Research Engineer Human Influence
UK government research organisation that evaluates advanced AI risks and develops and tests mitigations to inform governments.
About AI Security Institute
The AI Security Institute (AISI) is a research organisation within the UK Department for Science, Innovation and Technology. It conducts technical research, evaluates leading AI systems, develops risk mitigations, and shares evaluation infrastructure such as Inspect.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will build scalable systems for model evaluations, benchmarks, and LLM post-training. You will develop reinforcement-learning environments, apply interpretability methods, deliver engineering-heavy research projects, and analyse model behaviour to design mitigations.
Requirements
- AI safety
- Machine learning
- Large language model post-training
- Fine-tuning
- Reinforcement learning
- PyTorch
- Keras
- JAX
- Python
- Docker
- Kubernetes
- Ray
- FastAPI
- SLURM
- Model evaluation
- Benchmark
- Production code
Responsibilities
- Design and build reinforcement-learning environments to mitigate harmful model behaviour
- Apply interpretability methods to identify concerning model behaviour and design mitigations
- Build scalable architecture for model evaluations and benchmarks
- Deliver engineering-heavy research projects using post-training techniques on large compute clusters
Benefits
- Pre-release access to frontier models and compute
- Learning and development stipend
- Conference and external collaboration funding
- Hybrid working
- Occasional remote work abroad
- Work-from-home equipment stipend
- At least 25 days of annual leave
- 8 public holidays
- Additional team-wide breaks
- 3 volunteering days
- Paid parental leave
- Employer pension contribution
- Cycling, donation, retail, and gym discounts