Faculty develops frontier AI and machine learning solutions designed to address complex operational challenges across defense, healthcare, financial services, public services, and retail. The company works with organizations to apply advanced AI to areas such as national security, outcome prediction, and process improvement, with an emphasis on practical impact, responsible deployment, and sustainability.
AI Safety Research Scientist – Hybrid in London
Research Scientist role focused on understanding and steering transformer-based large language model behaviour at Faculty. The work combines rigorous AI safety research, reproducible tooling, and practical applications.
Description
- Investigate AI safety questions that support the responsible deployment of trustworthy AI
- Plan and execute experiments to understand and influence model behaviour
- Use white-box and black-box approaches to analyse transformer-based large language models
- Write modular, reproducible Python research software with tools such as PyTorch
- Create and support the infrastructure and technical tools required for research
- Share results through academic publications, technical reports, and presentations
- Work with senior scientists to identify gaps in the literature, test hypotheses, and scope research programmes
- Explain technical findings to frontier laboratories, government organisations, external partners, and safety institutes
Requirements
- Strong foundation in machine learning and transformer architectures, particularly large language models
- Working knowledge of white-box and black-box interpretability methods, including steering vectors and mechanistic interpretability
- Ability to derive well-formed hypotheses from structured technical challenges
- Advanced Python programming skills and experience with deep learning frameworks such as PyTorch
- Evidence of producing clean, modular research software
- Experience managing code with Git
- Clear communication skills for presenting complex technical ideas and experimental findings in reports and presentations
- Curious, rigorous, and analytical approach to research
- Familiarity with AI safety topics such as robustness, evaluation, or uncertainty calibration
- Ability to state whether UK work visa sponsorship is required
Benefits
- Part-time working arrangements may be discussed
- Every application is reviewed by a person rather than decided by AI
- AI tools may be used for research and interview preparation
- Interview AI note-taking can be declined