Mechanistic AI Interpretability Scientist

Anthropic; San Francisco (CA); Hybrid; USD 350,000 - 850,000; Full time

Job summary

Anthropic is seeking researchers and engineers for the Interpretability team to reverse-engineer how trained models work and develop mechanistic understanding. The role focuses on methods to understand LLMs, running robust experiments, and constructing interpretable circuits using features.

Based in San Francisco, exceptional candidates may be considered for remote work. This position offers visa sponsorship and collaboration with Alignment Science and Societal Impacts to enhance model safety.

Qualifications

• Strong track record of scientific research and interpretability work.
• Ability to collaborate across teams and communicate results clearly.
• Comfortable with experimental research and exploring new directions.
• Proficient in Python for research and tooling.

Responsibilities

• Develop methods for understanding LLMs by reverse engineering algorithms learned in their weights
• Design and run robust experiments, scalable to large models
• Create and analyze interpretability features and circuits to understand model behavior
• Build infrastructure for running experiments and visualizing results
• Communicate results internally and publicly

Skills

Scientific research

Interpretability

Team science

Experimental mindset

Coding & experiments

Python

Education

Bachelor's degree or equivalent

Job description

Anthropic is seeking researchers and engineers for the Interpretability team to reverse-engineer how trained models work and develop mechanistic understanding. The role focuses on methods to understand LLMs, running robust experiments, and constructing interpretable circuits using features.

Based in San Francisco, exceptional candidates may be considered for remote work. This position offers visa sponsorship and collaboration with Alignment Science and Societal Impacts to enhance model safety.