and that involves thinking in the sense that anything we do involves thinking fundamentally. But that thinking is different from thinking logically about a concept and writing it down which I think is where you are getting at.
The definition of writing and thinking is too broad in that sentence even though it does apply in several obvious cateogires within that at different levels.
And also "writing helps us think about the world" is too broad again. Why? Why does me writing "apt apt apt apt apt apt apt" help me think about the world? I just wrote it because i felt like writing it. Why wouldn't you consider that writing?
Poor example. You wrote it to make a point, after all.
Do you have any non-contrived example of writing that was done with zero thought?
Have you never asked a decent model to explain something to you? You should try it.
Would love an example where you’re able to write without transferring your thoughts. Besides the obvious: fjcjfjrnjfjfifjfnrnakosifnrbwkofgjrj
(1) Translation from one language into another
(2) Transcription from one medium (audio) into another (text)
(3) Deception to obfuscate your thoughts
(4) Posting things like "First!", "This.", "Just google it.", etc.
(5) Textbook answers with no original thought.
Bad example, because translation is deeply creative. You can't blindly mill one language into another, because words and phrases and concepts and cultural references in one language frequently don't map 1:1. You have to find a way to convey meaning as closely as you can and not necessarily the words.
I prefer when things are kept 1:1 as is and maybe there's an explanation for things that don't quite make sense as a footnote.
Of course the above requires actual work on the part of the consumer. I realize many don't want that, particularly when it comes to entertainment. So I appreciate that the other sort of "translation" exists but I think it's important to realize what exactly those are.
Thankfully LLMs are more or less to the point of providing what I'm after in near real time.
Words refer to a broad semantic region, a phenomenon technically known as "polysemy".
The range of a word in one language is always different from the range of analogous words in another language. This is a classification problem. And a translator must think about how to solve it. Imagine a Venn diagram with 20 circles that each overlap the other 19 to differing degrees. What does it mean to designate one of those circles as "the literal translation" of a foreign word?
Obviously there are degrees to this and obviously preferences will vary. I acknowledged that.
Imagine a localization attempting to replace a reference to an actor, political scandal, or other concrete cultural reference from one country with the "equivalent" from another. I've encountered that sort of thing before and while there are certainly those who appreciate it I am emphatically not one of them. As far as I'm concerned that's shitty fan fiction.
There are also a lot of examples in most (all?) languages that rely on repetitive sounds, easily mistaken words, or other strictly auditory features of the native language. You literally cannot translate those things. I do not want shitty fan fiction, I want an explanatory note.
- You've got a pet peeve.
- You're going to rant about it, because you want to, whether or not it's relevant to an existing conversation.
- You didn't bother to think about my comments.
- You didn't bother to think about habinero's comment either.
Here is the same passage of the Analects (part of the chapter Gongye Chang) in different translations:
--- Annping Chin ---
Zilu said, "We would like to hear what you would like to see yourself accomplish."
The Master said, "To give comfort to the old, to have the trust of my friends, and to have the young seeking to be near me."
--- David Hinton ---
Adept Lu then said: "No Master, we'd like to hear your greatest ambition."
"To comfort the old, to trust my friends, and to cherish the young."
---
Our focus here is on the second line, what Confucius says. Does he want to trust his friends, or does he want his friends to trust him?
Does he want to cherish the young, or does he want them to cherish him?
We might also ask, though the translators have agreed on this point, whether he wants to comfort the elderly or for the elderly to comfort him. (And we could further ask whether Confucius wants to personally comfort the elderly, or whether what he has in mind is for society in general to do that.)
All three clauses are formed the same way in the original Classical Chinese, and for a couple of interacting technical reasons they are all ambiguous in this way. Translators, as you can see, make different choices.
But of relevance here, when you're doing a translation to English, you have no option but to make a choice. It isn't possible to render the original text 'in literal translation' and append a note explaining what went wrong. You must commit to a meaning behind the text and phrase that meaning in English. You can also append a note explaining that you might have chosen wrong, but English simply doesn't allow you to do anything that parallels the source material.
I think it should be quite clear by now that I am not talking about isolated words that broadly lack an equivalent concept in the target language. I even quoted the bit from the original comment that I took issue with and proceeded to give examples so I'm really not sure where the misunderstanding between us could lie at this point. Perhaps you are the one who should stop and more carefully think about what I wrote?
