Research Scientist, Multimodal Alignment, Safety and Fairness, Deepmind

Google Google · Big Tech · Mountain View, CA +3

Research Scientist at Google DeepMind focusing on multimodal AI alignment, safety, and fairness. The role involves generating new ideas, implementing and evaluating sociotechnical AI systems, and contributing to the research community through publications. Key qualifications include experience with multimodal AI models, deep learning frameworks, and a publication record. Preferred qualifications include fine-tuning LLMs, developing generative AI and agentic AI solutions, and expertise in vision-language models.

What you'd actually do

  1. Generate new ideas, keep up with the state-of-the-art in the field, and discuss research directions with other researchers.
  2. Design, rapidly implement, and evaluate ideas, methods, interfaces, and tools to explore new sociotechnical AI systems.
  3. Report and present research findings and developments clearly and efficiently both internally and externally, verbally and in writing.
  4. Suggest and engage in inter and intra-team collaborations to meet ambitious research goals, while also driving significant individual contributions.
  5. Take ownership of substantial technical projects, from ideation and design to implementation and evaluation, often involving cross-functional collaboration.

Skills

Required

  • PhD degree in Computer Science, Machine Learning, or a related technical field, or equivalent practical experience.
  • Experience in developing multimodal AI models and systems.
  • Experience conducting research and development, including experimental design, implementation, and analysis.
  • Experience in Python and deep learning frameworks (e.g., JAX, Flax, or Gemma).
  • Publication record in machine learning conferences (e.g., NeurIPS, CVPR, ICML, ICLR, ICCV, ECCV).

Nice to have

  • Experience fine-tuning and post-training LLMs using Reinforcement Learning (RL) or other alignment methods.
  • Experience with developing generative AI architectures, techniques, and agentic AI solutions.
  • Experience with multimodal learning, integrating information from different data types (e.g., vision, audio, text).
  • Proven expertise in working with, tuning, and prototyping vision-language models (VLMs) using modern prompting strategies.
  • Strong interest in and awareness of the AI alignment, safety, responsibility, and landscape, with an excitement to collaborate across disciplines.

What the JD emphasized

  • multimodal AI models and systems
  • publication record in machine learning conferences
  • fine-tuning and post-training LLMs using Reinforcement Learning (RL) or other alignment methods
  • developing generative AI architectures, techniques, and agentic AI solutions
  • integrating information from different data types (e.g., vision, audio, text)
  • working with, tuning, and prototyping vision-language models (VLMs) using modern prompting strategies
  • AI alignment, safety, responsibility

Other signals

  • multimodal AI models
  • sociotechnical AI systems
  • AI alignment, safety, responsibility