We look for researchers who are excited to tackle unsolved problems. Predicting the future is only half the battle; the other half is identifying the actions that can alter it. Your mission is to build the planning layer on top of the LPM - conditioning the model on objectives and producing the actions that achieve them, from operational decisions to physical interventions. It is the capability that provides our models with interventional causality rather than merely observational causality, and it has no established playbook.
Responsibilities- Research and implement methods that turn a predictive physics model into one that reasons toward objectives - planning, control, and decision-making against a learned model of the world
- Develop approaches for decision-making under uncertainty in high-dimensional, continuous physical state spaces
- Build interfaces for specifying objectives and constraints, and methods for producing actions that satisfy them
- Run experiments and ablations that connect reasoning methods to decision quality
- Work across the full ML stack - data, model, eval, and infrastructure - to take ideas from prototype to scaled training runs
What we're looking forWe value a relentless approach to problem-solving, rapid execution, and the ability to quickly learn in unfamiliar domains.
- Strong grasp of machine learning fundamentals, with depth in at least one relevant area (e.g. reinforcement learning, planning and control, decision-making under uncertainty, model-based RL, post-training of large models)
- Experience training models and the ability to understand experimental results through careful analysis and ablation studies
- Familiarity with the challenges of reasoning, planning, or acting with learned models
- A track record of turning open-ended research problems into working systems