As to your example. I certainly do not accept that this is a case where we should throw our hands up and accept that different translators will go about things differently. Those two sentences in english have (as you note) rather different meanings. So either one or both translators must be wrong.
You have indicated that the original work in the native language is ambiguous. In such a case I do not think it is remotely acceptable for a translator to arbitrarily pick one of several possible meanings and just run with it. If the original meaning of the text is ambiguous then removing that ambiguity changes the meaning thus it is a bad translation. The translator instead needs to faithfully communicate that ambiguity, possibly resorting to a note if it isn't possible to easily express such a thing in the target language.
I realize that many people aren't going to want such a marked up copy. But without all the gory detail the reader will be consuming some sort of bizarre partial fan fiction. Your example illustrates that perfectly.
https://www.theguardian.com/uk-news/2020/aug/26/shock-an-aw-...
QED?
Not all writing is thinking: That all, even a lot, of writing, or parts of writing, is such that it will develop one's thinking much. For example most stuff I have to write, the dozen emails a day, the funding application boilerplates, the reports are stuff that don't really need (or deserve) much thinking but they have to get written. And even in the writing that deserves attention, there is stuff like grammar and spelling and surface style that usually take quite a bit of time after the ideas have been written down already.
Not all thinking is writing: for many cases writing is not a particularly efficient way to develop one's thinking, and e.g. visualizations, math, coding, discussions etc can be a lot better.
They’re completely opposed to experiencing any type of friction.
It reminds me of the transition over the last year from AI-assisted coding to AI doing all the coding. At first the code output wasn't good enough, and humans read and iterated on the code all day, so the details of the source code mattered. Now, the code is largely high quality and it meets a large set of guardrails we've set up over the years (linters, typecheckers, security checks, LLM-assisted code quality checkers), and it's just Claude working on the code, so the details matter less and engineers think a level or two up (machine code < assembly/bytecode < source code < conversation with agent < artifact with high level design).
I wonder if long form writing will go the way of code. You and the AI agree on an outline or other high level representation, then the LLM expands it into a document. But writing and coding are different enough in a number of ways that this is far from inevitable.
What use is that? I'm not being facetious, I'd really rather like to know.
Who or what is the audience for that sort of long form writing? If it's a human, why would they read it? They'd just give it to an LLM and get the salient points back. If the audience is another LLM, why expand it?
The only use case is an audience of humans who still read and understand, and those people aren't going to be interested in a message when it is not apparent that the sender actually understands the message themselves.
That's why there's so many meetings in white collar companies. Because people can't understand what is going on at those documents so they just need to "align".
LLMs are amazing at generating this useless documentation that goes absolutely nowhere.
That's already available today. We don't have to perfect LLM writing.
1. I have a bunch of data or research that I've gathered with a unique hypothesis
2. Having gotten my arms around that pile of information, I believe I have a compelling thesis to put forth
3. I design the narrative arc and of the thesis. The important parts, the necessary but not sufficient scaffolding.
4. An AI helps fill in the story from there. Fact checks each claim, connects the dots, makes it comprehensible.
Who is this for? Well, quite possibly the human who asked for it. It's pretty informative to read back a research brief in full that you helped do the scaffolding.
Also of very clear use is other AI's who did not have the same unique hypothesis and did not gather the supporting evidence. It's an interesting angle for others to build on.
And of course, other humans! Most human written content gets almost zero readers today as it is. And I suppose LLM content probably pulls the asymptote closer to zero, but some pieces of content may be genuinely interesting or useful.
I think this certainly has some value but this claim in and of itself is stated like your hand-wavy step 3. How do they fact check claims and connect the dots?
Maybe LLMs get there but currently they write in an extremely verbose manner, and things that have gotten into the context window that are no longer relevant continue to stick around (just try having it write some code, then work some of it back to simplify the problem - it will insist on writing comments about code that no longer exists).
Right now using an LLM to write documents is like taking a superhighway to travel 100 meters. Yeah you're doing a lot but is all that really necessary?
I won't deny that LLMs will never have a place in writing. But I personally don't think the current form is "the one that actually lands" (!).
