Which mentor is the AI Safety mentor focusing on scalable oversight and evaluation?

• A. Sam Bowman
• B. Maxim Massenkoff
• C. Saffron Huang
• D. Nicholas Carlini

Answer: (A)

Scalable oversight and evaluation is about building evaluation frameworks that can reliably monitor and guide AI behavior as models become more capable, using scalable metrics, human-in-the-loop feedback, and robust benchmarks.

Sam Bowman is the mentor who aligns with this focus because his work centers on how we measure and evaluate NLP models at scale. He emphasizes rigorous evaluation methods, reliable benchmarks, and how to compare model outputs in a way that remains meaningful as systems grow more capable. This directly supports the goal of scalable oversight: you need assessment processes that stay informative and efficient even as models become more complex.

The other mentors tend to concentrate on different safety angles, such as robustness, adversarial safety, or broader ethics and policy considerations, rather than the specialized evaluation and monitoring infrastructure that scalable oversight requires.