Researchers Warn Of Unchecked Toxicity In Ai Language Models

Researchers from MIT are using machine learning to develop a “red-team language model” that generates problematic prompts to test chatbots for safety. This approach is faster and more effective than traditional “red-teaming” methods.