Member of Technical Staff - Mechanistic Interpretability

What you'll do

• Extract training signals from language-model internal states and convert representations, features, circuits, and causal behaviors into intrinsic rewards.
• Compare interpretability-derived rewards with other reward sources and assess their quality, generalization, robustness, and resistance to reward hacking.
• Study representations during RL and post-training, build reproducible experimental infrastructure, and pursue a research agenda in open-ended learning.

What they ask for

• PhD or equivalent experience in machine learning, reinforcement learning, or a closely related field.
• Research excellence demonstrated through publications, open-source work, deployed AI systems, or other substantial technical contributions.
• Deep understanding of modern machine learning, especially reinforcement learning, representation learning, and large language models; strong familiarity with LLM post-training methods.
• Experience running rigorous ML experiments and expertise with Python and at least one major ML framework, such as PyTorch or JAX.

Nice to have

• Experience with mechanistic interpretability techniques, language-model agents, scalable RL infrastructure, reward models, verifiers, process supervision, or automated evaluation systems.
• Demonstrated software engineering ability and ability to communicate technical results to research and non-research audiences.

Pay

$300,000 - $500,000 USD (base salary)

Tools

Python, PyTorch, JAX

AI focus

Uses mechanistic interpretability to derive intrinsic rewards from language-model internals for reinforcement learning, improving training beyond imitation of human expertise.