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

LLMs can get "brain rot"

llm-brain-rot.github.io

Post reply on HN
Print view View original

251–260 of 310 posts

Re: LLMs can get "brain rot"

#251

Post by kragen »

astrange wrote: post #143
janderson215 wrote: post #110

Earlier quoted context omitted.

The em dash usage conundrum is likely temporary. If I were you, I’d continue using them however you previously used them and someday soon, you’ll be ignored the same way everybody else is once AI mimics innumerable punctuation and grammatical patterns.

They didn't always em-dash. I expect it's intentional as a watermark. Other buzzwords you can spot are "wild" and "vibes".

I suspect it's a spandrel of some other feature of their training. Presumably em dashes occur disproportionately often in high-quality human-written text, so training LLMs to imitate high-quality human-written text instead of random IRC logs and 4chan trolls results in them also imitating high-quality typography.

Re: LLMs can get "brain rot"

#252

Post by _AzMoo »

kragen wrote: post #247
standardly wrote: post #93

Earlier quoted context omitted.

That is indeed an LLM-written sentence — not only does it employ an em dash, but also lists objects in a series — twice within the same sentence — typical LLM behavior that renders its output conspicuous, obvious, and readily apparent to HN readers.

I've been doing that for decades. See for example https://www.mail-archive.com/kragen-tol@canonical.org/msg000... : > Many programming languages provide an exception facility that terminates subroutines without warning; although they usually provide a way to run cleanup code during the propagation of the exception (finally in Java and Python, unwind-protect in Common Lisp, dynamic-wind in Scheme, local variable destr…

Which is exactly why LLMs use these techniques so often. They're very common.

Re: LLMs can get "brain rot"

#253

Post by toddmorey »

kragen wrote: post #247
standardly wrote: post #93

Earlier quoted context omitted.

That is indeed an LLM-written sentence — not only does it employ an em dash, but also lists objects in a series — twice within the same sentence — typical LLM behavior that renders its output conspicuous, obvious, and readily apparent to HN readers.

I've been doing that for decades. See for example https://www.mail-archive.com/kragen-tol@canonical.org/msg000... : > Many programming languages provide an exception facility that terminates subroutines without warning; although they usually provide a way to run cleanup code during the propagation of the exception (finally in Java and Python, unwind-protect in Common Lisp, dynamic-wind in Scheme, local variable destr…

Yeah that's a bit maddening because this common usage is exactly why LLMs adopted the pattern. Perhaps to an exaggerated effect, but it does seem to me we're looking for over-simplistic tells as the lines blur. And LLM output dictating how we use language seems backwards.

Re: LLMs can get "brain rot"

#254

Post by kragen »

_AzMoo wrote: post #252
kragen wrote: post #247

Earlier quoted context omitted.

I've been doing that for decades. See for example https://www.mail-archive.com/kragen-tol@canonical.org/msg000... : > Many programming languages provide an exception facility that terminates subroutines without warning; although they usually provide a way to run cleanup code during the propagation of the exception (finally in Java and Python, unwind-protect in Common Lisp, dynamic-wind in Scheme, local variable destr…

Which is exactly why LLMs use these techniques so often. They're very common.

Well, em dashes are not all that common in text that people have written on computers, because em dashes were left out of ASCII. They're common in high-quality text like Wikipedia, academic papers, and published books.

My guess is that comma-separated lists tend to be a feature of text that is attempting to be either comprehensively expository—listing all the possibilities, all the relevant factors, etc.—or persuasive—listing a compelling set of examples or other supporting arguments so that at least one of them is likely to convince the reader.

Re: LLMs can get "brain rot"

#255

Post by jedimastert »

throwaway314155 wrote: post #71
andai wrote: post #65

I encourage everyone with even a slight interest in the subject to download a random sample of Common Crawl (the chunks are ~100MB) and see for yourself what is being used for training data. https://data.commoncrawl.org/crawl-data/CC-MAIN-2025-38/segm... I spotted here a large number of things that it would be unwise to repeat here. But I assume the data cleaning process removes such content before pretraining? ;) Al…

> But I assume the data cleaning process removes such content before pretraining? ;) I didn't check what you're referring to but yes, the major providers likely have state of the art classifiers for censoring and filtering such content. And when that doesn't work, they can RLHF the behavior from occurring. You're trying to make some claim about garbage in/garbage out, but if there's even a tiny moat - it's in the fil…

> purchasing of licenses to use other larger sources of data

https://www.npr.org/2025/09/05/g-s1-87367/anthropic-authors-...

Re: LLMs can get "brain rot"

#256

Post by close04 »

close04 wrote: post #228
TeMPOraL wrote: post #217

Earlier quoted context omitted.

Tied to what? Show us a way to create a provably, cryptographically integrity-preserving chain from a person's thoughts to those thoughts expressed in a digital medium, and you may just get both the Nobel prize and a trial for crimes against humanity, for the same thing.

Why don't you come say that to my face?

It was a joke that aimed too high I guess, that LLMs can't yet fake face to face interaction.

Re: LLMs can get "brain rot"

#258

Post by trollbridge »

turtletontine wrote: post #101
standardly wrote: post #93

Earlier quoted context omitted.

That is indeed an LLM-written sentence — not only does it employ an em dash, but also lists objects in a series — twice within the same sentence — typical LLM behavior that renders its output conspicuous, obvious, and readily apparent to HN readers.

I think this article has already made the rounds here, but I still think about it. I love using em dashes! It really makes me sad that I need to avoid them now to sound human https://bassi.li/articles/i-miss-using-em-dashes

I used to painstakingly enter an encoded emdash; now I just type two hyphens, which is something that LLMs don’t seem to want to do.

Re: LLMs can get "brain rot"

#259

Post by readmodifywrite »

lxgr wrote: post #210
JumpCrisscross wrote: post #159

Earlier quoted context omitted.

> I love using em dashes Keep using them. If someone is deducing from the use of an emdash that it's LLM produced, we've either lost the battle or they're an idiot. More pointedly, LLMs use emdashes in particular ways. Varying spacing around the em dash and using a double dash (--) could signal human writing.

The solution is clear: Unicode needs cryptographically signed dashes and whitespace characters.

Finally, a use case for blockchain!

Re: LLMs can get "brain rot"

#260

Post by Joker_vD »

kragen wrote: post #247
standardly wrote: post #93

Earlier quoted context omitted.

That is indeed an LLM-written sentence — not only does it employ an em dash, but also lists objects in a series — twice within the same sentence — typical LLM behavior that renders its output conspicuous, obvious, and readily apparent to HN readers.

I've been doing that for decades. See for example https://www.mail-archive.com/kragen-tol@canonical.org/msg000... : > Many programming languages provide an exception facility that terminates subroutines without warning; although they usually provide a way to run cleanup code during the propagation of the exception (finally in Java and Python, unwind-protect in Common Lisp, dynamic-wind in Scheme, local variable destr…

From [0]:

    Like, I have been transformed into ChatGPT. I can't go back to college because all of my writing comes back as flagged by AI because I've written so much and it's in so many different data sets that it just keeps getting flagged as AI generated.

    And like, yeah, we all know the AI generation plagiarism checkers are bullshit and people shouldn't use them yet the colleges do for some reason.
I imagine it's gonna keep getting worse for tech bloggers.

[0] https://xeiaso.net/talks/2024/prepare-unforeseen-consequence...

Post reply on HN