14 days ago
Seattle, WA, USA +2 moreMid Level
H1B Sponsor
Base Salary
$500k - $850k/yr
Responsibilities
- Build, maintain, and improve reinforcement learning and fine-tuning algorithms and systems used to train AI models.
- Improve the speed, reliability, robustness, and usability of model-training systems.
- Profile training pipelines and diagnose and fix performance degradation or other problems.
- Build systems that launch test training jobs and detect issues in training pipelines.
- Adapt fine-tuning systems for new model architectures.
- Implement instrumentation to detect and eliminate Python GIL contention.
- Implement stable and fast versions of new training algorithms proposed by researchers.
Requirements
- 4+ years of software engineering experience.
- Experience or interest in systems and tools that improve researcher productivity.
- Strong candidates may have experience with high-performance, large-scale distributed systems.
- Strong candidates may have experience with large-scale LLM training.
- Strong candidates may have experience with Python.
- Strong candidates may have experience implementing LLM fine-tuning algorithms such as RLHF.
- Bachelor’s degree or an equivalent combination of education, training, and/or experience in a relevant field or demonstrated through relevant coursework, training, or professional experience.
Benefits
- Competitive compensation and benefits.
- Optional equity donation matching.
- Generous vacation and parental leave.
- Flexible working hours.
- Office space for collaboration.
- Hybrid policy currently expects staff to be in an office at least 25% of the time.
- Visa sponsorship may be available, with immigration lawyer support.
Tech Stack
Categories
About Anthropic
We're an AI research company that builds reliable, interpretable, and steerable AI systems. Our first product is Claude, an AI assistant for tasks at any scale. Our research interests span multiple areas including natural language, human feedback, scaling laws, reinforcement learning, code generation, and interpretability.