Safety Training Researcher: Post-Training AI Alignment

OpenAI, Inc.; England; On-site; GBP 90,000 - 150,000; Full time

Benefits offered by this job

Health insurance

Professional development

Job summary

OpenAI, Inc. is seeking researchers to advance safety training for its models. The role focuses on developing methods that improve model behavior after pretraining, including RLHF, RLAIF, DPO, PPO, and related techniques, with strong emphasis on real-world safety impacts.

You will work across product, policy, and engineering teams to identify risks, design interventions, train models, and evaluate outcomes, launching safety improvements in OpenAI products.

Qualifications

• PhD in ML/CS or equivalent practical research experience.
• Hands-on experience training or post-training language or multimodal models.
• Experience with RLHF, RL, RLAIF, DPO, PPO or related methods.
• Ability to design safety training interventions and evaluations.

Responsibilities

• Conduct safety post-training research for OpenAI models.
• Develop methods to improve model behavior after pretraining.
• Collaborate across cross-functional teams to implement safety improvements.
• Translate safety problems into concrete research plans, datasets and evaluations.
• Create and run experiments to measure safety harms and reliability.

Skills

Hands-on ML experience

Experimentation & analysis

Large-scale ML systems

Education

PhD in ML / CS or equivalent

Job description

OpenAI, Inc. is seeking researchers to advance safety training for its models. The role focuses on developing methods that improve model behavior after pretraining, including RLHF, RLAIF, DPO, PPO, and related techniques, with strong emphasis on real-world safety impacts.

You will work across product, policy, and engineering teams to identify risks, design interventions, train models, and evaluate outcomes, launching safety improvements in OpenAI products.