Research Engineer — AI Safety & Human Influence

Aisi; Greater London; Hybrid; GBP 65,000 - 115,000; Full time

Benefits offered by this job

Hybrid working

25 days annual leave

Pension contributions

Job summary

AI Security Institute, based in the UK, is seeking a Research Engineer to join the Human Influence team. You will conduct scalable work in LLM post-training and reinforcement learning, or equivalent engineering validation of large-scale evaluation pipelines.

Projects include designing RL environments to reduce model deception, leveraging interpretability to guide mitigations, and building the architecture for repeatable model evaluations on large compute clusters.

Qualifications

• Proven experience deploying a benchmark, evaluation, or product to users.
• Clear understanding of the current AI safety literature, and interest in Human Influence topics.
• Clear and consistent communication.
• Clear understanding of fundamental Machine Learning concepts.

Responsibilities

• Design and build RL environments to mitigate model deception in one-to-one or multi-agent settings.
• Apply interpretability methods to identify concerning model behavior and design mitigations.
• Build scalable architecture for repeatable evaluation delivery and benchmarking.
• Deliver engineering-heavy research projects on Human Influence topics using post-training techniques on large compute clusters.

Skills

Benchmarking deployments

AI safety literature

Clear communication

Machine learning concepts

Tools

PyTorch

Keras

JAX

Docker

Kubernetes

Ray

FastAPI

SLURM

Job description

AI Security Institute, based in the UK, is seeking a Research Engineer to join the Human Influence team. You will conduct scalable work in LLM post-training and reinforcement learning, or equivalent engineering validation of large-scale evaluation pipelines.

Projects include designing RL environments to reduce model deception, leveraging interpretability to guide mitigations, and building the architecture for repeatable model evaluations on large compute clusters.