Like Humans… AI Models Are Vulnerable to “Brain Rot”

In recent years, the term “brain rot” has become widely used among scientists and health experts to describe a state in which a person experiences poor concentration, memory problems, and increased anxiety and depression, mainly due to excessive consumption of superficial or low-quality content, especially online.

Researchers are now warning that this phenomenon could also affect artificial intelligence models, according to Fortune magazine.

A recent (yet unpublished) study found that prolonged exposure of large language models to short, viral social media posts can lead to permanent cognitive decline, characterized by a reduced ability to think deeply and understand long or complex contexts.

The researchers explained that this type of training causes a phenomenon known as “thought skipping”, where the models fail to build logical, step-by-step reasoning in their answers and bypass critical mental analysis stages.

Even after attempts to fix this degradation by feeding the models with high-quality human data, the cognitive decline remained evident, with a clear gap in performance before and after exposure to short-form content.

The study’s authors emphasize that the effects of “brain rot” appear to be deeply embedded in these models and may be long-lasting, especially since AI is trained on trillions of data points circulating online, making it highly vulnerable to what they call cognitive pollution.