Evals Software Engineer

Apollo Research is seeking an Evals Software Engineer to join their team in London. The role focuses on building and maintaining software for AI safety evaluations, particularly related to deceptive alignment. Key responsibilities include extending internal libraries for language model evaluations, collaborating with researchers, and advocating for good software design practices. The ideal candidate should have at least 2 years of experience in Python development and a strong interest in AI safety.

The Evals team at Apollo Research works on conceptual safety cases, building evaluations for deceptive alignment properties, conducting evaluations on frontier models, and creating model organisms to demonstrate behaviors related to deceptive alignment. The role offers an opportunity to work on cutting-edge AI safety research and development.

Apollo Research aims for a culture emphasizing truth-seeking, goal-orientation, constructive feedback, and helpfulness. They welcome applicants from all backgrounds and offer a range of benefits including private medical insurance, flexible work hours, unlimited vacation, and a professional development budget.

The position is based in London, with the possibility of visa sponsorship for international candidates. The interview process includes multiple stages, focusing on practical skills related to the job rather than general coding challenges. Early applications are encouraged as reviews are conducted on a rolling basis.