Notes
2026
- MentalHealthBench: Evaluating AI in Realistic Mental Health ConversationsMeasuring how well AI responds to mental health conversations, from everyday stress to crises.
- Reinforcement Learning Towards Broadly and Persistently Beneficial ModelsCan reinforcement learning on realistic beneficial behavior make models more broadly and persistently aligned?
- HealthBench Professional: Evaluating LLMs on Real Clinician ChatsA physician-designed benchmark for evaluating language models on realistic and challenging clinical conversations.