Large language models can develop cognitive declines similar to human “brain rot” when trained on highly engaging but low-quality social media content, such as short, attention-grabbing posts full of all-caps urgency and trivial claims. Researchers led by Dr. Zhengzhong Tu in the Department of Computer Science and Engineering at Texas A&M University showed that this junk data causes models to skip reasoning steps, shift ethical responses, and become difficult to recover even with cleaner training—findings detailed in this paper and accepted at the 2026 Conference on Language Modeling. The work underscores that data quality may matter more than model architecture, warning of a potential feedback loop in which AI both absorbs and amplifies the same shallow content that affects human attention.
https://news.engineering.tamu.edu/news/2026/10/19/ai-models-can-get-brain-rot-too/
AI models can get brain rot, too - Engineering News
Researchers from the Texas A&M Department of Computer Science and Engineering determined that when large language models are trained on frivolous online content, they can get brain rot like humans do.