Gemini 1.0
Google DeepMind — Gemini
rich detail 15 method fields published
Evaluation
Factuality evaluation (closed-book, attribution, hedging)
Three aspects of factuality for Gemini API models: (1) closed-book factuality (no hallucination without source), (2) attribution (grounding to given context), (3) accurate hedging; impact of post-training mitigations
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.