Snapshot

Help us build generative models of the 3D world. World models power numerous domains, such as media generation, visual reasoning, simulation, planning for embodied agents, and real-time interactive experiences. Work with us to build better versions of Gemini, Genie, and Veo, while also exploring new, spatial modalities beyond images and videos.

The Role Key responsibilities: Conduct research to build generative multimodal models of the 3D world. Solve essential problems to train world models at massive scale: build and train large-scale systems for data annotation, curate and annotate training datasets, build and maintain large model training infrastructure, develop scaling ladders and training recipes, develop metrics for spatial intelligence, enable real-time interactive experiences, study the integration of spatial modalities with multimodal language models, and of course: actually train massive-scale models.

Areas of focus:

• 3D computer vision, spatial annotation systems

• Spatial representations

• Training large-scale transformers

• Generative pixel and latent models

• Infrastructure for large-scale data pipelines and annotation.

• Quantitative evals for spatial accuracy and intelligence.

• Model scaling, efficiency, distillation, training infrastructure

About you

We seek individuals who are passionate about large-scale generative models and believe spatial understanding and generation are on the path to intelligence. We strive for simple methods that scale and look for candidates excited to improve models through infrastructure, data, evals, and compute.

In order to set you up for success as a Research Scientist/Engineer at Google Deep Mind, we look for the following skills and experience:

• MSc or PhD in computer science or machine learning, or equivalent industry experience.

• Experience with large-scale transformer models and/or large-scale data pipelines.

• Track record of releases, publications, and/or open source projects relating to video generation, world models, multimodal language models, or transformer architectures.

• Exceptional engineering skills in Python and deep learning frameworks (e.g., Jax, Tensor Flow, Py Torch), with a track record of building high-quality research prototypes and systems.

• Demonstrated experience in large-scale training of multimodal generative models.

In addition, the following would be an advantage:

• Experience building training codebases for large-scale video or multimodal transformers.

• Expertise optimizing efficiency of distributed training systems and/or inference systems.

• Strong background in 3D representations or 3D computer vision

• Strong publication record at top-tier machine learning, computer vision, and graphics conferences (e.g., NeurIPS, ICLR, ICML, SIGGRAPH, CVPR, ICCV).

• A keen eye for visual aesthetics and detail, coupled with a passion for creating high-quality, visually compelling generative content.