Researcher, Multimodal Safety

📋 Description

• Research multimodal safety for text, vision, and audio; connect perception to safe model behavior.
• Build training/evaluation methods for vision-language models; post-training, safety evals, and
• Collaborate with internal teams to translate research into safer multimodal experiences.
• Role based in San Francisco with a hybrid work model (3 days in office/week) and relocation

🎯 Requirements

• Strong track record in multimodal models; depth in vision-language, video understanding, image
• Understanding of end-to-end multimodal systems: encoders, fusion, cross-modal reasoning, scaling
• Experience with post-training improvements (SFT, RL) and robust evaluation; data curation and
• Ability to form hypotheses, run decisive experiments, diagnose failures, and translate findings
• Excellent safety judgment; strong research and engineering judgment for open-ended safety problems.

🎁 Benefits

• Relocation assistance
• OpenAI-specific benefits and equal opportunity employer commitment
• Competitive compensation with equity options