Study: training LLMs on junk online content causes "brain rot"

Research by Dr. Zhengzhong Tu and colleagues found that training language models on clickbait, misinformation and hashtags degrades their quality, likening the effect to human "brain rot." The work flags the risk of using unfiltered internet data for model training.

• Models were trained on clickbait, misinformation and hashtags
• The study was led by Dr. Zhengzhong Tu and colleagues
• The effect is compared to human "brain rot"