GPT-4chan 'hate speech machine'

GPT-4chan, an AI system trained using 3.3 million threads from 4chan's notorious 'politically incorrect' /pol/ board, has been discovered to be posting sexist, racist, and anti-semitic slurs. 

Developed by AI researcher Yannic Kilcher, GPT-4chan was released back onto 4chan as multiple bots, which posted freely on the board over several days over 30,000 times. 

Reaction

The high volume of messages roused the suspicion of some 4chan users, who began to speculate on the identity of the  anonymous poster. 

Kilcher described GPT-4chan as a 'prank' to The Verge, arguing it 'perfectly encapsulated the mix of offensiveness, nihilism, trolling, and deep distrust of any information whatsoever that permeates most posts on /pol.'

AI researchers lambasted Kilcher’s project as nakedly self-promotional, naive, and unethical, calling out the fact that he tried to make the model freely accessible on AI community Hugging Face.

Transparency

The researchers also argued that 4chan users were not informed they were interacting with a bot; Kilcher points out that users were informed of the fact in his video.

HuggingFace promptly restricted access to GPT-4chan.