LLMs can get "brain rot"

#1

Post by tamnd » Tue, Oct 21, 2025, 2:24 PM UTC

LLMs can get "brain rot"
llm-brain-rot.github.io

Re: LLMs can get "brain rot"

#2

Post by AznHisoka » Tue, Oct 21, 2025, 2:30 PM UTC

Can someone explain this in laymen terms?

Re: LLMs can get "brain rot"

#3

Post by PaulHoule » Tue, Oct 21, 2025, 2:47 PM UTC

AznHisoka wrote: post #2

Can someone explain this in laymen terms?

They benchmark two different feeds of dangerous tweets:

  (1) a feed of the most popular tweets based on likes, retweets, and such
  (2) an algorithmic feed that looks for clickbait in the text

and blend these in different proportions to a feed of random tweets that are not popular nor clickbait and find that feed (1) has more of damaging effect on the performance of chatbots. That is, they feed that blend of tweets into the model and then they ask the models to do things and get worse outcomes.

Re: LLMs can get "brain rot"

#4

Post by ForHackernews » Tue, Oct 21, 2025, 3:29 PM UTC

PaulHoule wrote: post #3

AznHisoka wrote: post #2

Can someone explain this in laymen terms?

They benchmark two different feeds of dangerous tweets: (1) a feed of the most popular tweets based on likes, retweets, and such (2) an algorithmic feed that looks for clickbait in the text and blend these in different proportions to a feed of random tweets that are not popular nor clickbait and find that feed (1) has more of damaging effect on the performance of chatbots. That is, they feed that blend of tweets into…

Blended in how? To the training set?

Re: LLMs can get "brain rot"

#5

Post by pixelmelt » Tue, Oct 21, 2025, 3:33 PM UTC

Isn't this just garbage in garbage out with an attention grabbing title?

Re: LLMs can get "brain rot"

#6

Post by sailingparrot » Tue, Oct 21, 2025, 3:33 PM UTC

AznHisoka wrote: post #2

Can someone explain this in laymen terms?

train on bad data, get a bad model

Re: LLMs can get "brain rot"

#7

Post by CaptainOfCoit » Tue, Oct 21, 2025, 3:37 PM UTC

> continual exposure to junk web text induces lasting cognitive decline in large language models (LLMs).

TLDR: If your data set is junk, your trained model/weights will probably be junk too.

Re: LLMs can get "brain rot"

#8

Post by philipallstar » Tue, Oct 21, 2025, 3:39 PM UTC

pixelmelt wrote: post #5

Isn't this just garbage in garbage out with an attention grabbing title?

Attention is all you need.

Re: LLMs can get "brain rot"

#9

Post by b0gb » Tue, Oct 21, 2025, 3:41 PM UTC

AIs need supervision, just like regular people... /s

Re: LLMs can get "brain rot"

#10

Post by wat10000 » Tue, Oct 21, 2025, 3:44 PM UTC

pixelmelt wrote: post #5

Isn't this just garbage in garbage out with an attention grabbing title?

Considering that the current state of the art for LLM training is to feed it massive amounts of garbage (with some good stuff alongside), it seems important to point this out even if it might seem obvious.