FirstPrinciples Research Fellow

FirstPrinciples is a non-profit organization building an autonomous AI Physicist to understand the nature of reality. The Research Fellow will own a well-scoped research direction that contributes directly to the AI Physicist's ability to reason about physics, working on novel model architectures and collaborating with engineering teams to implement state-of-the-art methods into production.

Responsibilities:

• Research, design, and test novel model architectures that combine academic literature, NLP, symbolic reasoning, and structured scientific workflows
• Prototype and build embedding representations for physical concepts, mathematical objects, and logical structures, enabling models to reason over equations, abstractions, and scientific constraints rather than surface text alone
• Investigate alternatives to transformer-based architectures and deliver concrete recommendations
• Design and run targeted experiments to evaluate new architectural ideas, using empirical results to guide the development of next-generation model architectures
• Develop reinforcement learning loops that enable models to run internal and independent thought experiments
• Design and automate scalable data ingestion pipelines that aggregate scientific literature, metadata, equations, and experimental data
• Create custom benchmarks to measure physical understanding, mathematical reasoning, and failure modes in scientific reasoning and abstraction
• Refine and release curated datasets and baselines once internal validation is complete
• Run and track model training jobs while managing compute usage and budget constraints
• Design sandbox environments for controlled autonomous exploration
• Build evaluation frameworks using visual and statistical tools to identify strengths and blind spots
• Implement tests and guardrails that flag low-quality or unsafe outputs
• Maintain internal issue tracking with clear failure modes and fixes
• Work closely with engineers to ensure research is feasible and production-ready
• Communicate technical trade-offs clearly to non-technical stakeholders
• Present regular research updates tied to defined milestones