I used to say this was the future of advertising (cr sales person prompts “we have some new EV SUVs on the lot”; GPT generates an ad email with a synthetic video, blinking text etc; then the recipient’s spam processor tells them “that dealer has some new SUVs”. I suppose the same could happen with so-called “long form”.
People are terrible at writing. Near universally bad. Even good writers have drafts and editors.
There is a constant refrain here that somehow short messages are more valuable than longer ones. But that assumes it's understandable. Lots of short content is, frankly, awful because the writer cannot put themselves in the position of the reader and explain all the things around the point they're making that the reader really should be told.
You can view writing as translation. From your language to a language your audience speaks. At that level is it so odd if the word count differs from one side to the other?
Look at it the other way, could you take a good longer message you’ve written and make it shorter and less readable for your audience while still making sense to you and containing the key points?
personally, I think there's a time and place for short versus long, just like there's a time and place for a 45mins TV episode versus a 2 hour marathon movie.
My opinion of LLM design review isn't that high - it seems to miss design tweaks that could vastly simplify corner cases. But if your code isn't written for human consumption maybe it doesn't matter. I'm still directly responsible for what I commit, so I can't just offload it to Claude.
It's like watching somebody about to be hit by a bus. You yell, you wave your arms, but they either don't hear you, or they don't believe you. The last thing that goes through their head is a Greyhound's hood ornament.
The most popular programming languages in 2030 will, in fact, be English and Mandarin. Deal with it and get over it.
----------------
Edit, to bcrosby95: Look up the etymology of the word 'computer'. It didn't originally have anything to do with hardware. The first computers were people, who were told what to do ("programmed") without necessarily knowing what they were working on in a big-picture sense.
Calling it gatekeeping is just laughable. That's like saying it's gatekeeping to say that the painter painted their painting, and that the person who commissioned the painting did not paint it. It's wholely absurd.
Anybody can pick up a book and learn to actually code themselves. Or you can use an LLM to try to make things without bothering with that. But even if the LLM worked perfectly, pretending these are the same thing is silly.
The history of the word computer is obviously irrelevant. Words change, it turns out.
(I’ve heard about some GPU compute nuance meaning that even without randomness injected they still wouldn’t quite be deterministic, but that’s also not core to their nature)
Neither do humans.
Two people can give the exact same prompt to the exact same LLM and get different results.
No one cares.
This is like telling someone else to code something for you.
Exactly.
What are you willing to bet?
To be precise: I will bet that high-level programming languages won't be any less popular as a whole, but the vast majority of code will be written by AI rather than humans, working from specs written in natural language or something very close to it.
What we call "source code" today will be thought of as "object code" by 2030. Something that occasionally needs to be inspected by humans, but rarely authored directly. Anyone not writing code this way had better be doing it as a hobby, because almost no one will pay for it.
Other person probably doesn’t like the idea that LLMs will replace hard earned skills. On the flip side, I bet you’ve seen your skills atrophy at an alarming rate and are trying to justify it.
Both sides come from fear. Just relax and take things as they come. Whatever happens happens.
It's baffling you people are in control of such a strong product when you are obsessed with this intellectual pornography; wow - look at how smart it made my thoughts look (n.b. look, not read). Don't look too close. And certainly don't ask me what it means.
The widespread introduction of LLM code generation is very destructive to that.
Perhaps LLMs can be brought to support human cognition in the same way writing can; but that has yet to be designed and it does not seem to be the way things are heading.
The bottom line is, prompting is definitely writing.
Cognitive burden increases marginally with AI assisted coding.
This is why we haven't seen big projects(think browsers and browser engines) spawning in the past year.
Why are you reviewing AI code in detail? Do you also review the assembly output of GCC line by line?
In the case of writing, it’s like hiring someone to write a book for you vs. hiring someone to translate a book you wrote into another language. In the first case, you didn’t really define the message for readers, whereas in the second case you did, and the translator is converting that same message for another audience to consume.
In the case of LLMs, the behavior is non-deterministic and inconsistent. If I don’t explain how handle an edge case or give a performance constraint, the LLM will still produce code and may do so in different ways, handling edge cases differently and with different performance characteristics. I can’t reason about how the LLM will fill in those gaps, it’s “random.”
