From the source
The concept of red-teaming large language models, methods for jailbreaking, and the challenges of evaluating model safety.
From the source
Red-teaming is a form of evaluation that elicits model vulnerabilities that might lead to undesirable behaviors.
huggingface.co
