Research Scientist Control

Research Scientist Control Overview

Join Apollo Research as a Research Scientist in Control, where you will be part of a pioneering team focused on enhancing AI safety through empirical research. This role is integral to the development of our monitoring tool, Watcher, aimed at mitigating risks associated with AI systems. You will collaborate closely with a dedicated team of researchers and engineers, contributing to the transformation of theoretical AI control research into practical applications.

Key Responsibilities

• Design and implement control protocols for AI systems, testing them on real-world production environments.
• Conduct experiments to evaluate the effectiveness of monitoring systems across various failure modes and agent behaviors.
• Develop and maintain frameworks for assessing progress in monitoring capabilities.
• Create and curate high-quality datasets for training and testing monitoring systems.
• Iterate on monitoring strategies based on empirical findings, focusing on balancing detection accuracy with computational efficiency.
• Contribute to a shared library of coding agent failure modes as they are identified during the design and evaluation of monitoring systems.
• Stay updated on the latest research regarding AI safety, agent failures, and detection methodologies.
• Engage with research on coding security and safety vulnerabilities.
• Fine-tune open-source models to develop efficient monitors suitable for high-volume production settings.
• Design and implement autonomous monitoring systems that analyze logs to detect both known and new failure modes.
• Integrate insights from red-teaming exercises into the design and evaluation of monitoring systems.

Job Requirements

• A minimum of 2 years of experience in empirical research involving large language models or AI systems.
• Proven experience with AI coding agents, including extensive use and comparison of leading coding agents.
• Familiarity with LLM-as-a-judge frameworks or broader AI monitoring practices.
• Experience in designing and executing experiments, analyzing outcomes, and refining approaches based on empirical data.
• Strong proficiency in Python programming.
• Ability to work independently on complex research challenges.

Benefits

• Competitive salary and equity options.
• Salary range: $204,000 – $385,000 in San Francisco; £136,000 – £258,000 in London, with plans for future salary increases.
• Flexible working hours and schedules.
• Unlimited vacation and sick leave.
• Up to 6 months of paid parental leave.
• Comprehensive health, dental, and vision insurance.
• Retirement savings plan with competitive employer matching.
• Free meals and snacks provided during workdays.
• Paid travel for work-related events, including retreats and conferences.
• Annual professional development budget of $1,000.
• Relocation assistance and coverage for visa fees if applicable.

Working Pattern

This position is based onsite in either our London or San Francisco office, with flexible working hours available.

Visa Sponsorship

Apollo Research sponsors visas in both the UK and US. While sponsorship is not guaranteed for every role or candidate, we will assist you in finding the appropriate visa route if you receive an offer.

How to Apply

Interested candidates should complete the application form along with their CV. A cover letter is not required, but links to relevant work samples are welcome. The interview process includes a screening interview, a take-home test, three technical interviews, and a final interview with the CEO. We do not conduct traditional coding interviews, and you may use AI tools during the take-home test.

Degree Requirement: Not Specified

Visa Sponsorship Promising