Research Engineer / Research Scientist, RL Frontiers
$500k - $850k • Remote • San Francisco, CA | New York City, NY | Seattle, WA
Posted 18d ago
Job Location
San Francisco, CA | New York City, NY | Seattle, WA
Tech Stack
Remote Work Policy
Fully remote
Categories
AI Research Engineer
About the job
Anthropic is dedicated to building AI systems that are safe, reliable, and steerable, aiming to benefit both users and society. This role is central to advancing Reinforcement Learning (RL), the core technology enabling Claude's reasoning, coding, and autonomous capabilities. The RL Scaling team focuses on how RL performance, stability, and learning efficiency evolve with increasing model size, episode length, and compute power. You will be instrumental in developing and scaling next-generation RL algorithms and architectures, from initial research to large-scale deployment, while deeply understanding and addressing the challenges that arise at scale. This position bridges research and engineering, requiring you to build the systems that accelerate iteration and experimentation, and to tackle complex, unsolved problems in Anthropic's largest RL runs.
Responsibilities
- Analyze how RL training and sampling scale with model size, context length, and compute, identifying algorithmic and systems improvements for efficiency.
- Develop and optimize next-generation model architectures and RL algorithms for frontier-scale execution.
- Scale promising small-scale research results to frontier-scale runs, diagnosing and resolving discrepancies in behavior.
- Build experimental infrastructure to accelerate research velocity and enable reproducible comparisons of architecture and algorithm variants at scale.
- Oversee the end-to-end performance of large-scale RL runs, from research code to hardware.
- Develop performance and cost models for proposed architecture and algorithm changes to guide scaling decisions.
- Investigate and trace root causes of training dynamics issues at scale, such as instabilities and divergence.
Requirements
- Deep understanding of modern transformer language models, including architecture, training dynamics, and large-scale optimization.
- Hands-on experience training large models in a distributed setting, with knowledge of data, tensor, and pipeline parallelism tradeoffs.
- Demonstrated original technical contributions in ML training or systems (e.g., new methods, architectures, optimizations) through research, open-source, or production impact.
- Proficiency in designing rigorous large-scale experiments with appropriate baselines, ablations, and statistical rigor.
- Ability to quantitatively analyze compute, memory, and communication costs of models and algorithms.
- Strong programming skills in Python and JAX or PyTorch, with comfort across the full software stack.