Physical Superintelligence (PSI) Fellowship

About PSI

Physical Superintelligence PBC is building AI systems that accelerate physics discovery. With roots at Harvard, MIT, Johns Hopkins, Oxford, the Institute for Advanced Study, and the Perimeter Institute, we’re assembling an exceptional team of physicists, AI researchers, and engineers.

We believe AI will transform scientific discovery, but physics presents unique challenges that require deep domain expertise to solve. We’re building the systems to overcome those challenges.

Fellowship Overview

The PSI Fellowship is designed for exceptional physicists who want to work at the frontier of AI-accelerated scientific discovery. We’re looking for researchers who can bridge rigorous physics thinking with modern AI capabilities—people with deep domain expertise and hands-on experience building AI systems.

Fellows will work directly with PSI’s founding team on core research challenges, with the goal of producing substantive contributions to our technical roadmap and, where appropriate, publishable research.

Program duration: 3–6 months full-time, depending on project scope and mutual fit

What to Expect

• Direct collaboration with PSI leadership and technical advisors on high-impact research

• Flexibility to work on projects aligned with your expertise—from benchmark development to agent architectures to physics evaluation frameworks

• Access to compute and tooling for AI-physics research

• Potential path to full-time roles for exceptional performers (though not guaranteed)

Research Areas

Fellows may work on projects spanning:

• AI-for-Physics Benchmarks: Developing rigorous evaluations that measure whether AI systems can genuinely reason about physics vs. pattern-match to solutions

• Physics Discovery Agents: Building and evaluating agentic systems that can formulate hypotheses, design experiments, and interpret results

• Verification & Interpretability: Creating frameworks to verify that AI-generated physics insights are sound and to understand how models represent physical knowledge

• Post-Training for Scientific Reasoning: Techniques for improving foundation models’ ability to reason about physics, including RLHF, constitutional methods, and domain-specific fine-tuning

• Human-AI Scientific Collaboration: Designing workflows where physicists and AI systems collaborate effectively on discovery

Ideal Candidates

Required:

• Physics PhD or advanced PhD candidate (all but dissertation)

• Demonstrated interest and hands-on experience with modern AI/ML (not just “AI-curious”—we’re looking for people who have built things)

• Strong programming skills (Python required; experience with deep learning frameworks preferred)

• Available for full-time commitment (3–6 months)

• Excited about the possibility that AI could fundamentally transform how physics is done

Strong candidates may also have:

• Experience with LLM fine-tuning, RLHF, or post-training methods

• Background in agentic AI systems or multi-step reasoning

• Published ML research or significant open-source contributions

• Experience with AI-assisted scientific workflows

• Track record of crossing disciplinary boundaries

Logistics

Location: Remote-friendly within the US, with optional in-person collaboration in Boston or San Francisco.

Work authorization: You must have US work authorization for the duration of the fellowship. We are not able to sponsor visas for fellowship positions. This role may involve access to technology or information subject to U.S. export control regulations. Employment eligibility and role assignments may be affected based on candidate status under applicable legal requirements.

Start dates: We are accepting applications on a rolling basis for start dates in 2026.

We review applications on a rolling basis and will respond within two weeks.

We care about depth of physics expertise, breadth of intellectual curiosity, and ability to ship.