3 ms·

...I find it a bit surprising that Paul Graham of all people, writing in May of 2025, managed to get through this entire essay without mentioning LLMs. Because

...I find it a bit surprising that Paul Graham of all people, writing in May of 2025, managed to get through this entire essay without mentioning LLMs. Because I think LLMs provide a clear counterexample to his thesis. They are quite good at the craft of writing--not perfect, but probably much better than the median human--and they are just as good when the content is true as when it's false. This quality ruins a lot of my heuristics for evaluating whether writing is trustworthy, because LLMs are so good at bullshitting. So while I agree that for humans, writing that sounds good tends to also be logically correct, that clearly isn't inherent in all writing.

pvg 1y ago
Most people don't want to spend much time reading 'median human' writing so the claim that LLMs are 'better', even if true, doesn't really say that much. We don't listen to median human music either.
- Wowfunhappy 1y ago
  > We don't listen to median human music either. Don't we? What about the music that plays in the elevator, or while you're on hold on the phone? Similarly for writing, I would imagine you read plenty of emails that are of more-or-less median writing quality. And yet, these emails may discuss pivotal decisions, where it is very important whether their arguments are logically correct.
  - pvg 1y ago
    Get a median human to write some elevator or hold music and see how that goes. There's an entire business of making and licensing such music which would not be the case if any rando could just crank it out. Emails, etc, are just 'communication but over text' - it's 'writing' in the very basic sense but it's not the sort of writing people concerned with style and quality care about, either as consumers or producers. Me talking to the cashier at the store is not 'public speaking', neither I nor the cashier care it's not the Gettysburgh Address.
    - the_af 1y ago
      I think emails summarizing meetings, or making a case for something, or describing something, match the kind of writing PG means in his essay. And like the commenter you're replying to, I think that LLMs today can write far better than your average coworker (assuming they don't go on a hallucinated tangent; then again some coworkers also do!). So it's fair to compare LLMs and work emails. Not to speak of work wiki pages (Confluence etc) describing technical things, decisions, policies, etc.
      - pvg 1y ago
        Why do you think pg is writing about emails and meeting notes? I don't readily see anything in the piece to suggest that. 'making a case for something' or 'describing something' covers the bulk of writing, the piece is quite explicit about not being about all writing.
        Wowfunhappy 1y ago
        Why do you think he's not? It seems to me there is no difference in kind between an email arguing for or examining an important decision or idea, and an "essay" on an important decision or idea. I definitely don't think these documents are just "written speech". Some emails are—a quick message asking if you'll be in the office tomorrow, for example—but major ones require significantly more thought.
        pvg 1y ago
        It seems to me there is no difference in kind between an email arguing for or examining an important decision or idea, and an "essay" on an important decision or idea. Why are there people who make a living as, say, newspaper opinion columnists or magazine staff writers? Or hell, why aren't we all rich substackers given that writing a substack is literally email?
pdonis 1y ago
> LLMs provide a clear counterexample to his thesis. No, they don't. His thesis is not that writing that looks good--that seems plausible and convincing, like the output that LLMs often produce--actually is. He says explicitly in the article: "we can't safely conclude that beautiful writing is true". His actual thesis is in the very next clause of that sentence: "it's usually safe to conclude the converse: something that seems clumsily written will usually have gotten the ideas wrong too."
- the_af 1y ago
  To be fair, most people commenting on it here on HN seem to understand the first meaning (looks good => is good). I can guess what this says about the readership of HN, but what does it says about the clarity of PG's essay?
  - pdonis 1y ago
    > what does it says about the clarity of PG's essay? This seems pretty clear to me: "[W]hile we can't safely conclude that beautiful writing is true, it's usually safe to conclude the converse: something that seems clumsily written will usually have gotten the ideas wrong too." And I'm not the only one who has quoted that passage in this discussion.
    - the_af 1y ago
      It's the second to last paragraph, when someone reading the essay may have already formed a hard to dislodge opinion on the kind of writing PG means. And indeed, many people here on HN did so. The order of paragraphs, how early in the text ideas or clarifications are introduced, etc, all make for clear writing. I'd say this is an example of an essay that lacks clarity, i.e. one that is not "good writing".
- aspenmayer 1y ago
  > His actual thesis is in the very next clause of that sentence: "it's usually safe to conclude the converse: something that seems clumsily written will usually have gotten the ideas wrong too." This analysis in the article feels woefully uncharitable and is incompatible with HN guidelines, ironically, but if internalized leads directly to the nitpicking style frequently seen on HN as a substitute for elucidation of ideas not well expressed on the way to deeper analysis. Dismissing mediocre writing out of hand seems classist more than anything.
envirogis 1y ago
I had the same reaction. I spend an increasingly large part of my day reading and being required to edit AI slop. Part of what makes that hard as well as annoying is that it is all reasonably well written and plausible. Just not factual. That’s the real problem I think all of society is likely to face soon if not already. Not to mention the upcoming problem of new AI models trained on the internet of slop.
- Wowfunhappy 1y ago
  ...can I ask what your job is where you're spending a significant portion of your day editing an AI's prose?
SoftTalker 1y ago
> LLMs are so good at bullshitting They are trained on it.

pvg 1y ago

Most people don't want to spend much time reading 'median human' writing so the claim that LLMs are 'better', even if true, doesn't really say that much. We don't listen to median human music either.

