AI Brain Rot: Nowadays the word brain rot is being discussed a lot on social media. It refers to a situation when people constantly consume poor, mindless and low-effort content, which affects their ability to concentrate and memory. Till now this problem was considered limited to humans, but a new research has proved that AI is also not untouched by this digital disease.
Cornell University’s study raised the curtain
This study has been published by Cornell University. The research focused on the LLM Brain Rot Hypothesis, according to which if a Large Language Model (LLM) is continuously trained on poor or confusing Internet text, its ability to think and reason gradually begins to decline.
How does brain rot start in AI?
During the experiment, researchers continuously showed posts from X (formerly Twitter) to an AI model. These posts were selected that were very viral, clickbait or had sensational words like TODAY ONLY or WOW. After providing this data, the reasoning and understanding ability of AI was evaluated on benchmarks like ARC and RULER.
The results were surprising; the score on ARC test (reasoning ability) dropped from 74.9 to 57.2. The score on the RULER test (long-context understanding) fell from 84.4 to 52.3.
AI became angry and insensitive
According to the study, AI influenced by such data started thought-skipping i.e. skipping the process of logical thinking before giving answers and started giving wrong or incomplete information. Apart from this, negative behavioral traits also started developing in the model such as narcissism and psychopathic tendencies increased while cooperative and responsible nature was seen to decrease.
And the most worrying thing is that even when the model was later trained with higher quality data, the effect of the old junk data was not completely eliminated.
Can this AI brain rot be stopped?
This study makes it clear that if not only humans but also Artificial Intelligence is being affected by poor online content, then it is a serious warning for the future. Amidst the daily flood of fake, sensational and misinformation on the internet, providing clean data to AI has become a difficult but necessary challenge.
According to researchers, companies should now reconsider the process of AI data collection. Simply collecting data by scraping the internet is no longer enough. Quality control systems, filtering mechanisms, and data validation have become extremely important for AI so that it can avoid “cumulative harm” in the long run.
Also read:
WhatsApp’s new feature Dhamaka! Now those who send repeated messages will be banned, know how this rule will work
