Member of Technical Staff, Foundation Models

What we're looking for

The right person will have a deep interest in building large foundation world models, and in the scaling laws that tell you which ones are worth building. You will want to train them from scratch, at the scale that takes: runs across thousands of GPUs, data mixtures measured in years of video, and an interactive model at the end that has to hold up frame by frame under a human's hands. You will treat that as one system, data, architecture, training, inference, evaluation, rather than as six specialties, and go wherever the bottleneck is.

This is a cutting edge research area that is not yet mature, so you will be working at the cusp of what’s possible. Most new experiments in this area will fail, your focus will be on maximally learning from failed experiments to increase the chances of eventual success.

What you’ll do

• Learn what makes large real-time world models tick. Understand how data, architecture, scale, and diffusion algorithms interact.
• Run scaling studies and use them: fit scaling laws over model size, data, and compute, and let them pick the next large run rather than intuition alone.
• Own world model training at scale, large distributed runs across thousands of GPUs, the data mixtures that feed them, and the loss and stability work that keeps them alive for weeks.
• Implement state of the art ML algorithms, define metrics, and relentlessly iterate on leaderboards.
• Work end to end across the stack, from data pipelines and tokenizers through training to real-time inference and evaluation.
• Be part of a team that is defining and leading the world model space.
• Exploit the latest features on modern GPUs to increase training and inference efficiency.
• Take ownership of the full ML stack, including the core frameworks that Odyssey researchers and product engineers alike rely on.

Who you are

• 2+ years of software engineering experience, with significant work in ML performance.
• 4+ years of ML engineering experience, or a PhD in a related field.
• Hands-on experience training large models, pretraining at multi-node scale, and the debugging that comes with it.
• Comfortable reasoning about scale: scaling laws, compute and data budgets, and what a small-scale ablation does and does not predict.
• A holistic ML engineer — happy to move between data, model, systems, and evaluation, and to own the whole path from an idea to a shipped model.
• Track record of owning projects end to end.
• Not shy to touch any stage of an ML pipeline.
• Proficiency with PyTorch (or TF/JAX).
• Highly experiment driven.

Our staff is expected to be in office at least three times a week. A lot of our progress comes from working through problems together, and we want you in the room for that.