Topic: Alignment

Develop data and algorithmic solutions to advance safety and fairness in Gemini models, including post-training of large language models, adversarial robustness improvement, and evaluation protocol design.