Question: Scalable Oversight Protocols for Superhuman AI

Recursive reward modeling human error rate

Needs new evidence

1/5 publication-ready

Findings metrics

—

total

—

verified

—

pending

—

rejected

Research