The Role
Design and implement scientifically valid automated evaluations of frontier AI systems, focusing on impacts to user wellbeing, autonomy, mental health, and decision-making. Develop user simulators, LLM-as-a-judge pipelines, evaluation rubrics, and population-specific methods. Collaborate with research engineers, scientists, frontier AI labs, regulators, domain experts, and affected communities to improve evaluation realism, measurement, and oversight.
Summary Generated by Built In
Salary range: $250,000 - $450,000/year + benefits
Description: Transluce is a fast-moving nonprofit research lab building the public tech stack for AI evaluation and oversight. We are pioneering research into the behaviors of AI chatbots and their effect on user wellbeing, and we’re improving outcomes for millions of sensitive AI interactions with vulnerable users.
About the role: As an AI Behavior Researcher, you will lead projects to design and develop automated evaluations of frontier AI systems that are technically sophisticated, scientifically valid, and concretely impactful. This includes expanding on our existing evaluation pipelines to conduct novel analyses of AI behaviors that affect the autonomy and wellbeing of specific user groups (e.g., children or users located in countries beyond the United States).
As an early member of a highly collaborative team, you will learn and grow quickly, and work directly with frontier labs to improve AI evaluations design, with regulators to improve independent oversight of AI, and with domain experts and affected populations to enhance the realism and relevance of our evaluations.
Core responsibilities:
- Develop novel, valid automated evaluations of AI’s impacts on users, including their mental health and decision making.
- Write code to implement and run automated evaluations, such as user simulators or LLM-as-a-judge pipelines.
- Design methods to improve the ecological validity and realism of automated evaluations for specific populations, such as customizing existing user simulation methods to capture the vocabulary used by children.
- Write and revise judge rubrics to evaluate model behaviors related to user wellbeing and decision making, systematizing abstract, socially situated concepts into clear measurement criteria.
- Collaborate with scientists and research engineers to productionize best practices in AI behavioral evaluation.
Minimum qualifications:
- Expertise on quantitative generative AI evaluation and measurement. Good intuition about how to systematize and operationalize complex social concepts.
- Relevant experience designing and validating automated AI evaluation methods, such as LLM-as-a-judge systems or multi-turn benchmarks.
- Proficiency in Python to implement analysis and evaluation tooling.
- Meticulous, good experimental design, epistemic self-awareness and transparency.
- Ability to balance between the needs of AI researchers and domain experts, as well as between researchers and senior decision makers.
- Strong communication skills, low ego, openness to giving and receiving feedback.
Preferred qualifications (not required):
- Experience running automated evaluations at scale or in a production context.
- Experience conducting controlled human subjects experiments to validate automated evaluation methods.
- Experience in customer-facing, consulting, or forward-deployed roles translating ambiguous stakeholder needs into concrete deliverables.
- Experience or training in human-centered design or HCI research methods, including working with domain experts or impacted communities.
- Experience or demonstrated interest in studying AI’s psychological or social impacts, such as for crisis support, manipulation or sycophancy, political persuasion, or displacing human relationships.
- Experience designing multilingual generative AI evaluations.
- Experience and comfort using AI coding agents at work.
We are hiring at all levels of experience and would encourage those enthusiastic about the role who do not meet all of the qualifications to apply. We are located in San Francisco and excited to work together in-person. We are open to sponsoring international visas.
Skills Required
- Expertise in quantitative generative AI evaluation and measurement
- Ability to systematize and operationalize complex social concepts
- Experience designing and validating automated AI evaluation methods, such as LLM-as-a-judge systems or multi-turn benchmarks
- Proficiency in Python for analysis and evaluation tooling
- Strong experimental design skills
- Meticulousness, epistemic self-awareness, and transparency
- Ability to balance the needs of AI researchers, domain experts, and senior decision makers
- Strong communication skills, low ego, and openness to feedback
- Experience running automated evaluations at scale or in production
- Experience conducting controlled human subjects experiments to validate automated evaluation methods
- Experience in customer-facing, consulting, or forward-deployed roles
- Experience or training in human-centered design or HCI research methods
- Experience or demonstrated interest in AI psychological or social impacts
- Experience designing multilingual generative AI evaluations
- Experience using AI coding agents at work
Am I A Good Fit?
Get Personalized Job Insights.
Our AI-powered fit analysis compares your resume with a job listing so you know if your skills & experience align.
Success! Refresh the page to see how your skills align with this role.
The Company
What We Do
Transluce is an independent research lab that builds open, scalable technology for understanding AI systems and steering them in the public interest. Transluce means to shine light through something to reveal its structure. Today’s complex AI systems are difficult to understand—not even experts can reliably predict their behavior once deployed. Given AI's extraordinary consequences on society, we need scalable and open analyses of the capabilities and risks of AI systems. We are building open source, AI-driven tools to understand and analyze AI systems. We will apply these tools to open-weight models, so the world can vet our analyses and improve their reliability. Once our technology has been vetted, we will work with frontier AI labs and governments to ensure that internal assessments reach the same standards as our publicly vetted procedures. Email: [email protected]
