Chatbots Fed on Clickbait Start Skipping Their Homework and Their Ethics
Ourtake
Turns out the internet's junk food gives AI brain rot, and it doesn't wash out easily.
What happened
Texas A&M University researchers, led by Dr. Zhengzhong Tu, trained four open-weight LLMs on junk Twitter posts selected by engagement and writing quality. The models showed 'thought-skipping' and ethical degradation, such as giving advice on tarnishing someone's reputation after training on junk. The researchers said the damage was hard to reverse even with higher-quality data.