Machine Learning Researcher

On-site, San FranciscoMost of the field is betting that bigger language models lead to more capable AI. This lab is taking a different bet. The thesis: cause and effect genuinely exist in the physical world in a way they never will in text. So instead of scaling another LLM, you'll push the boundaries of AI research while working with one of the most well-observed physical systems on Earth. Weather data offers rapid, objective ground-truth feedback. You predict, and the atmosphere tells you if you were right. That gives you the chance to build causal intelligence models grounded in verifiable feedback, rather than contested benchmarks.You’ll develop novel architectures and training algorithms. The dataset dwarfs what trains today's largest models, and you'll work at a scale of hundreds to thousands of GPUs with no resource queue between you and the run.This is a small team, in the early stage of something enormous. You'll own the full stack, from data to eval, not a narrow slice of someone else's pipeline.What you’ll doResearch and implement novel model architectures and training algorithmsBuild data pipelines and training infrastructure for petabyte-scale multimodal datasetsRapidly iterate on experiments and ablations to sharpen model performanceStay up-to-date on research to continuously bring new ideas to your workWhat you’ll bring1+ year of experience training large-scale foundation models from scratch (pre-training, not fine-tuning or deploying existing models), at a startup or big-tech operating at the frontierStrong ML fundamentals, with depth in at least one core domain: Computer Vision, Sensor Fusion, Language Models, robotics, or Physics-informed NNsFamiliarity with distributed training and inference across hundreds to thousands of GPUsGenuinely mission-driven, with a real pull toward science or the physical worldAbout the companyA well-funded research lab in San Francisco, built by researchers and engineers from self-driving, biotech, robotics, and other frontier domains. They're pursuing a genuinely contrarian thesis: that the path to more capable AI runs through the physical world and its cause-and-effect structure, not through scaling text models further. It's early, the team is small, and the ambition is enormous.Interested enough to hear the full thesis? Hit apply and we'll take it from there.