Brain Rot Is Infecting AI Too: How Doomscrolling Is Breaking Human and Machine Minds

People are writing research papers on which biryani (Indian-flavored rice) is the best, but more on that later. 😅

This might be the most important paper on AI we will read. Scientists are showing how large language models can rot their own minds, in the same way humans get brain rot from endless junk scrolling.

Yes, it’s true. The digital debris that we as humans haven’t been able to escape for the last two decades has now started infecting AI. If OpenAI models like Grok, Gemini, or ChatGPT start feeding on viral social media content, researchers noticed that the cognition of these models collapses. Viral content was defined as short, punchy, hyper-engaging posts that get a lot of engagement. The results were striking, reasoning dropped by 23%, long-context memory fell by 30%, and personality markers spiked toward narcissism and psychopathy.

Here’s the sad part, even after retraining the models on clean, high-quality data, the damage didn’t fully reverse. The “rot” stayed.

I was recently playing with a few children on the streets of Egypt using a soft toy, which my son told me was Labubu, the brain rot toy. It took me a while to reconcile that the brain rot we are all experiencing online was now being manifested in our hands as a physical toy LOL.

And well, if this paper is any proof, the AI version of doomscrolling is real too, and we’ve already let it begin. The bigger problem? Low-quality, high-engagement content isn’t just harmless noise, it can actually reshape how an AI thinks and reasons. That means we risk getting insights from LLMs that aren’t just wrong — they could even be dangerous.

Read the paper HERE.

LLMs Can Get “Brain Rot”!

I asked Notebook LLM to generate an Infographic and I think it did a good job. The point is what is input is what you output.

If AI eats junk, it spits out junk.

AI On Digital Junk Diet Leads To Brain Rot

Featured Image: Labubu

– 0 –