Who is the AI Safety mentor focused on scalable oversight and evaluation?

• A. Maxim Massenkoff
• B. Sam Bowman
• C. Peter McCrory
• D. Saffron Huang

Answer: (B)

Scalable oversight and evaluation is about finding practical ways to supervise and judge AI behavior as systems get more capable, without relying on constant, costly human input. It combines designing robust benchmarks, developing evaluation pipelines, and using techniques like reward modeling to ensure models behave safely and as intended even when they’re powerful enough to surprise us.

Sam Bowman’ s work centers on rigorous evaluation frameworks for language models—creating meaningful benchmarks, probing how models respond, and measuring outputs in reliable, scalable ways. This focus directly supports scalable oversight because it provides concrete methods to assess and steer model behavior as capabilities expand, making him the best fit for a mentor in this area.

The other researchers work in related domains within AI safety or ML methodology, but their primary emphasis isn’t as closely aligned with the evaluation-heavy, scalable oversight niche.