The role
Turn open questions about agent learning into experiments we can run, reproduce, and learn from. Your work will shape training environments across long-horizon coding, AI4Science, AI4AI, knowledge work, and tool use.
What you’ll work on
- Design RL experiments to study task difficulty, curricula, reward design, and generalization.
- Develop verifiers and evaluations that separate useful capability gains from reward exploitation.
- Investigate long-horizon learning and AI research workflows, including recursive self-improvement.
- Work with engineers to turn findings into better environments, datasets, and reproducible research.
What you bring
- Hands-on experience with reinforcement learning, model post-training, or agent evaluation.
- Strong Python and experience implementing experiments in PyTorch or a similar framework.
- Sound experimental judgment: clear baselines, controlled comparisons, and careful failure analysis.
- Ability to explain your results and build research code that others can run and extend.
Show us your work
Share a paper, experiment, benchmark, or codebase you’re proud of. Tell us what you tried, what failed, and what you learned.