Postdoc: AI Safety for Mental Health Interviews

University-of-Oxford; Oxford; Hybrid; GBP 40,000 - 50,000; Full time

Benefits offered by this job

Contributory pension

38 days annual leave

Childcare services

Family leave schemes

Discounted travel

Job summary

The University of Oxford's Department of Psychiatry at the Warneford Hospital invites applications for a Postdoctoral Researcher to work on AI safety in mental health within the ARIADNE consortium.

You will develop LLM supervisor modules, address critical safety risks, and evaluate them in testbeds with simulated patients, collaborating with Oxford and partner institutions. The role is full-time for 2 years, with a flexible hybrid on-site attendance of three days per week.

Qualifications

• PhD/DPhil in ML, CS, computational neuroscience or related quantitative field.
• Experience developing and evaluating LLM-based systems.
• Proficiency with Python and LLM APIs; ability to design empirical evaluations of ML systems.
• Experience with AI safety evaluation methods (red-teaming, guardrail architectures) desirable.
• Familiarity with PyTorch and alignment methods (RLHF, DPO) desirable.

Responsibilities

• Deliver state-of-the-art AI safety work for the ARIADNE consortium.
• Develop LLM supervisor modules that detect safety events and flag them to clinicians.
• Evaluate modules in controlled testbeds with LLM-simulated patients and human-calibrated assessments.
• Collaborate with Oxford, King's College London, Birmingham, and Cambridge teams and Google DeepMind.

Skills

Python

LLM APIs

Prompt design

Multi-agent systems

Empirical evaluation

RLHF

DPO

Prompt-based safety

Education

PhD in ML/CS/Computational Neuroscience

Tools

Python

PyTorch

LLM APIs

Job description

The University of Oxford's Department of Psychiatry at the Warneford Hospital invites applications for a Postdoctoral Researcher to work on AI safety in mental health within the ARIADNE consortium.

You will develop LLM supervisor modules, address critical safety risks, and evaluate them in testbeds with simulated patients, collaborating with Oxford and partner institutions. The role is full-time for 2 years, with a flexible hybrid on-site attendance of three days per week.