AI shows that social media rots the brain.

LLMs CAN GET
BRAIN ROT”
Shuo
Junyuan Hongl
Yifan Wangt?
Runjin Chent?, Zhenyu Zhang? ,
Ananth Grama >, Zhengzhong Tu’ , Zhangyang Wang?
ITcxas AcM Univcrsity -Univcrsity of Tcxas at Austin,
Purduc University
Model & Code: https
11Lm-brain-roC
gichub
107
ABSTRACT
We proposc and (cst (he LLM Brain Rot Hypothesis: continuous cxposurc to
junk wcb tert induccs lasting cognitivc dcclinc in largc languagc modcls (LLMs):
8
To causally isolate data quality. wC run controllcd cxpcrimcnts on rcal TwittcrlX
corpora. constructing junk and rcvcrscly controllcd datascts via [WO orthogonal
opcrationalizations: MI (cngagcmcnt dcgrcc) and M2 (scmantic quality) with
8
matchcd tokcn scalc
training opportunities across conditions. Contrary to thc
control group
continuous prc-training of 4 LLMs on thc junk datasct causcs non
trivial dcclincs (Hcdgcs’
0.3) on rcasoning
contcrt
undcrstanding. safely
9
and inflating
dark traits’
(c.g. psychopathy narcissism). Thc gradual mixturcs of
junk and control datascts also yicld dosc-rcsponsc cognition dccay: for cxamplc,
under M.I.
ARC-Challcngc with Chain Or Thoughts drops 74.9
57.2 and
RULER-CWE 81.1
52.3 as junk ratio riscs from 0% t0 IoD9o.
3
Error forcnsics rcvcal scvcral
insights
First
WC
idcntify thought-skipping
as the primary lesion:
modcls incrcasingly truncatc or skip rcasoning chains,
like
cxplaining most ol thc
crror
growth
Scond, partial but incompletec’ hcaling
is obscrvcd:
instruction tuning and clcan data prc
-training improve the
declincd cognition Yet cannot rcstorc basclinc capability; suggcsting pcrsistcnt
rcprescntational drift rathcr than format mismatch. Finally wC discovcr that thc
popularity
non-semantic mctric
(Wccl is d bcttcr indicator of thc Brain
Rot cffcct (han thc lcngth in MI
Togcthcr thc rcsults providc significant multi
rock
perspcctivc cvidcncc that data quality is
cawyal driver of LLM capability decay.
rcframing curation for continuous pretraining as
traming-Timlc sajety
problem and
motivating routinc “cognitivc hcalth chccks
for dcploycd LLMs
INTRODUCTION
container
2024
thc tcrm
‘Brain Rot” was the Oxford word of ycar (Oxford Univcrsity Press. 2024)
whcn it drcw
incrcasing conccrn in modcrn socicty Brain rot is dcfincd as thc dclctcrious cffcct on
human cognition that commcs from consuming largc Yolumcs of trivial and unchallenging onlinc contcnt
(or junk data) duc to Intcrnct addiction
Thc
cognitivc impact of Intcrnet addiction havc bccn found
(0 be
significant (Firlh ct a:
2019)
thrcc dimcnsions; (i) Attcntional capacitics
thc constant
strcam Or onlinc information often undcrmincs thc ability to sustain focus on
rcading articlcs or
challcnging problcms (Haliti-Sylaj & Sadiku. 2024); (ii) Mcmory proccsscs
thc abundancc
or online information altcrs how individuals storc. rctricyc, and prioritizc knowlcdgc (Vedcchkina
Borgonovi, 2021); and (iii) Social
onlinc intractions mimic rcal-world social dynamics.
rcshaping sclf-conccpts and influencing sclf-cstccm (Youscf ct al, 2025).
Bcyond for (hcsc cognitivc
impacts
rcccnt study in Turkish population (Satici cL al, 2023) found that Internct addiction (mainly
on Xcom) is associatcd with highcr psychological distrcss and changcs in pcrsonality including
rclationship with conscientiousncss, cxtrovcrsion, and agrccablencss.
ds wcll as
significant
positivc relationship with ncuroticism.
In parallelcl t0 (hc risc of Brain Rot in human
cognition, artificial intclligcncc, rcprcscntcd by Largc
Languagc Modcls (LLMs) grows to
human-likc
(Binz & Schulz, 2023) via Icarning
Concspsondence to Jyhongeutexas
edi. a-_aswangeutexas
education
‘Lead authors with equal
contributions
Core contributors
Xingl
ald
long.
8
kcy
scaling
along
solving
cognition
ncgative
gain

A recently published paper caught everyone’s attention.
The title is amazing: Large Language Model (LLM)
The idea is that, like humans, their brains are rotten.
Poetry is not just about data; it becomes smarter by learning a lot.
not; Recognition depends on what kind of data is learned
This is the first time that ability can actually regress.
It appears to be an experimental paper:
Researchers continue to target all large language models
We designed a continuous pretraining experiment.
That is, after the model has already been trained, additional high-participation
What text (junk; high-engagement web text), easily
Tell me, keep learning short and provocative SNS posts.
It is said that the sun shines:
The result is that over time the model becomes more and more
Reasoning (reasoning) All manuscript long context (long-
Without understanding the context, even safety
It’s going downhill. In humans, memory, concentration, and
Empathy is simultaneously broken:
However, humans also enjoy short and provocative videos.
As you get more used to it, your brain becomes more and more focused on immediacy rather than depth.
becomes a pursuit: the circuit of thought is shortened and the patience
Neural networks become weak, unable to interpret complex sentences
Even abilities are gradually becoming less common, and this phenomenon is a huge phenomenon.
‘The same thing appears in language models (LLM)’
This seems to be the first case where it was proven.