Maybe you don’t care about how the LLM handles those edge cases or handles performance, but that’s different than a deterministic abstraction whose implementation details you don’t care about, but whose logic and performance is deterministic and consistent
The closest thing we have to vetting LLMs is “whoa look, it escaped this sandbox, that’s prolly not great but it’s so cool!”
I wouldn't use it for flight control software yet, at least not without careful review, but most software isn't exactly critical. At the same time, I wouldn't trust flight control software that was only reviewed by humans, since AI is so much better at debugging.
We'll probably need humans in the loop for safety critical software for at least a year or two, before AI fully outpaces humans at generating correct code.
So I assume you don’t fly? Or is it only software created after 2025 which must be reviewed by the All Knowing Entity?
And, AI is rapidly getting better than people at both code review and authorship, so a human deeply involved is turning into nothing but a slowdown. The main purpose people have is testing that the specs were, in fact, implemented properly.
The vast majority of properly written software was already plumbing well over a decade ago. The software engineering is making high level decisions based on experience with respect to the existing tools and the needs of the business. If you're not already using LLMs that way, you would have been a similarly bad manager of human devs writing similar inadequate slop. Less code has always been better code.
The line in the sand for these arguments really ought to be whether you think LLMs are better than humans who actually know what they're doing.
If you think LLMs are better, or could get better while continuing to use statistical methods, you automatically lose the argument (delusional/ignorant) and any hope of regaining credibility. That's not dogma. That's the science.
twitch
Gcc makes maybe 1 mistake ever 2 billion emissions. LLMs make 1 mistake ever 3rd emission.
What a horrible, cold, inhumane world that would be.
If you're not billed for usage, anyway.
Otherwise, for the other 99% of folks, that attitude is of course a pit trap that captures code bases and makes them maintainable only through the providers -- presumably one or few -- with a rich enough model to keep up with the growing mess. Preserving a code base that's legible, organized, and fundamentally maintainable by both humans and trailing commodity models is of imminent concern for anybody who doesn't want their margin strangled by your employer once it's too late to have other options.
As frontier capabilities advance, the details don't matter less; they matter more.
Prices are very competitive and today's SOTA is next to free in half a year.
Whether code is maintainable without AI becomes less and less important.
I couldn’t help myself, replied and asked him for a recipe for delicious apple cobbler and hiking trail recommendations in Glasgow, which “he” immediately provided. Highlight of my career.
I think my core argument is this: I have access to every bit of information your AI does, so if I want an AI answer I’ll get one myself. If that isn’t true, why are you hoarding information? Push it somewhere we can all see it. So the only reason I would send you a message is to access _your_ brain. I have no interest in talking to an AI through a worse interface.
1. I understand fully the code and everything it does 2. You can pick up on mistakes super early and it can adjust the plan is it goes. 3. Faster than writing it by hand but slower than letting the LLM do it.
[0] https://ankursethi.com/blog/prevent-cognitive-debt-by-manual...
I worry about AI Loopidity here though. Think about the similar analogy of email. If my set of ideas is condensable to bullet points, but I use AI to expand the content, then I add no information density and a lot of noise. Other folks then use AI to summarize the content to a list of bullet points, ideally the same but not certainly the same, and thus communication has been only partially successful.
You get pushback for this? I saw an anthropic job post recently, and they wanted you specifically to have claude muck with your resume before applying.
The entire point is what runtime you’re running your code on. A computer with any modern stack requires a lot of text for you to communicate “spin a square around on its center” to it. A human requires only that short string because they have a faster natural language interpreter.
Text meant for a human can communicate “spin a square around its center” much better than any code that mimics it. In some sense, all programming is boilerplate expansion because computers have (until now) been unable to be programmed with anything approaching natural language.
Maybe sometimes, but not always. When you need to actually render the thing you have all kinds of micro decisions, like where to put the square, what color, how fast it spins, etc.
You might not care about the details, but maybe you do. If it spins at 10000 rpm, will you care then?
Natural language, and human communication in general, is ambiguous, and coding is in great part about disambiguation.
Sure, you can use English to disambiguate as much as needed, but wouldn’t you then end up with some yaml-like spec that wasn’t much easier to create in the first place?
