Job description

WHO WE ARE Odyssey odyssey.ml is an AI lab pioneering general world models: causal, multimodal systems that learn to predict and interact with the world over long horizons. This foundational technology promises to revolutionize robotics, science, healthcare, education, gaming, defense, and beyond. Odyssey’s founders previously pioneered the most complex application of physical AI: self-driving cars. They’ve now brought together a world-class research team from DeepMind, Tesla, Waymo, Meta, Apple, and Wayve, who have made significant contributions to language models (DeepMind Gemini), video models (DeepMind Veo), world models (Wayve GAIA), and autonomous systems (Tesla FSD). Odyssey has raised significant venture capital from GV, Amazon, AMD, EQT, NVIDIA, Natural Capital, In-Q-Tel, Elad Gil, Jeff Dean, Guillermo Rauch, Garry Tan, Kyle Vogt, and researchers from OpenAI, DeepMind, MSL, Recursive, and Thinking Machines. WHAT WE'RE LOOKING FOR We hire deeply technical staff working in world models, video generation, multimodal, robotics, autonomous vehicles, and adjacent fields. Applied Research sits where the models meet real applications. Whether you're building from scratch, scaling training runs, or working on inference, we hire people who can take an idea and make it real. WHAT YOU’LL DO - Push what world models can do past where current diffusion and transformer approaches stop, with an emphasis on conditioning and control. That might mean new architectures, new training paradigms, or new ways of representing time, dynamics, and interaction. - Train the conditioning methods and control policies that give real handles on generation: action, camera, physics, style, and the long-horizon consistency that holds them together. - Take ideas from hypothesis through experiment to working prototype, grounded in a concrete application. - Design and run the experiments, distributed training, and inference systems that make breakthroughs possible, and ship the results into models real users touch. - Define what "controllable" means in measurable terms. Build the evals that tell us whether a method holds up in the applications it's meant for, not just on a held-out set. - Work shoulder-to-shoulder across research, engineering, and product. Strong work here is recognized by all sides as theirs. - Contribute to the lab's scientific and engineering culture: publishing where it matters, building tools that outlast the project, and engaging with the broader community shaping this field. WHO YOU ARE - Deeply technical, with a track record in world models, generative modeling, video, multimodal learning, large-scale ML systems, or adjacent areas. Experience with conditioning, controllability, or learned control policies is a strong signal. You've either originated work that moved the field, shipped systems at scale, or both. - Comfortable working from first principles. You can define a problem, design experiments to test it, and build the minimal system that proves what's possible. - Fluent across the research-to-application gap. You care how and why models work, and you care that they hold up under real constraints, on real content, at real latency. - Energized by ambiguity. You've operated in areas without precedent and built the tools, frameworks, and patterns as you went. - Have a real view on where this field needs to go next, and can defend it. "Bigger model, more compute" isn't an answer that lands here. - Thrive in small, focused teams that value autonomy, speed, and tight collaboration over process. - Excited to help define a new medium, how AI perceives, learns, and interacts through pixels, over the next decade. 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.

Requirements

Must-haves

Train the conditioning methods and control policies that give real handles on generation: action, camera, physics, style, and the long-horizon consistency that holds them together.
Take ideas from hypothesis through experiment to working prototype, grounded in a concrete application.
Design and run the experiments, distributed training, and inference systems that make breakthroughs possible, and ship the results into models real users touch.
Define what "controllable" means in measurable terms. Build the evals that tell us whether a method holds up in the applications it's meant for, not just on a held-out set.
Work shoulder-to-shoulder across research, engineering, and product. Strong work here is recognized by all sides as theirs.
Contribute to the lab's scientific and engineering culture: publishing where it matters, building tools that outlast the project, and engaging with the broader community shaping this field.
Comfortable working from first principles. You can define a problem, design experiments to test it, and build the minimal system that proves what's possible.
Fluent across the research-to-application gap. You care how and why models work, and you care that they hold up under real constraints, on real content, at real latency.
Energized by ambiguity. You've operated in areas without precedent and built the tools, frameworks, and patterns as you went.
Have a real view on where this field needs to go next, and can defend it. "Bigger model, more compute" isn't an answer that lands here.
Thrive in small, focused teams that value autonomy, speed, and tight collaboration over process.
Excited to help define a new medium, how AI perceives, learns, and interacts through pixels, over the next decade.
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.

Nice-to-haves

None flagged.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

mlaigo

Platform: ashby
Effort: quick