Google DeepMind is facing a significant threat to the intellectual property and integrity of its foundational models due to model distillation techniques being used to steal critical model capabilities. The company needs to build a state-of-the-art detection system to identify and prevent these security threats in real time.
Requirements
- Demonstrated research or product expertise in machine learning, with a focus on classifier training, classifier evaluation and model evaluation.
- Deep expertise in training and evaluating a wide range of classifier architectures.
- Proven experience building and scaling ML training and evaluation infrastructure.
- Experience building data pipelines using LLM autoraters.
- Strong understanding of model distillation, model stealing, and other capability extraction techniques.
- Strong software engineering skills and experience with ML frameworks like JAX, PyTorch, or TensorFlow.
- A track record of landing research impact or shipping production ML systems in a multi-team environment.
Responsibilities
- Design and build scalable pipelines for training and evaluating many novel classifiers.
- Train and evaluate classifiers for distillation detection and other key GDM security priorities.
- Collaborate with model training, infrastructure, and deployment teams to deploy classifiers into our production models.
- Research, design, and train novel classifiers to identify distillation and other security threats in real time.
- Build the core detection system that protects Gemini.
- Define how we monitor and protect GDM models by building the core systems that identify sophisticated threats.
Other
- Ph.D. in Computer Science or a related quantitative field, or a B.S./M.S. in a similar field with 2+ years of relevant industry experience.
- Passionate about building the detection systems that protect the future of AI.
- Thrive on ambiguity and enjoy the challenge of iterating to achieve state-of-the-art detection capabilities.
- Value diversity of experience, knowledge, backgrounds and perspectives.
- Commitment to equal employment opportunity regardless of sex, race, religion or belief, ethnic or national origin, disability, age, citizenship, marital, domestic or civil partnership status, sexual orientation, gender identity, pregnancy, or related condition (including breastfeeding) or any other basis as protected by applicable law.
