Tech Stack

Description

You will join Anthropic's Interpretability team to reverse-engineer how trained neural networks work, focusing on mechanistic interpretability to make AI systems safe and trustworthy. Your work involves discovering how neural network parameters map to meaningful algorithms.

Requirements

• Strong background in neural networks and interpretability
• Ability to conduct research in mechanistic interpretability
• Experience with publications in interpretability or related fields
• Proficiency in programming and model analysis

Responsibilities

• Reverse-engineer how trained neural networks work
• Develop mechanistic understanding of AI systems
• Apply interpretability methods to improve AI safety
• Build tools for analyzing neural network internals