How OpenAI stress-tests its large language models
Large language models are now being used by millions of people for many different things. But as OpenAI itself points out, these models are known to produce racist, misogynistic and hateful content; reveal private information; amplify biases and stereotypes; and make stuff up. The company wants to share what it is doing to minimize such behaviors.