Spatial intelligence is an umbrella category. For hiring, that is a problem: two candidates who both describe themselves as world-model researchers may have very different strengths, experimental methods and deployment experience.
01 / Rendering is not the same as planning.
A team developing generated 3D worlds may prioritise computer graphics, multimodal generation, 3D reconstruction and scene consistency. A team using learned dynamics for robot control may need stronger state estimation, physical reasoning, policy learning and evaluation on actions. These disciplines overlap, but they are not interchangeable.
02 / Ask what the representation is for.
Is the model creating environments for human consumption, producing training data, forecasting future states or assisting an embodied agent? The answer changes the relevant research history, the compute requirements and what successful evaluations look like.
03 / Evaluate the complete research loop.
Useful evidence includes training data provenance, preprocessing decisions, model architecture, experiments that failed, baselines and downstream impact. For research engineering roles, ask how the candidate made the pipeline reproducible, instrumented and tractable at scale.
04 / Build adjacent hiring maps.
Map graphics, vision, generative modelling, multimodal systems, simulation, GPU systems and robot learning separately before combining them. This avoids rejecting candidates who have the right fundamentals simply because their title is not “world-model researcher.”
World Labs describes its work in spatial intelligence and world models, including its Atlas research model. This is an external technical reference, not a claim that World Labs is an Impax client.
Building a spatial AI team?
We can help define which research and engineering specialisms your next milestone actually requires.