Wowfunhappy 1y ago
> We don't listen to median human music either. Don't we? What about the music that plays in the elevator, or while you're on hold on the phone? Similarly for writing, I would imagine you read plenty of emails that are of more-or-less median writing quality. And yet, these emails may discuss pivotal decisions, where it is very important whether their arguments are logically correct.
- pvg 1y ago
  Get a median human to write some elevator or hold music and see how that goes. There's an entire business of making and licensing such music which would not be the case if any rando could just crank it out. Emails, etc, are just 'communication but over text' - it's 'writing' in the very basic sense but it's not the sort of writing people concerned with style and quality care about, either as consumers or producers. Me talking to the cashier at the store is not 'public speaking', neither I nor the cashier care it's not the Gettysburgh Address.
  - the_af 1y ago
    I think emails summarizing meetings, or making a case for something, or describing something, match the kind of writing PG means in his essay. And like the commenter you're replying to, I think that LLMs today can write far better than your average coworker (assuming they don't go on a hallucinated tangent; then again some coworkers also do!). So it's fair to compare LLMs and work emails. Not to speak of work wiki pages (Confluence etc) describing technical things, decisions, policies, etc.
    - pvg 1y ago
      Why do you think pg is writing about emails and meeting notes? I don't readily see anything in the piece to suggest that. 'making a case for something' or 'describing something' covers the bulk of writing, the piece is quite explicit about not being about all writing.
      - Wowfunhappy 1y ago
        Why do you think he's not? It seems to me there is no difference in kind between an email arguing for or examining an important decision or idea, and an "essay" on an important decision or idea. I definitely don't think these documents are just "written speech". Some emails are—a quick message asking if you'll be in the office tomorrow, for example—but major ones require significantly more thought.
        pvg 1y ago
        It seems to me there is no difference in kind between an email arguing for or examining an important decision or idea, and an "essay" on an important decision or idea. Why are there people who make a living as, say, newspaper opinion columnists or magazine staff writers? Or hell, why aren't we all rich substackers given that writing a substack is literally email?

pdonis 1y ago

> LLMs provide a clear counterexample to his thesis. No, they don't. His thesis is not that writing that looks good--that seems plausible and convincing, like the output that LLMs often produce--actually is. He says explicitly in the article: "we can't safely conclude that beautiful writing is true". His actual thesis is in the very next clause of that sentence: "it's usually safe to conclude the converse: something that seems clumsily written will usually have gotten the ideas wrong too."

the_af 1y ago
To be fair, most people commenting on it here on HN seem to understand the first meaning (looks good => is good). I can guess what this says about the readership of HN, but what does it says about the clarity of PG's essay?
- pdonis 1y ago
  > what does it says about the clarity of PG's essay? This seems pretty clear to me: "[W]hile we can't safely conclude that beautiful writing is true, it's usually safe to conclude the converse: something that seems clumsily written will usually have gotten the ideas wrong too." And I'm not the only one who has quoted that passage in this discussion.
  - the_af 1y ago
    It's the second to last paragraph, when someone reading the essay may have already formed a hard to dislodge opinion on the kind of writing PG means. And indeed, many people here on HN did so. The order of paragraphs, how early in the text ideas or clarifications are introduced, etc, all make for clear writing. I'd say this is an example of an essay that lacks clarity, i.e. one that is not "good writing".
aspenmayer 1y ago
> His actual thesis is in the very next clause of that sentence: "it's usually safe to conclude the converse: something that seems clumsily written will usually have gotten the ideas wrong too." This analysis in the article feels woefully uncharitable and is incompatible with HN guidelines, ironically, but if internalized leads directly to the nitpicking style frequently seen on HN as a substitute for elucidation of ideas not well expressed on the way to deeper analysis. Dismissing mediocre writing out of hand seems classist more than anything.

envirogis 1y ago

I had the same reaction. I spend an increasingly large part of my day reading and being required to edit AI slop. Part of what makes that hard as well as annoying is that it is all reasonably well written and plausible. Just not factual. That’s the real problem I think all of society is likely to face soon if not already. Not to mention the upcoming problem of new AI models trained on the internet of slop.

Wowfunhappy 1y ago
...can I ask what your job is where you're spending a significant portion of your day editing an AI's prose?

SoftTalker 1y ago

> LLMs are so good at bullshitting They are trained on it.