🚀 At the start of the month, we submitted comments to National Institute of Standards and Technology (NIST)'s request for information on developing standards and best practices for safety and security and reducing risk as part of its assignments under the #AIExecutiveOrder Our feedback focuses on providing information on Model Risks, Model Modalities, Modes of Model Safety Operationalisation, Benchmarking, Red-Teaming, and enabling the responsible development and deployment of generative models. Based on our rich experience in algorithm auditing, risk management, and assurance, we recommend: → Grounding approaches in applied academic and industry research → Investing in and furthering existing research initiatives on benchmarking → The standardization and systematization of Red-Teaming → Consideration of nuances in different modalities → Using the NIST AI 100-1 to guide the responsible release of generative models Check out our full response below 👇 #NIST #AIgovernance #AIstandards #llms #ethicalAI #AIsafety
