Research Scientist, Superalignment

About the Team

OpenAI’s Superalignment Team is working on technical approaches to ensure that superintelligence–an AI system vastly smarter than humans–follows human intent. 

Through scientific experimentation, we explore the scalability of alignment techniques and identify potential breaking points. Our approach to alignment research includes a range of different projects; some of these will help us improve the alignment of our models and others will allow us to validate how aligned our models actually are:

• Scalable oversight: How can we best leverage AI systems to assist evaluation of other AI systems on difficult tasks?
• Generalization: Can we understand and control how our models generalize from easy tasks that humans can supervise to hard tasks that humans cannot?
• Automated interpretability: Can we use AI to explain how LLMs work internally?
• Robustness: How can we train our models to be aligned in worst-case situations?
• Adversarial testing: If we deliberately train deceptively aligned models as testbeds,  can our oversight techniques, interpretability tools, and evaluations detect this misalignment?

We want to figure out how to spend vast amounts of compute to solve this problem, in particular by automating alignment research itself.

About the Role

As a Research Scientist here, you will develop innovative machine learning techniques and advance the research agenda of the Superalignment team, while also collaborating with peers across the organization. We are looking for people who want to discover simple, generalizable ideas that work well even at large scale, and form part of a broader research vision that unifies the entire company.

We are seeking Research Scientists to help design and implement experiments for alignment research. Responsibilities may include:

• Designing experiments to measure the effectiveness of scalable oversight techniques such as AI-assisted feedback and Debate
• Studying generalization to see when AI systems trained on easy problems can solve hard problems
• Managing large datasets from interpretability experiments and creating visualizations to explore interpretability data
• Developing experiments to test how well chain of thought reasoning reflects model cognition
• Investigating situations when training against a reward signal causes model outputs to deteriorate
• Exploring methods to understand and predict model behaviors, such as finding inputs causing anomalous circuits or catastrophic outputs
• Designing novel approaches for using LLMs in alignment research

You might thrive in this role if you:

• Are excited about OpenAI’s mission of building safe, universally beneficial AGI and are aligned with OpenAI’s charter.
• Have a track record of coming up with new ideas or improving upon existing ideas in machine learning, demonstrated by accomplishments such as first author publications or projects
• Possess the ability to own and pursue a research agenda, including choosing impactful research problems and autonomously carrying out long-running projects
• Possess a strong curiosity about aligning and understanding ML models, and are motivated to use your career to address this challenge
• Enjoy fast-paced, collaborative, and cutting-edge research environments
• Have experience implementing ML algorithms (e.g., PyTorch)
• Can develop data visualization or data collection interfaces (e.g., JavaScript, Python)
• Want to ensure that powerful AI systems stay under human control