Member of Technical Staff, Mechanistic Interpretability

Responsibilities

• Design and execute experiments to uncover features, circuits, and mechanisms driving model behavior
• Build infrastructure for mechanistic interpretability, including activation analysis and causal interventions
• Investigate how internal representations relate to downstream performance, reasoning, and robustness
• Develop new techniques to improve evaluation, reliability, safety, and model design
• Study how multimodal genome language models represent biological concepts and abstractions
• Collaborate with teams in model architecture, training, systems, safety, and biology

Requirements

• Strong background in machine learning research (LLMs, representation learning, or mechanistic interpretability)
• Deep understanding of modern deep learning architectures like transformers
• Proficiency in Python and PyTorch
• Strong experimental skills and scientific judgment
• Excellent written and verbal communication skills

Preferred Qualifications

• Experience with sparse autoencoders, activation patching, or circuit discovery
• Research experience in AI safety, alignment, or model evaluations
• Experience with large-scale training systems or distributed computing
• Background in computational biology, genomics, or neuroscience
• Contributions to open-source ML research or publications in interpretability/AI safety

About the Company

Radical Numerics is an AI research lab building general biological intelligence. The team created Evo, a pioneer in generative genomics, and focuses on bringing rigor from distributed systems and numerics to the challenges of biology to reduce human suffering.