Physical Superintelligence - Member of Technical Staff, AI Research
Requirements
• PhD in machine learning, computer science, physics, mathematics, or a related quantitative field, with a track record of recent publications at top venues (NeurIPS, ICML, ICLR, or comparable physics-ML venues). You have produced original research that the community recognizes. • Hands-on track record building agents and training models with reinforcement learning, ideally for science, mathematics, code, or other complex-reasoning domains. You have shipped working RL systems that beat non-trivial baselines, with rigorous experimental methodology. • Proficiency with modern ML frameworks and distributed training. You can move from a single GPU to a cluster without rewriting your code, and you understand what breaks at each scale. • A physics or mathematics background providing intuition for physical reasoning and scientific tool use. You can hold a substantive conversation with a domain physicist. • Hands-on experience with modern RL algorithms (PPO, SAC, MuZero, multi-agent self-play, search-augmented methods, or comparable). • Deep fluency with PyTorch or JAX, plus distributed training via Ray, XLA, Accelerate, or comparable. • Experience applying agents to simulators, scientific tools, games, or rigorous benchmark suites. • Open-source contributions, conference presentations, or shipped research artifacts that the community has adopted. • How We Work • 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.
Responsibilities
• Build and train AI agents and training systems that learn to do physics. Focus on the core research questions: how agents acquire physical reasoning, how to design action spaces for scientific tool use, how to structure rewards that survive long-horizon discovery tasks, and how training infrastructure scales without breaking the science. • Design evaluation workflows and benchmarks for physics reasoning. Distinguish genuine reasoning from pattern matching and benchmark gaming. Build the instrumentation that makes agent behavior interpretable, not opaque. • Publish results that advance the field of AI for science. Develop training curricula, reward structures, and architectures for discovery tasks; iterate on what works in practice; share what works at top ML venues where it serves the mission. • Collaborate with physicists who design verification harnesses and with engineers who build training infrastructure. Ship working systems end-to-end, not isolated research artifacts.
Apply in one click
Upload My Resume
Drop here or click to browse · Tap to choose · PDF, DOCX, DOC, RTF, TXT