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 talked like that before this happened, and now I just feel like my diction has been maligned :p

I think it’s because I was a pretty sheltered kid who got A’s in AP english. The style we’re calling “obviously AI” is most like William Faulkner and other turn-of-the-20th-century writing, that bloggers and texters stopped using.

Don't forget the "it's not just X, it's Y" formulation and the rule of 3.

Lol. This is brilliant. I'm not sure if anyone else has this happen to them, but I noticed in college my writing style and "voice" woukd shift quite noticeably depending on whatever I was reading heavily. I wonder if I'll start writing more like an LLM naturally as I unavoidably read more LLM-generated content.

hehe, I see what you did there.

why do they always say "not only" or "it isn't just x but also y and z"? I hated that disingenuous verbosity BEFORE these LLMs out and now it'll all over the place. I saw a post on linked in that was literally just like 10+ statements of "X isn't just Y, it's etc..." and thought I was having a stroke

LLM slop is not just bad—it's degrading our natural language.

thanks, I hate it.

It’s a text generator regurgitating plausible phrases without understanding and producing stale and meaningless pablum. It doesn’t even know what the intended information is, and judging from the above neither did the human involved.

It doesn’t help writing it stultifies and gives everything the same boring cheery yet slightly confused tone of voice.

> If it conveys the intended information then what's wrong with that?

Well, the issue is precisely that it doesn’t convey any information.

What is conveyed by that sentence, exactly ? What does reframing data curation as cognitive hygiene for AI entails and what information is in there?

There are precisely 0 bit of information in that paragraph. We all know training on bad data lead to a bad model, thinking about it as “coginitive hygiene for AI” does not lead to any insight.

LLMs aren’t going to discover interesting new information for you, they are just going to write empty plausible sounding words. Maybe it will be different in a few years. They can be useful to help you polish what you want to say or otherwise format interesting information (provided you ask it to not be ultra verbose), but its just not going to create information out of thin air if you don't provide it to it.

At least, if you do it yourself, you are forced to realize that you in fact have no new information to share, and do not waste your and your audience time by publishing a paper like this.

Nothing wrong with using LLMs—until every paragraph sounds like it’s A/B tested for LinkedIn virality. That’s the rot setting in.

The problem isn’t using AI—it’s sounding like AI trying to impress a marketing department. That’s when you know the loop’s closed.

The problem is that writing isn't only judged on whether it conveys the intended information or not. It's also judged on whether it does that well, plus other aesthetic criteria. There is such a thing as "good writing", distinct from "it mentioned all the things it needed to mention".

If you can’t understand the irony inherent in getting an LLM to write about LLM brainrot, itself an analog for human brainrot that arises by the habitual non use of the human brain, then I’m not sure what to tell you.

Whether it’s a tsunami and whether most people will do it has no relevance to my expectation that researchers of LLMs and brainrot shouldn’t outsource their own thinking and creativity to an LLM in a paper that itself implies that using LLMs causes brainrot.

What information is conveyed by this sentence?

Seems like none to me.

it’s not really clear whether it conveys an “intended meaning” because it’s not clear whether the meaning - whatever it is - is really something the authors intended.

The brainrot apologists have arrived

Style is important in writing. It always has been.

Because it sounds like shit? Taste matters, especially in the age of generative AI.

And it doesn’t convey information that well, to be honest.

I like to call it tuning, it's more accurate to the way they "learn" by adjusting coefficients and also there's no proven similarity between any existing AI and human cognition.

Sometimes I wonder if any second order control system would qualify as "AI" under the extremely vague definition of the term.

>Turns out that LLMs learn a lot better and faster from educational content as well. This is partly because the average Common Crawl article (internet pages) is not of very high value and distracts the training, packing in too much irrelevant information.

>The average webpage on the internet is so random and terrible it's not even clear how prior LLMs learn anything at all. You'd think it's random articles but it's not, it's weird data dumps, ad spam and SEO, terabytes of stock ticker updates, etc. And then there are diamonds mixed in there, the challenge is pick them out.

https://x.com/karpathy/status/1797313173449764933

Context: FineWeb-Edu, which used Llama 70B to [train a classifier to] filter FineWeb for quality, rejecting >90% of pages.

https://huggingface.co/spaces/HuggingFaceFW/blogpost-fineweb...

From the current WSJ front page:

Paul Ingrassia's 'Nazi Streak'

Musk Tosses Barbs at NASA Chie After SpaceX Criticism

Travis Kelce Teams Up With Investor for Activist Campaign at Six Flags

A Small North Carolina College Becomes a Magnet for Wealthy Students

Cracker Barrel CEO Explains Short-Lived Logo Change

If that's the benchmark for high quality training material we're in trouble.

Yes but the other interesting bit which is not clearly addressed is that increasing the garbage in to 100% does not result in absolute garbage out. So visibly there is still something to learn there.

In case anyone missed the reference: https://arxiv.org/abs/1706.03762

> (...) We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely.

In today's hyper saturated world, attention is everything:

- consumer marketing

- politics

- venture fundraising

When any system has a few power law winners, it makes sense to grab attention.

Look at Trump and Musk and now Altman. They figured it out.

MrBeast...

Attention, even if negative, wedges you into the system and everyone's awareness. Your mousey quiet competitors aren't even seen or acknowledged. The attention grabbers suck all the oxygen out of the room and win.

If you go back and look at any victory, was it really better solutions, or was it the fact that better solutions led to more attention?

"Look here" -> build consensus and ignore naysayers -> keep building -> feedback loop -> win

It might not just be a societal algorithm. It might be one of the universe's fundamental greedy optimization algorithms. It might underpin lots of systems, including how we ourselves as individuals think and learn.

Our pain receptors. Our own intellectual interests and hobbies. Children learning on the playground. Ant colonies. Bee swarms. The world is full of signals, and there are mechanisms which focus us on the right stimuli.

6-7 ٩(●•)_

I don't think anyone is throwing raw datasets into LLMs and hoping for high quality weights anymore. Nowadays most of the datasets are filtered one way or another, and some of them highly curated even.

Garbage in -> Magic -> Hallucinated Garbage out

It is a blog post, it was published as a Github page and on arXiv.

I think it's intended as a catchy warning to people who are dumping every piece of the internet (and synthetic data based on it!) that there are repercussions.

Yeah, I don't think surrealism or constructed is good in the early data mix, but as part of mid or post-training seems generally reasonable. But also, this is one of those cases where anthropomorphizing the model probably doesn't work, since a major negative effect of Cocomelon is kids only wanting to watch Cocomelon, while for large model training, it doesn't have much choice in the training data distribution.

I am mostly concerned with the irreversibility part. More developed brains probably would not be affected as much.

Agreed, “ Popularity as a better indicator”. Hypothetically you could look at popularity over time to filter out viral rot content and work out if people feel the content is useful.

Blended in how? To the training set?

> train on bad data, get a bad model

Right: in the context of supervised learning, this statement is a good starting point. After all, how can one build a good supervised model if you can't train it on good examples?

But even in that context, it isn't an incisive framing of the problem. Lots of supervised models are resilient to some kinds of error. A better question, I think, is: what kinds of errors at what prevalence tend to degrade performance and why?

Speaking of LLMs and their ingestion processing, there is a lot more going on than purely supervised learning, so it seems reasonable to me that researchers would want to try to tease the problem apart.