Research Scientist, Interpretability

Details

Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. The Interpretability team focuses on mechanistic interpretability: reverse-engineering how neural network parameters map to meaningful algorithms. The team studies neural network components (neurons, attention heads, features, circuits) and builds tools and experiments—ranging from toy models to large-scale analyses—to discover mechanisms that explain model behavior and improve model safety.

Responsibilities

• Develop methods for understanding large language models (LLMs) by reverse engineering algorithms learned in model weights
• Design and run robust experiments both in toy scenarios and at scale in large models
• Create and analyze interpretability features and circuits to better understand how models compute
• Build infrastructure for running experiments and visualizing results
• Communicate results clearly with colleagues and publicly (writing up results, presentations)

Requirements

• Strong track record of scientific research (in any field); some prior work on interpretability is expected
• Comfortable with messy experimental science and iterative investigation
• Research and engineering skills: every team member writes code, designs and runs experiments, and interprets results
• Familiarity with Python is required
• Education: at least a Bachelor's degree in a related field or equivalent experience

Role-specific location policy

• This role is based in the San Francisco office; Anthropic is open to considering exceptional candidates for remote work on a case-by-case basis
• Location-based hybrid policy: currently, staff are expected to be in one of Anthropic's offices at least 25% of the time

Compensation

• Annual Salary: $315,000 - $560,000 USD
• Total compensation package for full-time employees includes equity, benefits, and may include incentive compensation

Logistics

• Visa sponsorship: Anthropic does sponsor visas for some roles and will make reasonable efforts to assist when an offer is made
• Applicants are encouraged to apply even if they do not meet every qualification listed

How we work / Culture

• The team values collaborative, high-impact research that treats AI research as an empirical, large-scale scientific effort
• Strong communication skills and willingness to publish and share research are emphasized

Additional notes

• The role references work and publications related to transformers, transformer circuits, and mechanistic interpretability. Candidates should be prepared to engage with both theory and engineering aspects of interpretability research.