AI RESEARCH
How catastrophic is your LLM?
Amazon Science
•
A new framework provides a statistical method for estimating the likelihood of catastrophic failures in large language models in adversarial conversations.