Sydney Levine

LitmusValues: Will AI Tell Lies to Save Sick Children? Litmus-Testing AI Values Prioritization with AIRiskDilemmas

PluriHarms: Benchmarking the Full Spectrum of Human Judgments on AI Harm

MoReBench: Evaluating Procedural and Pluralistic Moral Reasoning in Language Models, More than Outcomes

Language Model Alignment in Multilingual Trolley Problems

Hypothesis-Driven Theory-of-Mind Reasoning for Large Language Models

SafetyAnalyst: Interpretable, Transparent, and Steerable Safety Moderation for AI Behavior

Can Language Models Reason about Individualistic Human Values and Preferences?

SafetyAnalyst: Interpretable, transparent, and steerable LLM safety moderation