How to make a chatbot that isn’t racist or sexist | MIT Technology Review
Tools like GPT-3 are stunningly good, but they feed on the cesspits of the internet. How can we make them safe for the public to actually use?
§
Sometimes, to reckon with the effects of biased training data is to realize that the app shouldn't be built. That without human supervision, there is no way to stop the app from saying problematic stuff to its users, and that it's unacceptable to let it do so.