Research Scientist (AI)

Physical Superintelligence•San Francisco, CA

1d•Onsite

About The Position

Physical Superintelligence is a startup with roots at Harvard, MIT, Johns Hopkins, Oxford, the Institute for Advanced Study, and the Perimeter Institute is building AI systems to discover new physics at scale. We are seeking AI researchers to develop reinforcement learning agents and training systems for scientific discovery.

Requirements

Experience building agents and training models with reinforcement learning.
Proficiency in modern machine learning frameworks.
Understand distributed training systems.
Track record shipping working AI systems.
Hands-on experience with modern reinforcement learning algorithms including PPO, SAC, MuZero, and multi-agent self-play and other methods
Proficiency with PyTorch or JAX, distributed training using Ray, XLA, or Accelerate, and modern pretraining workflows

Nice To Haves

Physics or mathematics background providing intuition for physical reasoning and mathematical modeling
Experience applying agents to simulators, games, scientific tool use, or benchmark design with rigorous experimental methodology

Responsibilities

Build and train AI systems for physics discovery, working with physicists who design verification harnesses and engineers who build training infrastructure.
Focus on core AI research questions including how agents learn physics reasoning, action space design for scientific discovery, reward structure development, and training systems that scale.
Build and train reinforcement learning agents using modern approaches including PPO, SAC, MuZero, and multi-agent self-play and other methods
Design agent architectures for physics reasoning and scientific tool use
Implement training curricula and reward structures for discovery tasks
Develop evaluation workflows and benchmarks for physics reasoning capabilities
Build instrumentation to understand agent behavior and learning dynamics
Collaborate with physicists and engineers on system design and architecture