Remote Research Scientist: Mechanistic AI Interpretability

PrincInt; Greater London; Hybrid; GBP 93,000 - 126,000; Full time

Benefits offered by this job

Unlimited PTO

Flexible working hours

High ownership & autonomy

Contractor option

Annual health budget

Team meetups

Job summary

PrincInt is hiring Research Scientists to work on Ambitious Mechanistic Interpretability (AMI) with fully remote setup. Preference for time spent in London or Bay Area; visa sponsorship available for London/Berkeley candidates.

Start date January 2027, salary USD 125k–170k gross annually. The role focuses on building data models inspired by physics, quantifying data structure relationships, and developing synthetic datasets to benchmark interpretability tools, with mentorship of junior

Qualifications

• PhD or equivalent research experience in math/physics/ML/CS/neuroscience/complexity.
• Strong ability to communicate complex ideas to technical and non-technical audiences.
• Hands-on training with neural networks and deep learning.
• Experience with scientific computing in Python.
• Comfort with fully remote work.
• Motivation to contribute to AI safety.

Responsibilities

• Develop tractable data models inspired by statistical physics.
• Quantify the relationship between data structure and learned features.
• Develop synthetic datasets to benchmark interpretability tools.
• Own projects end-to-end and mentor junior researchers.

Skills

Neural networks

Python

Remote work

Communication

Physics/Math background

Education

PhD or equivalent research experience in math/physics/ML/CS/neuroscience/complexity

Job description

PrincInt is hiring Research Scientists to work on Ambitious Mechanistic Interpretability (AMI) with fully remote setup. Preference for time spent in London or Bay Area; visa sponsorship available for London/Berkeley candidates.

Start date January 2027, salary USD 125k–170k gross annually. The role focuses on building data models inspired by physics, quantifying data structure relationships, and developing synthetic datasets to benchmark interpretability tools, with mentorship of junior