Frontier AI Research Scientist, DeepMind

Google•Mountain View, CA

3d•$147,000 - $210,000

About The Position

As a Research Scientist at DeepMind, you will conduct research that advances AI development and supports product innovation. The role involves setting up large-scale tests, quickly deploying promising ideas, managing deadlines and deliverables, and applying current theories to develop new products, processes, and technologies. Work may include experiments, prototypes, and new architectures addressing real-world computer science problems. You will contribute to the wider research community by sharing and publishing findings, with ideas arising from internal projects and collaborations with universities and technical institutes. The position focuses on foundational post-training and advanced reasoning across large language models and diffusion architectures, translating algorithmic innovation into deployed products and services.

Requirements

Bachelor's degree in Computer Science, Computer Engineering, Electrical Engineering, a related technical field, or equivalent practical experience.
At least 2 years of experience with ML/AI algorithms, tools, deep learning, or other AI/ML modeling.
At least 2 years of experience with Python and a related machine learning package, such as Keras, PyTorch, or Hugging Face Transformers.
At least 1 year of experience developing algorithms for intelligent agents, including planning, reinforcement learning, or multi-agent systems.
Experience in artificial intelligence or machine learning.

Nice To Haves

Master's degree or PhD in Computer Science, Artificial Intelligence, Machine Learning, or a related technical field.
Deep expertise in large language model (LLM) post-training techniques, particularly reinforcement learning and reward modeling at scale.
Track record of publications at NeurIPS, ICLR, or ICML.

Responsibilities

Drive foundational post-training and advanced reasoning research across large language models (LLMs) and diffusion architectures.
Develop next-generation post-training techniques, including self-distillation methods, scaling RLxF for complex agentic capabilities, and novel environment-in-the-loop RLxF recipes.
Partner with GDM verticals to integrate research breakthroughs into products and services, including Thinking Strike, Veo/Omni, and Science.
Take end-to-end ownership of algorithms and ensure the work continuously feeds GDM production systems.
Contribute to internal research manuscripts and publish research in venues such as ICLR, ICML, and NeurIPS, subject to GDM approvals.