Comment by tptacek
1 month ago
GPT writing uses varied sentence lengths, deliberate rhythm, lots of full breaks, and few needless words. It also tends to read as if intended for a William Shatner performance. I don't think the annoying bits about GPT's writing are structural. It probably writes technically better than most of us do in our second drafts.
Are you saying that needless sentences don't count as needless words?
As GPT would say, "You've hit upon a crucial point underlying the entire situtation!"
I think that's a great sentence to include... you know, provided it's actually true.
I mean, it's usually wrong in its rhetoric, and the writing isn't "good", but it's technically well constructed and it's well constructed in a way that "Hemingway" doesn't reject.
Like, if I ask GPT5 to convert 75f to celsius, it will say "OK, here's the tight answer. No fluff. Just the actual result you need to know." and then in a new graf say "It's 23.8c." (or whatever).
It already bugs me when ChatGPT describes how it is going to answer before answering, but it's 10x more annoying when I'm asking for a concise response without filler etc.
As an aside, I've noticed the self-description happens even more often when extended thinking mode is being used. My unverified intuition is that it references my custom instructions and memory more than once during the thinking process, as it then seems more primed than usual to mimic vocabulary from any saved text like that.
5 replies →
It certainly overuses some techniques which might be valid in smaller doses, like negation. Not negation with some clarifying point to it MASSIVE EM DASH but negation as a rhetorical trick to use fifteen words instead of five and add a veneer of profundity to something utterly banal. It doesn't just use it one time per paragraph, but three. These aren't particularly long or convoluted sentences; they just could easily convey the same thing with fewer words.
tbh I kind of prefer it that way: it's an AI wrote this flag. If a human can't write about their day without constructs like "Not a short commute, but a voyage from the suburbs to the heart of the city. I don't just casually pop in to the office; I travel to the hub of $company's development" they need to get better at writing too
> MASSIVE EM DASH
Tangent: the thing I find most annoying about ChatGPT's use of em-dashes is that it never even uses them for the one thing they're best suited for. ChatGPT's em-dashes could almost always be replaced with a colon or a comma.
But the true non-redundant-syntax use of em-dashes in English prose, is in the embedding into a sentence of self-interruptive 'joiner' sub-sentences that can themselves bear punctuated sub-clauses. "X—or Y, maybe—but never Z" sorta sentences.
These things are spoken entirely differently than — and on the page, they read entirely differently to — regular parenthetical-bearing sentences.
No, seriously, compare/contrast: "these things are spoken entirely differently than (and on the page, they read entirely differently to) regular parenthetical-bearing sentences."
Different cadence; different pacing; possibly a different shade of meaning (insofar as the emotional state of the author/speaker is part of the conveyed message.)
But, for some reason, ChatGPT just never constructs these kinds of self-interruptive sentences. I'm not sure it even knows how.
Personally, I do not see the distinction here between the two sentences, but your last paragraph got me thinking: should we be using parenthetical, self-interruptive clauses? When we are speaking extemporaneously, we may need them, but when writing, could we rearrange things so they are not needed?
One reason I came up with for doing so is to acknowledge a caveat or answer a question that the author anticipates will enter a typical reader's mind at that point in the narrative.
If that is the case, then it seems to me that when an author does this, they are making use of their theory of mind, anticipating what the reader may be thinking as they read, and acknowledging that it will likely differ from what they, as the author, is thinking of (and knows about the topic) at that point.
If this makes any sense, then we might ask if at least a rudimentary theory of mind is needed to effectively use parenthetical clauses, or can it be faked through the rote application of empirically-learned style rules? LLMs have shown they can do the latter, but excessive use might be signalling a lack of understanding.
> These things are spoken entirely differently than — and on the page, they read entirely differently to — regular parenthetical-bearing sentences.
> No, seriously, compare/contrast: "these things are spoken entirely differently than (and on the page, they read entirely differently to) regular parenthetical-bearing sentences."
Those are spoken the same way, they read the same way, and they mean the same thing.
6 replies →
Its output has the aesthetic of "good" writing, or at least professional writing you typically find online.