Machine Learning Systems Engineer, Anthropic

• You want to build the cutting-edge systems that train AI models like Claude
• You're excited to work at the frontier of machine learning, implementing and improving advanced techniques to create ever more capable, reliable and steerable AI
• Have 2+ years of software engineering experience
• Like working on systems and tools that make other people more productive
• Are results-oriented, with a bias towards flexibility and impact
• Pick up slack, even if it goes outside your job description
• Enjoy pair programming (we love to pair!)
• Want to learn more about machine learning research
• Care about the societal impacts of your work

Desirable

• As an ML Systems Engineer on our Reinforcement Learning Engineering team, you'll be responsible for the critical algorithms and infrastructure that our researchers depend on to train models
• Your work will directly enable breakthroughs in AI capabilities and safety. You'll focus obsessively on improving the performance, robustness, and usability of these systems so our research can progress as quickly as possible
• You're energized by the challenge of supporting and empowering our research team in the mission to build beneficial AI systems
• Our finetuning researchers train our production Claude models, and internal research models, using RLHF and other related methods
• Your job will be to build, maintain, and improve the algorithms and systems that these researchers use to train models
• You’ll be responsible for improving the speed, reliability, and ease-of-use of these systems
• Profiling our reinforcement learning pipeline to find opportunities for improvement
• Building a system that regularly launches training jobs in a test environment so that we can quickly detect problems in the training pipeline
• Making changes to our finetuning systems so they work on new model architectures
• Building instrumentation to detect and eliminate Python GIL contention in our training code
• Diagnosing why training runs have started slowing down after some number of steps, and fixing it
• Implementing a stable, fast version of a new training algorithm proposed by a researcher
• Currently, we expect all staff to be in one of our offices at least 25% of the time. However, some roles may require more time in our offices