Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Good Writing

paulgraham.com

Post reply on HN
Print view View original

111–120 of 345 posts

Re: Good Writing

#111

Post by pdonis »

breckinloggins wrote: post #69

Didn’t the ancients define “sophistry” as (in part) exactly the kind of writing one does when influence is more important than truth? It might be that it’s hard to create sophistry accidentally in one’s writing, but it’s certainly a possible - and common - trick. The danger is when you convince yourself that what you’re writing isn’t sophistry… because - after all - it looks good.

PG agrees that you can't conclude that beautiful writing, writing that seems plausible and convincing, actually is. Sophistry is writing (or speaking) that seems plausible and convincing, but is false. PG is not saying that's not possible.

What he's saying is that writing that is ugly is highly likely to be wrong. Which has nothing to do with sophistry.

Re: Good Writing

#112

Post by pdonis »

Wowfunhappy wrote: post #19

...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…

> 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."

Re: Good Writing

#113

Post by pdonis »

baxtr wrote: post #32

Maybe it’s true that writing that sounds good is also more likely to be right. But I’m pretty sure this doesn’t hold for speech.

> Maybe it’s true that writing that sounds good is also more likely to be right.

That's not what PG is saying in the article. He explicitly says that you can't conclude this.

Re: Good Writing

#114

Post by envirogis »

Wowfunhappy wrote: post #19

...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…

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.

Re: Good Writing

#117

Post by spyckie2 »

While writing, you edit your writing towards both sounding good and having solid ideas.

That's the entire essay's point and I largely agree with it.

Its why I think current LLMs are bad writers, not because their prose or ideas are off, but AI generated writing does not have the same quality of robustness from ideas that are thoroughly vetted through the author's editing of it.

At the end of the day, good writing takes a lot of time thinking through what you are really saying and standing behind it. LLMs cannot do that for you (yet).

LLMs can certainly be a helpful tool, mostly by unblocking authors via creating prose, honing in on accurate expressions, researching edge cases and suggesting arguments or counter arguments.

But the craft of sharpening an idea to a very fine, meaningful, well written point is something that is still far off.

Of course, until the next research paper completely proves me wrong.

Step 1. RLHF on real time edited documents Step 2. Profit??

Re: Good Writing

#119

Post by logic_node »

This essay nails it, clear thinking really does lead to clear writing. It’s a good reminder that writing is less about sounding smart and more about being understood.

Re: Good Writing

#120

Post by andrewrn »

The point about end-notes being a mechanism to ease the strain of fitting tree-like ideas into a linear essay is lovely. It brings to mind David Foster Wallace's writing, which is obsessively end-noted and if you listen to his speeches, you can see that he basically tortures himself in sanding down his ideas, much like PG says.

PG's ideas in here, to the extent that I agree with them (which is not fully), does break down for ideas. Example being: brilliant engineers who are incredibly capable at having ideas and executing against them but incredibly incapable of communicating said ideas. Their ideas are very true, evidenced by their ability to produce real results, but also oftentimes ugly when communicated.

A final counterpoint is JFK's eulogy, which sounds amazing, but, after the initial emotional appeal wore off, I realized doesn't really have a strong unified thread running through it, and is thus forgettable in terms of the truths it ostensibly delivers. https://www.youtube.com/watch?v=DOiDUbaBL9E. Compare to "This Is Water" by DFW, which doesn't have the same epic prose, but is maybe the most true-seeming speech I've ever heard https://www.youtube.com/watch?v=DCbGM4mqEVw. It could be that PG's ideas were never intended for spoken speeches, but whatever this is still an adjacent truth != beauty example.

Post reply on HN