Member of Technical Staff - Post-Training (Reinforcement Learning & Generalization)

Role overview Research and develop post-training methods that help large language models acquire skills and apply them to unfamiliar tasks. You will design reinforcement learning experiments, training environments and evaluations that distinguish robust learning from memorization or reward exploitation, then turn effective approaches into reproducible recipes for production models.

Responsibilities - Design, run and analyze large-scale reinforcement learning post-training experiments, including approaches based on human feedback or verifiable rewards. - Develop methods to improve generalization across distributions, using reward design, varied tasks and environments, curricula or regularization. - Build held-out evaluations for unseen tasks, domains and formats, detecting benchmark overfitting and reward hacking. - Investigate training instability, excessive reward optimization and regressions in model capabilities. - Convert successful research into reliable, repeatable training procedures and collaborate with data infrastructure, evaluation and pre-training teams.

Requirements - Strong research and software engineering skills, including hands-on experience training large language models. - Applied reinforcement learning experience with language models; examples include PPO, GRPO, RLHF or RL from verifiable rewards. - Evidence of work on generalization, robustness or distribution shift, through research publications or deployed models. - Careful experimental practice, including controlled comparisons, useful ablations and sound interpretation of noisy results. - Strong understanding of data quality and how dataset characteristics can affect model training. - Clear communication skills.

Nice to have - Experience with distributed training systems such as JAX, PyTorch, FSDP or Megatron.

Benefits and work setup - Remote-friendly, with office options and no minimum in-office requirement; expected time-zone alignment may depend on the posting. - Benefits include health and dental coverage, mental health support, retirement contributions, parental leave top-up, six weeks of paid vacation, learning and enrichment funds, and a lunch stipend. Remote staff may also receive coworking and home-office support, plus travel funds for visiting other offices.