jobs
Research Engineer, Interpretability
Anthropic’s Interpretability team works to reverse-engineer how trained neural networks operate, with the goal of developing a mechanistic understanding that can contribute to the safety of advanced…
Auf Englisch
- Company
- Anthropic
- Location
- San Francisco, CA, USA
- Status
- Closed
- Posted
- 2026-11-18 03:34 UTC
Anthropic is hiring a research engineer for its Interpretability team, which studies how trained neural networks work internally and supports safety audits of frontier models. The role involves building and maintaining training and inference infrastructure, including instrumented passes, activation extraction, and steering vectors; profiling and optimizing for scale; and creating researcher-facing tools. The posting asks for five to ten or more years of software experience, strong programming and Python skills, and collaboration with researchers. Useful extras include distributed systems, transformer models, PyTorch or CUDA, and JAX or XLA.
Original job posting