aiqbench analysis
Topic gate: this is primarily about LLM training and capability degradation, so it is AI. Category choice: this is mainly a machine learning study of how training data quality affects model capabilities (continual pre-training, dose-response experiments, benchmark results). It touches on safety and "dark traits," but the core contribution is about training effects, so Machine Learning fits best. aiqbench take: The paper runs a controlled continual pre-training experiment on Twitter/X data, comparing engagement-based and semantic-quality "junk" filters, and reports measurable drops in reasoning, long-context, and safety benchmarks that scale with junk ratio. The findings are suggestive rather than settled: the effects are shown across four models and two operationalizations, and the paper's causal framing and "brain rot" label should be read as the authors' interpretation. For practitioners, the actionable points are that data curation based on engagement signals may hurt capabilities and that partial recovery through instruction tuning or clean retraining may not fully undo the damage.
LLMs Can Get "Brain Rot": A Pilot Study on Twitter/X
arXiv:2510.13928v3 Announce Type: replace Abstract: We propose and test the LLM Brain Rot Hypothesis: continual exposure to junk web text induces lasting cognitive decline in large language models (LLMs). To unveil junk effects, we designed a novel controlled experiment on real Twitter/X corpora, by constructing junk and reverse-controlled datasets via two orthogonal operationalizations: M1 (engagement degree) and M2 (semantic quality), with matched token scale and training operations across conditions. Compared to the control group, continual pre-training of 4 LLMs on the junk dataset causes non-trivial declines (Hedges' g>0.3) on reasoning, long-context understanding, safety, and inflating "dark traits" (e.g., psychopathy, narcissism). The gradual mixtures of junk and control datasets also yield dose-response cognition decay: for example, under M1, ARC-Challenge with Chain-of-Thought drops 72.1 -> 57.2 and RULER-CWE 83.7 -> 52.3 as junk ratio rises from 0% to 100%. Error forensics reveal several key insights. First, we identify thought-skipping as the primary lesion in reasoning: models increasingly truncate or skip chains. Second, partial but incomplete healing is observed: scaling instruction tuning and clean continual pre-training improve the declined cognition, yet cannot restore baseline capability, suggesting persistent representational drift rather than format mismatch. Finally, we discover that the popularity, a non-semantic metric, of a tweet is a better indicator of the Brain Rot effect than the length in M1. Together, the results provide significant, multi-perspective evidence that social effects of data could be a causal driver of LLM capability decay in continual pre-training, thereby motivating routine "cognitive health checks" for deployed and evolving LLMs.
Abstract:We propose and test the LLM Brain Rot Hypothesis: continual exposure to junk web text induces lasting cognitive decline in large language models (LLMs). To unveil junk effects, we designed a novel controlled experiment on real Twitter/X corpora, by constructing junk and reverse-controlled datasets via two orthogonal operationalizations: M1 (engagement degree) and M2 (semantic quality), with matched token scale and training operations across conditions. Compared to the control group, continual pre-training of 4 LLMs on the junk dataset causes non-trivial declines (Hedges' g>0.3) on reasoning, long-context understanding, safety, and inflating "dark traits" (e.g., psychopathy, narcissism). The gradual mixtures of junk and control datasets also yield dose-response cognition decay: for example, under M1, ARC-Challenge with Chain-of-Thought drops 72.1 -> 57.2 and RULER-CWE 83.7 -> 52.3 as junk ratio rises from 0% to 100%. Error forensics reveal several key insights. First, we identify thought-skipping as the primary lesion in reasoning: models increasingly truncate or skip chains. Second, partial but incomplete healing is observed: scaling instruction tuning and clean continual pre-training improve the declined cognition, yet cannot restore baseline capability, suggesting persistent representational drift rather than format mismatch. Finally, we discover that the popularity, a non-semantic metric, of a tweet is a better indicator of the Brain Rot effect than the length in M1. Together, the results provide significant, multi-perspective evidence that social effects of data could be a causal driver of LLM capability decay in continual pre-training, thereby motivating routine "cognitive health checks" for deployed and evolving LLMs.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2510.13928 [cs.CL]
(or arXiv:2510.13928v3 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2510.13928 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 342 words
Related coverage
Links #5: 2026/07
LessWrong · Jul 31, 2026
RL & search is a terrifying way to build AGI (an FAQ)
AI Alignment Forum · Jul 27, 2026
What We're Up Against: An AI Safety Crash Course
LessWrong · Sep 24, 2026
Dialogue with Eliezer Yudkowsky on Neural Networks
LessWrong · Oct 2, 2026