Research Scientist, Gemini Safety
Core
Advancing safety, fairness, and alignment of state-of-the-art AI models (Gemini) through post-training, instruction tuning, and adversarial robustness research.
Role type
Research Scientist (AI Safety & Alignment)
Builds
Foundational safety and alignment technology for Gemini App, Cloud API, and Search
Domain
Artificial Intelligence / Large Language Models / AI Safety
Deliverable
production ML models
Required skills
LLM post-training, instruction tuning, adversarial robustness research, evaluation protocol design, experimental planning, Supervised Fine Tuning, Reinforcement Learning fine-tuning
Preferred skills
Reward modeling, Long-range Reinforcement learning, Safety/Fairness/Alignment research, publication record in top ML conferences, concept-to-product execution, applied research leadership, JAX
Technologies
JAX, LLMs, Reinforcement Learning frameworks
Responsibilities
Post-train and instruction-tune state-of-the-art LLMs across text, image, video, and audio modalities; explore data and algorithmic solutions to ensure model safety and helpfulness; improve adversarial robustness against high-stakes abuse risks; design and maintain evaluation protocols for safety and fairness gaps; execute experimental plans to address gaps or build new capabilities; drive innovation in Supervised Fine Tuning and Reinforcement Learning at scale
