Member of Technical Staff, AI Research

Physical Superintelligence is a startup with roots at Google, NVIDIA, Harvard, Meta, MIT, Oxford, Johns Hopkins, Cambridge, and the Perimeter Institute building AI systems to discover new physics at scale. We are seeking AI researchers to build the agents and training systems that learn to do physics.

Our mission is to discover and commercialize transformative physics breakthroughs at scale with artificial superintelligence, safely, verifiably, and for broad public benefit.

The last century's golden age of physics gave us transistors, lasers, and nuclear energy. We believe artificial superintelligence will unlock the next one. We're creating the infrastructure to industrialize scientific discovery and usher in this new era.

• →

  Do original research on agents and training systems that learn to do physics, and publish it where it serves the mission.

• →

  Build what you research: training runs, evaluations and harnesses that ship as working systems, not isolated artifacts.

• →

  Go deep in one lane: reinforcement learning and post-training; agents, evaluation and verification; model architectures and training dynamics; scientific and generative modeling.

• →

  Work with the physicists who design verification harnesses and the engineers who build training infrastructure.

• Three or more years of research experience, with recent publications at top venues (NeurIPS, ICML, ICLR, or comparable physics-ML venues). You can defend the result and you can name the claim you would retract.

• Three or more years of hands-on research engineering with the modern ML stack: PyTorch or JAX, training and post-training frameworks, inference engines for rollouts, experiments run on multi-GPU clusters. You write research code that others reuse, and at least one system you built this way beat a non-trivial baseline and shipped as a working artifact rather than a notebook.

• Depth in at least one of the following. We hire for one lane and expect you to work across the others:

  • Reinforcement learning and post-training: policy optimization, reward design, RLHF, RLVR, GRPO, RL for language models and for control.

  • Agents, evaluation and verification: tool use, long-horizon reasoning, memory and harnesses, evals, verifiers, reward models, instrumentation that separates reasoning from gaming.

  • Model architectures and training methods: transformers, state-space and graph models, neural operator and surrogate architectures for physical fields and meshes, scaling laws, optimization, training dynamics.

  • Scientific and generative modeling: learning from simulators, physics-informed and conservation-aware models, out-of-distribution behaviour, diffusion and generative models, code generation with checkable results.

• A physics or mathematics background: you can hold a substantive conversation with a domain physicist.

• Distributed training at cluster scale: single GPU to cluster without rewriting, and what breaks at each step.

• Experience applying agents to simulators, scientific tools, games, or rigorous benchmark suites.

• Open-source contributions or research artifacts the community has adopted.

We hold a high technical bar and give people full ownership of their work, from spec to ship to on-call. We write contracts before logic, test against real systems instead of mocks, and favor simple designs that ship over clever ones that do not. Our development process is AI-native: we work with agentic coding tools daily, write specs that are legible to humans and agents alike, and lead with leverage.