LLM-Agent Oversight on HotpotQA
0.436Scalar RegretScalar routing
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Scalar routingIntervention Type=Scalar2026.06 | 0.436 | — | — | — | |
| Action-conditioned (prefix) routingIntervention Type=Action-conditioned (prefix)2026.06 | — | 0.417 | 0.019 | 0.02 |