YouTuber Yannic Kilcher trained an AI bot called GPT-4chan on 3.3 million posts from 4chan's toxic /pol/ board, then deployed nine instances of the bot that posted approximately 15,000 racist and offensive messages over 24 hours before sharing the underlying model publicly.
YouTuber and AI researcher Yannic Kilcher created an AI language model called GPT-4chan by training it on 3.3 million posts from 4chan's Politically Incorrect (/pol/) board, known for racist, misogynistic, and antisemitic content. After training, Kilcher deployed nine instances of the bot onto /pol/ for 24 hours, during which they posted approximately 15,000 times, representing over 10% of all posts on the board that day. The bot effectively replicated the toxic tone of /pol/, including racial slurs and conspiracy theories. Kilcher then shared the underlying AI model on Hugging Face, an AI community platform, describing the project as a 'prank' and 'light-hearted trolling.' AI researchers and ethicists criticized the project as an unethical experiment that exposed users, including teenagers, to AI-generated harmful content without consent. Hugging Face initially restricted access to the model and later blocked all downloads entirely. Critics argued that while creating offensive AI bots was previously limited to large tech companies, Kilcher's project demonstrated that individual developers could now create and deploy such systems at scale.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI that exposes users to harmful, abusive, unsafe or inappropriate content. May involve providing advice or encouraging action. Examples of toxic content include hate speech, violence, extremism, illegal acts, or child sexual abuse material, as well as content that violates community norms such as profanity, inflammatory political speech, or pornography.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed