According to a new study, Large Language Models like ChatGPT are susceptible to “brain rot,” just like humans, if it is subjected to “trivial” content.
Brainrot has become one of the internet’s favorite terms. As you might think, it relates to rotting your brain with content that might not make any sense or have any benefits for consumption. It’s incredibly popular on TikTok, for example, where short videos can be consumed in quick succession.
It is something that LLMs like ChatGPT can also become obsessed with, too. A study from researchers from Texas A&M University, the University of Texas at Austin, and Purdue University fed different artificial intelligence models with “trivial” content.
This junk data was split into two categories, with the first being “highly engaging” social media posts that have gone viral and longer-form content that has a superficial amount of content in it.
AI models can get brainrot too
As per the findings, all four of the models – Llama3 8B, Qwen2.5 7B/0.5B, Qwen3 4B that were tested showed some sort of cognitive decline.
Meta’s Llama model was the most affected by its diet of brainrot content, as its reasoning capabilities suffered drops, it struggled to understand context, and ignored safety standards. Qwen 3 4B was better at dealing with the content it was being fed, but still suffered from the brainrot.
“Brain rot worsens with higher junk exposure — a clear dose-response effect,” Junyuan “Jason” Hong said on X (formerly Twitter).
ChatGPT itself was not tested in the research, but it shows how much AI can suffer when fed content that doesn’t teach it anything of substance.
The artificial intelligence chatbot has become a daily part of many people’s lives, and has even started helping the United States military make some key decisions.
Major General William ‘Hank’ Taylor told Business Insider that he has become “really close” with ChatGPT recently.
Texas lawmakers double down on Discovery, call for DOJ investigation into Smithsonian