Prioritizes rigorous, large-scale red-teaming benchmarks, particularly focusing on preventing bias and societal harm in model outputs.
Open the full topic