Key details
- Role type
- Full-time
- Compensation
- $600,000–$1,800,000/yr
- Work arrangement
- Remote
- Category
- technology
- Openings
- 1
About this role
Role Overview
Advance the post-training research that strengthens reasoning, generalization, and agent capabilities in large language and multimodal models. This highly experimental role connects foundational machine learning research with large-scale model development, giving you ownership of hypotheses, ambitious experiments, and scalable training methods.
Key Responsibilities
- Research, implement, and scale post-training methods including supervised fine-tuning, reinforcement learning, and on-policy distillation.
- Investigate learning objectives, optimization dynamics, exploration strategies, and credit-assignment mechanisms to improve training stability, sample efficiency, and capability acquisition.
- Develop training recipes by jointly optimizing data mixtures, sampling strategies, curricula, and evaluation methodologies.
- Create approaches for learning from expert demonstrations, human preferences, corrections, and model-generated trajectories, including identifying where human feedback has the greatest impact.
- Study the co-evolution of generators, reward models, and graders to improve model performance and the quality of training signals.
- Design rigorous experiments, ablation studies, and evaluation frameworks for generalization, reasoning, and realistic downstream-task performance.
- Lead research initiatives end to end, from identifying promising directions and implementing algorithms through large-scale experimentation and result validation.
Qualifications
- Strong foundation in machine learning, deep learning, optimization, and modern model-training techniques.
- Hands-on experience training, fine-tuning, or post-training large-scale language or multimodal models.
- Deep expertise in one or more of reinforcement learning, model distillation, reward modeling, data selection, or learning dynamics.
- Strong Python and PyTorch skills, including the ability to implement, debug, and optimize complex training pipelines.
- Experience designing controlled experiments, interpreting model behavior, and turning empirical findings into actionable research directions.
- Ability to independently navigate ambiguous research problems, form hypotheses, and iterate quickly using experimental evidence.
- Analytical, research-driven approach with an emphasis on reproducibility, scientific rigor, and measurable improvement.
- Experience with RLHF, RLAIF, or preference-based methods is a plus.
Work Terms
- Full-time, remote role.
Compensation
- Annual compensation range: $600, 000 to $1, 800, 000.