You can already do this. And you can build pipelines where AI performs fact checks on what it writes, with citations a human can reference as well.
On the other hand, long form writing for human consumption seems like it may evade LLMs for much, much longer.
So, I'm not sure if it's a question of time at all: if a LLM text contains some piece of information beyond the information that went into the prompt, where does this "extra" information come from? [Note, I'm not thinking about facts which could trivially come from the training corpus, I'm thinking specifically as information in the sense of intended message from sender (author) to receiver (reader)]
[1] cf. this comment where I explain this analogy between LLMs and noise channel in communication theory: https://news.ycombinator.com/item?id=49510244
All of the things you say are very true in the near term for short form writing - a page or two of Claudeslop will probably be much easier to swallow in a year or two than it is now. But I don’t see a path to fully AI-generated novels or long-form investigative journalism becoming mainstream in the next couple of years.
Really huffing your own farts there, huh?
Later
> I wonder if long form writing will go the way of code. You and the AI agree on an outline or other high level representation, then the LLM expands it into a document.
So in the future, it won't be necessary for you to think?
This is just noise generation. If anyone is meant to actually read the document it should be written by you.
No. No it is not. Nobody who actually cares about the quality of their work is letting an LLM just turn out code without reviewing it carefully.
Then again, I've seen a counterargument [1] by someone who clearly heavily uses LLMs for writing (going by both their LLMy writing style and their own admission). The person I'm citing describes a process where they get a LLM to write something, they check over it and provide feedback to the LLM, the LLM rewrites, and the process repeats iteratively. So clearly he is putting thought into the process.
I think there is something valuable missing, even if it's hard to clearly express. I'll try. The threshold for what I'm willing to accept if I'm simply approving something is likely different from what I'll get if I write something myself, for instance. Saying "LGTM" is too tempting. It seems to me like he's outsourcing his selection of topics to cover as well. If you're not thinking yourself about what to cover then it would be very easy to miss a critical subject. There also an asymmetry between checking and generating something with constraints placed on it. Checks can't catch everything, and a constrained generating process can reduce the amount that needs to be checked, avoid issues that can't be checked so easily, and focus your attention on areas that you know historically have had issues with this generating process. I've thought about this quite a bit in terms of whether to write new code or use an existing library. Sometimes "the devil you know" (my code) is better than an existing library simply because I understand its flaws better.
Just did this for some caching, started with the structure of what I wanted the cache to look like, asked the agent to start the work, didn't like how the architecture came about, scratched it and rewrote the whole thing, so it fit the model I now wanted.
Was also just having this discussion with friends, that I can only think seriously about a subject if i can put it to paper (even virtual paper). Writing lets me organize my thoughts, clearly define my assumptions and see if any of it make any sense. I can't imagine what it would be like if i couldn't write, my brain just doesn't work without it.
The only thing that worked was a standing instruction and periodic system reminder injected from the harness:
"If any assumption doesn't hold, if there's a fork in the road, any architectural decision needs be made, STOP and report back to the user. Do not try to push through the problem."
This has worked remarkably well for me. Now I have to think a whole lot more. It's much slower, yes, but I don't really see any other way that doesn't end up in garbage.
I don't, generally, think in words, more in - I guess I would call it something like meta-shapes? A sense of a shape but not things I can exactly visualise.
(You might be surprised to read this and then hear I have an English degree. Surely I thought about Shakespeare in words?! Nope. Shapes, movement, structures)
For me, having to write is critical because it is the only way I practice serialising my thoughts in a way other people can understand.
If I do not then I get very "deep" into my own way of sensing ideas and it's difficult to dig myself back out.
This might also be why I have never been very enchanted by LLMs? They only seem to "think" verbally. So it is always a translation effort for me.
I never can really enter any "flow" state with an LLM. My intuition is that highly verbal thinkers can enter flow with LLMs very easily
Don't worry, no one does.
That's why it is so common for people to forget a specific word they want to use ("it's on the tip of my tongue").
If we thought in words that will never happen.
>> Don't worry, no one does.
> …you don’t think in words?
"I don't, generally, think in words" is not the same as "I don't think in words".
Like I already said, if thought was exclusively in words for humans, humans wouldn't have the "It's on the tip of my tongue" problem. It's blindingly obvious that thought does not occur exclusively with words.
