AI Research Scientist

About the role

Suited to a current or former ML research engineer, scientist or applied scientist who can judge what stumps a frontier model. Most of the thinking goes into task design, not model training.

At a glance

What you will do

• Write machine learning engineering tasks for a long-horizon research project
• Make each task hard enough that frontier models solve it at most twice in eight tries
• Work on inference, evaluation and systems problems that run on both GPU and CPU
• Explain in clear terms why a given problem is difficult for a frontier model

What they look for

• Experience training, fine-tuning or optimising models in PyTorch or JAX
• Depth in one of: LLMs, GPU inference and performance, reinforcement learning, or scientific ML
• Ability to say plainly what makes a problem hard for a frontier model

Nice to have

• Experience writing benchmark or evaluation questions