Member of Technical Staff, Research

About FirstPrinciples: FirstPrinciples is a non-profit organization building an autonomous AI Physicist to understand the nature of reality: the underlying structure, governing principles, and fundamental laws of our universe. We're developing an intelligent system that can explore theoretical frameworks, reason across disciplines, and generate novel insights to tackle the deepest unsolved problems in physics. By combining AI, symbolic reasoning, and autonomous research capabilities, we're developing a platform that goes beyond analyzing existing knowledge to actively contribute to physics research. Our goal is to accelerate progress on the questions that have captivated humanity for centuries. We operate as a global nonprofit organization , with a Canadian foundation, a US-based 501(c)(3). Job Description: We are looking for a Member of Technical Staff, Research to investigate, design, test and develop state of the art (SOTA) methods and applications, which can be integrated into the broader AI engine FirstPrinciples is developing. You will collaborate with cross-functional teams and your work will flow straight into production, helping advance the way scientific research is performed. Your work will impact the wider academic community through the development of unique solutions to usher in a new era of scientific discovery. The ideal candidate has a proven track record in AI research, who can combine strategic thinking with technical depth to bring complex ideas to life. Key Responsibilities: Fundamental Model Research: Research, design, and test novel, research‑specific model architectures that integrate academic literature, natural language processing (NLP), symbolic reasoning, and other methods to orchestrate the scientific process. Prototype and build custom tokenizers for LaTeX symbols and physical units to be treated as tokens. Explore alternatives to transformers through in-depth research and provide practical recommendations for model development. Develop reinforcement-learning loops to enable models to run independent and internal thought experiments. Multimodal Data & Benchmarking: Design and automate data ingestion pipelines in collaboration with our Data Scientists & Engineers that aggregates science literature, metadata, experimental data, equations and other data sources in a robust and scalable manner. Establish custom benchmarks to assess the models’ understanding of physical concepts, mathematical reasoning abilities, and ability to minimize hallucinations for the benefit of scientific reliability. Refine and release datasets and baselines once internal tests are stable. Training, Testing & Safety: Run and track model training jobs while leading the technical team through set-up, monitoring progress, and constraining costs within budget. Develop approaches to stage “practice runs” in a sandbox environment to develop the model’s abilities to explore ideas independently while logging results for later review. Develop a framework to evaluate the models’ learning using visual and statistical tools to spot patterns and blind spots. Add guard-rails and tests that flag poor quality model output. Maintain internal tools to track lists of known issues, noting failures, clear fixes, and improvements to be integrated into future development Collaboration & Technical Guidance: Work with the engineering team to ensure product feasibility and robust architecture. Translate technical trade-offs to non-technical stakeholders in clear terms. Present findings in clear updates to the technical team in order to keep the broader team appraised of progress against research milestones.