Job description

Cambridge University, Department of Engineering • Cambridge, UK

Cambridge, UK

In this studentship, you'll pursue a PhD exploring large language model safety through mechanistic interpretability and behavioural research.
Investigate Chain-of-Thought faithfulness and detect deceptive behaviour via perturbation methods and mechanistic analysis.
Monitor LLM behaviour at inference time and develop risk reduction strategies.
Apply either perturbation techniques to test CoT meaning or train models for transparency using human predictor evaluation.
Collaborate with your supervisor to define research direction after completing initial 1.5-year projects.

Applications are handled by the employer.

AI Safety Careers does not process applications directly.

PhD Studentship, Monitoring and Increasing LLM Safety