MentalHealthBench released openly for researchers and model developers
The announcement describes comparisons across 10 behaviors: context-seeking varies substantially across models, while empathy and support are more consistent.
TLDR
The MentalHealthBench announcement says the benchmark is being released openly so researchers and model developers can build on it. It notes that models with similar overall scores can have different strengths. Across the 10 behaviors examined, context-seeking varies substantially between models, it says, while empathy and support are more consistent.
MentalHealthBench released openly for researchers and model developers
The announcement describes comparisons across 10 behaviors: context-seeking varies substantially across models, while empathy and support are more consistent.
TLDR
The MentalHealthBench announcement says the benchmark is being released openly so researchers and model developers can build on it. It notes that models with similar overall scores can have different strengths. Across the 10 behaviors examined, context-seeking varies substantially between models, it says, while empathy and support are more consistent.
