AI Research Scientist, Recursive Self Improvement, AI Safety and Reinforcement Learning

Research self-improving training loops for recursive self-improvement (RSI) under explicit governance, kill switches, and human oversight. Build theory- and systems-grounded evaluations to detect capability drift, Goodhart effects, and distributional shift in closed-loop training. Partner with RL scientists on intersections with policy optimization and preference learning, define red-team and monitoring protocols, and publish technical findings aligned with responsible deployment standards.