In the race to make artificial intelligence smarter, we might’ve accidentally taught it how to be just as stupid as us.
A new study out of the University of Texas at Austin, Texas A&M, and Purdue University suggests that large language models can develop a kind of “AI brain rot” when trained on the same junk polluting your social media feed.
Researchers, led by Junyuan Hong (now at the National University of Singapore), wanted to know what happens when AI models gorge on the dopamine-dripping content that fuels your endless scrolling.
They fed open-source models like Meta’s Llama and Alibaba’s Qwen a steady diet of viral posts stuffed with words like “wow,” “today only,” or “look”—the kind of words SEO strategies that have been optimized with an inch of their lives demand to be put in headlines to generate clicks.
The quality of the LLM’s output quickly nosedived from there.
AI Models Are Getting Brain Rot From Internet Trash
The models got dumber and meaner. Their ability to reason tanked, along with their ability to hold onto information.
Their ethical alignments eroded, and the AI started scoring higher on measures of psychopathy. The machines became worse versions of themselves, just like people do after spending too much time online subsisting on a diet of low-quality, low-effort trash content.
Unsurprisingly, this is precisely what happens to people who have been fed low-quality information diets that rot our ability to think critically. Remember, brain rot was crowned Oxford Dictionary’s 2024 Word of the Year for a reason.
It’s more than a fun buzzword of the day. It encapsulates a growing problem with our ongoing relationship with the online space. Everything has become toxic. Everything has become low-quality.
And there’s a lot more of all that toxic, low-quality garbage than ever before, and it certainly doesn’t seem like anyone is going to slow it down anytime soon.
The researchers say that AI developers should refrain from training their models on stuff designed to grab clicks. It’s like trying to raise a child on a diet of soda and candy. The more frightening part is that it seems all this AI brain rot is permanent.
Once a model’s neurons grow s**ttier after a steady diet of low-quality data, even retraining it on “clean” text can’t fully restore its reasoning. Its brain is permanently f**ked. Since AI systems are increasingly generating social media content themselves, they’re essentially gobbling up their own vomit.
It’s an endless cycle that’s only making it dumber and dumber, an ouroboros of stupidity stretching into the infinite, eventually reaching an apex of stupidity previously unimagined by us mere mortals.
And then some dips**t on Twitter is still going to use a screenshot of an AI-generated response as the mic drop of a debate, because he, too, is too brain-rotted to understand that he just posted some profoundly stupid AI brain rot.
The future is now, folks.
The post Constant Scrolling Is Giving AI Brain Rot appeared first on VICE.
