MentalHealthBench
An AI evaluation benchmark designed with input from mental health professionals, used to measure whether large language model responses in mental health dialogue scenarios appropriately balance helpfulness and safety. It covers realistic conversational situations and the ability to identify risk factors.
Timeline (last 90 days)
MentalHealthBench is an evaluation benchmark designed with expert involvement to measure whether AI systems' responses in real mental health conversation scenarios are both helpful and sufficiently safe.
MentalHealthBench emphasizes a dual standard of "helpful and safe," measuring both helpfulness and safety simultaneously.
MentalHealthBench covers realistic mental health conversations, including ambiguous expressions, multi-turn interactions, and emotional fluctuations.
If widely adopted, MentalHealthBench has the potential to provide quantitative reference criteria for minimum safety thresholds before AI mental health products go live.
All Facts (4)
MentalHealthBench is an evaluation benchmark designed with expert involvement to measure whether AI systems' responses in real mental health conversation scenarios are both helpful and sufficiently safe.
50%UnverifiedMentalHealthBench emphasizes a dual standard of "helpful and safe," measuring both helpfulness and safety simultaneously.
50%UnverifiedMentalHealthBench covers realistic mental health conversations, including ambiguous expressions, multi-turn interactions, and emotional fluctuations.
50%UnverifiedIf widely adopted, MentalHealthBench has the potential to provide quantitative reference criteria for minimum safety thresholds before AI mental health products go live.
50%