Research Engineer, AI Safety & Alignment

About this role

Develop and implement AI safety techniques to make large language models more reliable, honest, and aligned with human values. This role combines cutting-edge research in model alignment and interpretability with practical engineering to protect millions of users.

What you'll do

• Design evaluation methodologies and metrics to assess safety and alignment of large language models
• Research and develop alignment techniques including value learning, interpretability, and RLHF
• Conduct adversarial testing to identify vulnerabilities and failure modes
• Mitigate biases, toxicity, and harmful behaviors in models through fine-tuning and reinforcement learning
• Translate safety research into scalable solutions with engineering and product teams
• Publish findings and contribute to AI safety research community

What they're looking for

• Machine learning and transformers architecture
• Reinforcement learning from human feedback (RLHF)
• Production code development (Python or similar)
• GPU training, serving, and debugging
• Data pipelines and infrastructure
• Model interpretability and explainability
• Adversarial testing and vulnerability assessment
• Distributed model training

Benefits

• Work on critical AI safety challenges affecting millions of users
• Contribute to academic publications and present at conferences
• Collaborate with leading AI research team at a unicorn AI company
• Located in Redwood City, CA
• Opportunity to shape responsible AI development