SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Physical Superintelligence is building AI systems to discover new physics at scale. The company was founded by researchers from Google, NVIDIA, Harvard, Meta, MIT, Oxford, Johns Hopkins, Cambridge, and the Perimeter Institute. The mission is to discover and commercialize transformative physics breakthroughs using artificial superintelligence, safely and verifiably.
As a Member of Technical Staff in AI Research, you will conduct original research on agents and training systems that learn to do physics, with an expectation to publish work that serves the mission. You will build what you research—training runs, evaluations, and harnesses that ship as working systems rather than isolated research artifacts. You will go deep in one specialized lane while collaborating across others: reinforcement learning and post-training; agents, evaluation and verification; model architectures and training dynamics; or scientific and generative modeling. You will work closely with physicists designing verification harnesses and engineers building training infrastructure.
The company emphasizes high technical standards, full ownership from spec to ship to on-call support, and an AI-native development process using agentic coding tools daily. They favor simple designs that ship over clever ones that do not, and prioritize research depth, experimental rigor, systems fluency, and shipping velocity.
REQUIREMENTS:
- Three or more years of research experience with recent publications at top venues (NeurIPS, ICML, ICLR, or comparable physics-ML venues). You must be able to defend your results and identify claims 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, and experiments on multi-GPU clusters. You write research code that others reuse, and at least one system you built has beaten a non-trivial baseline and shipped as a working artifact (not a notebook).
- Depth in at least one of the following lanes: (1) Reinforcement learning and post-training (policy optimization, reward design, RLHF, RLVR, GRPO, RL for language models and control); (2) Agents, evaluation and verification (tool use, long-horizon reasoning, memory, harnesses, evals, verifiers, reward models, instrumentation); (3) Model architectures and training methods (transformers, state-space and graph models, neural operators, surrogate architectures for physical fields, scaling laws, optimization, training dynamics); or (4) Scientific and generative modeling (learning from simulators, physics-informed models, out-of-distribution behavior, diffusion, generative models, code generation).
NICE TO HAVE:
- Physics or mathematics background enabling substantive conversations with domain physicists.
- Distributed training at cluster scale with understanding of what breaks at each step.
- Experience applying agents to simulators, scientific tools, games, or rigorous benchmark suites.
- Open-source contributions or research artifacts adopted by the community.
