Conceptual Reasoning Researcher (Contract)
$0k - $0k • Remote • Remote • Contract
Posted 3d ago
Remote Work Policy
Fully remote
Employment Type
Contract
Categories
AI Research Engineer
About the job
Trajectory Labs is seeking researchers to design evaluations for conceptual reasoning in AI safety, focusing on complex, open-ended problems in AI governance, alignment, macrostrategy, and economics. These evaluations will be paired with rubrics to distinguish genuine reasoning from superficial responses. The role involves working 20-40+ hours per week, either remotely or on-site in Berkeley. As an early-stage startup, responsibilities and priorities may evolve rapidly. The company's mission is to automate AI safety work through AI models, building safety and alignment evaluations, red-teaming programs, and RL environments for leading AI labs.
Responsibilities
- Identify challenging conceptual questions in AI safety (alignment, macrostrategy, economics, governance) where current frontier models exhibit poor reasoning.
- Develop each question into an evaluation task, including a well-defined problem and a detailed rubric that rewards strong reasoning and penalizes weak responses.
- Execute these tasks on frontier models and iterate on the rubrics to effectively differentiate between high-quality and low-quality answers.
- Engage in detailed quality assurance processes, with tasks undergoing review before finalization.
Requirements
- Original analysis in AI risk, alignment, macrostrategy, economics, or governance.
- Ability to articulate and encode criteria for good reasoning into a rubric.
- Capacity to revise judgments based on strong counterarguments and defend them with sound reasoning.
- Proficiency in working with frontier models and agentic tools, distinguishing correct output from plausible-sounding but incorrect output.
- Domain expertise in AI governance (compute governance, international agreements, AI law), AI safety & alignment (theory, control, evaluations, AI welfare), philosophy & macrostrategy (ethics, AI moral status, takeoff dynamics), or economics & decision theory (growth theory, AI's economic effects, game theory, decision theory).
Benefits
- $80–150/hr compensation
- Contract length of ~3 months with potential for renewal
- Flexible hours (20-40+ hours/week)
- Remote work option
- Desk space available at Constellation in Berkeley