New AI tools like GPT-3 (a language processing algorithm) are trained with trillions of bytes of text to generate new written content, like articles or image descriptions. This new text reflects the language the AI has been trained on, and can mirror biases in the training data and text sources. AI is also used to automatically curate news articles in social media feeds, and have been known to reflect bias in the content displayed.
Real toxicity promptsPre-trained neural language models (LMs) are prone to generating racist, sexist, or otherwise toxic language which hinders their safe deployment. We investigate the extent to which pre-trained LMs…
Yeah, great start after sacking human hacks: Microsoft’s AI-powered news portal mixes up photos of women-of-color in article about racismBlame Reg Bot 9000 for any blunders in this story, by…