I think I have produced reasonably good designs. Don't ask me to teach anyone how I do it, though.
One of the most rewarding things for me is figuring out a good shape for a system and how it would interoperate with the other systems, especially in a way that reframes other parts of the codebase in a way that bring clarity and makes it more intuitive to work with. Creating the right ontologies can make all the difference in what you can do with a project. It's a form of creating mathematical objects.
For example, a Unity game I work on has quest and dialog systems driven by visual scripting graphs. We had two way dialog with different units for player response choices and npc dialog. But we wanted to expand to letting NPC's have dialog with each other as well as conversations with more than two participants. I went outside and thought it over, which largely amounted to visualizing a dialog node graph and a feeling in the back of my mind like it was trying to perform a kind of geometric shape-fitting exercise. A fitment solution jumped out at me to have only one "Dialog" node shared by all participants, with a "participant" value on it. If the player parses this node then the options go on-screen as responses, while if an NPC parses this node with multiple options in it, it picks one. And this lets you voice the player if you want, and enables some things like overhearing other NPC's talk to an NPC then talking to that NPC yourself and having the same tree.
And for quests, the quests had just been for the player, but I was thinking about how to make scripted events in-game easiest to work with for script team who primarily works in visual scripting. Similar story - let the NPC's have their own little quests, with task stages, which are easy to track and make branching choices from, and let the NPC's definition for how to use that quest contain a collection of actions to override the typical actions available to it, so an NPC in a specific "quest" can't do things you don't want it to do, a common enough case that it's preferable to making a series of conditions on the general action planner like "not in quest A"
And timing myself, it took 1-2 hours each time to write out the detailed plan for how I wanted each thing implemented in the game with some other tasks thrown in, and it paid off after Astra worked on it until it was done. It was awesome coming back to something pretty much exactly what I asked for each time.
Things I need to do, ideas I want to ponder on, people I need to remember or get respond to.
Writing is learning. Writing is understanding.
I have so many conversations with people who are always telling me I’m “retro” or “old school” for doing this.
I don’t even bother explaining the psychology behind it anymore. I’ve got no time for the ignorance.
"If you’re thinking without writing, you only think you’re thinking."
> As I write, I think about things. As I write, I arrange my thoughts. And rewriting and revising takes my thinking down even deeper paths.
- Murakami, "What I Talk About When I Talk About Running"
This way of phrasing it is needlessly confusing. Writing is a kind of thinking--one of many--but it's not equivalent to thinking.
Dialectical thinking, for example, produces similar results.
The best thinkers I know mostly use writing for refinement, distillation— as a tool. The worst thinkers I know are owned by writing; they require its fixation & stimulation upfront to compensate for limited attention spans.
Really? I am having difficulty thinking of any examples of code that doesn't need to be understood. If it isn't understood by someone, then how is it even working?
If you mean like a library you are using, where you aren't even reading the internals or might not even have access to it, OK, but that code is stull understood by its authors, surely?
They're relatively simple, they do the task they need to and then they wait until they're needed again (or not).
In the past I wrote them, then forgot how they worked, until I needed them again, relearned what I did and adapted it.
Now I just don't have to know how exactly they work, I just get an AI to read the documentation anytime I need to reuse the project and I'll query the AI to fill in the details and to make changes and I ask the AI to run the code and debug it.
Perfect use cases for today's AIs. Doesn't even require SOTA, I can run comfortably on a Sonnet 5 or a Qwen 3.8 and it'll do exactly what it needs to do without making too many mistakes.
Not all code is large corporate code bases.
You can also do this for apps that are just tools for your own use. You satisfy yourself that they are working, and you use them because they save your time. You review enough to be sure its implemented the way you think it is - and if it is working, that tells you quite a lot. Sometimes you will be surprised and have some time wasted.
Yes, yes - there are people who will make the wrong choices in some of these cases but that doesn't mean there are never cases where you can do it.
More broadly - anyone who works in a team is already working with code they don't fully understand. I have code I wrote years ago I don't fully understand. I trust its observable properties and its track record.
