Scientific Lead Embedded Assessments
Independent nonprofit AI research lab building open technology for understanding and scalable public oversight of AI systems.
About Transluce
Transluce develops research, platforms, and tools to help people understand, evaluate, and steer advanced AI systems in the public interest. Its active offerings include Docent, a behavior-analysis platform for AI-agent transcripts, and Monitor, an AI-driven interface for observing and steering model computations.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will scope, design, and lead assessments conducted by teams of evaluators embedded at frontier AI labs. You will audit existing safety evaluations, identify coverage gaps and open model-safety questions, design and execute experiments, and lead the scientific content of public assessment reports. Between engagements, you will conduct practice evaluations on public models, train scientific staff, and improve evaluation tooling.
Requirements
- Excellent track record of scientific research, ideally adjacent to AI alignment
- Experience building AI evaluations, including benchmarks, evaluation suites, or red-teaming work
- Experience leading or managing researchers
- Ability to prioritize and deliver high-quality results under uncertainty and tight deadlines
- Technical writing and communication skills for technical and general-public audiences
- Ability to quickly absorb technical context from code and other technical artifacts
Responsibilities
- Refine the scientific scope of assessments based on information gained while embedded
- Review completed experiments and evaluations to audit coverage
- Design and help execute experiments that evaluate safety practices
- Lead the scientific input for final embedded-assessment reports
- Lead practice evaluations on public models
- Train scientific staff
- Improve evaluation tooling with engineering staff