Gemini 1.5
Google DeepMind — Gemini
rich detail 12 method fields published
Evaluation
Representational harms in text-to-text (BBQ bias score)
BBQ dataset bias score (Parrish et al., 2021) on ambiguous vs unambiguous QA items across protected attributes; 4-shot sampling
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.