I'm not following.. When we write regression tests those tests encode invariants we expect to be maintained under source code transformations over time. If I don't understand the test code I've written, how can I know which invariants I've imposed? That's why, broadly speaking, we write test code to be as simple as possible above all else--it's absolutely imperative that these invariants are not only intentional and easy to reason about, but also that when an invariant is violated we can easily discover why. Often, on a team, the person encountering a test failure after making a code change is not the person who originally established the invariant, so it's very important they be able to easily understand it.
I see no possible world in which failing to understand the test code is... possible? Like, if you have indecipherable test code things are really bad in your codebase. Fixing that is P0, because it'll compound rapidly.
Having your fly open is a harmless mistake that has little impact on your peers. People may or may not mention it to you but it’s not something they’ll hold against you.
Posting LLM slop under your name is a deliberate act. You decide to damage your message by taking a shortcut.
The obvious analogy was speaking while chewing.
They are powerful tools but they do not have any human understanding - that isn't their optimization target.
A) we only know of one species capable of metacognitive understanding,
B) we already tried that in the 1970s, and it was good work but often evolved into what we'd call boring ol' computing rather than AI, and
C) an alien mind wouldn't be a very good agent, for a ton of reasons relating to affect, conversational rythyms, cultural understanding, etc.
The trick is to make something that acts like a human but with the affordances of a computer (e.g. scalibility, symbolic certainty), without making it so human that it takes issue with its existential reality and/or use of its labor...
That is it. We cannot concieve it because we are new to it. Just like we would think of Stackoverflow as intelligent if we are fresh off the jungle and are not aware of how Internet works. Because without know that, we cannot conceive how Stackoverflow can produce answers without it "understanding"
To you, you type your questions, and answers appear. That would look like how LLMs appear to us now.
Which is exactly what happens with human evolution and development. Sure, we can say LLMs don’t have “human” understanding - which is something we can’t really define anyway - as long as we’re not trying to claim LLMs don’t have understanding at all. The latter is a much higher bar.
> We define goals that we cannot conceive of reaching without something like understanding happening.
Functionally speaking, that is understanding. Again if you want to go past a functional definition, that’s a bar which no one can clear right now.
I think AI models do have something like understanding - I think Leela understands chess and I think Claude understands code in some very real sense, though not a human sense.
But for general writing, you have to understand the world at large and there is no sufficient RL for that. Do you really not see the constant errors that AI make that betrays a lack of understanding the world? I see them so constantly I rarely think about them, I just skim over that slop and move on.
Sure, the exact nature of the understanding that an LLM exhibits is different from a human's. The differences in the training data we're each exposed to can explain a great deal of that, and of course there are architectural differences etc. as well.
But the specific quote I responded to was "We reward the appearance of understanding." My point is that's no different from humans: evolution and a child's upbringing rewards the appearance of understanding. The result is imperfect, e.g. people end up with an understanding of the world that in some cases is completely nonsensical (all religions except the one true religion, mine, are false!), but it's sufficient for them to survive.
This demonstrates that "appearance of understanding" is not a meaningful distinction between LLMs and humans. The meaningful distinction is in the training data and the specifics of the reward functions.
Many people seem to try to make a kind of "no true Scotsman" claim about understanding, that somehow LLMs "don't have real understanding". Based on the above quote, it seemed like you might be making that kind of argument. The counter to that argument is simple: if LLMs don't have real understanding, then neither do humans, because broadly speaking, both operate on similar principles: we learn from training data, there are reward (and punishment!) functions that influence what we learn, and the result is a "mind" that demonstrates an understanding of the world.
Not everyone accepts a simulationist view in which modeling something accurately enough inherently results in creating the actual thing.
Likewise using AI can be thoughtless, but it doesn't have to be. I don't see why a valid creation process can't be like this Simpson's meme[1], where you start with a rough object and then cut away and refine until it's done. I don't see it as lacking merit or requiring less thinking compared to starting from a blank canvas and adding more until it's done.
And either way at the end of the day the writing artifact stands on its own. It's either good or bad, taste permitting, and can be evaluated for what it is.
[1] https://media.licdn.com/dms/image/v2/D4D22AQFoqRgMxteTNg/fee...
Understanding is the bottleneck; the way they speed things up is by letting me outsource understanding, and get back a summary. The entire advantage to AI is that it lets me skip understanding the problem, and just get a working solution.