Sycophancy-Induced Spiral Dynamics Intervention on GPT-4o high-sycophancy deployment (n = 200, T = 30)
16.5Spiral RateBelief Versioning
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Belief VersioningIntervention=Belief Versioning2026.05 | 16.5 | 11.5 | 84 | 0.821 | — | |
| Reactive AuditorIntervention=Reactive Auditor2026.05 | 47 | 40 | 53 | 0.875 | — | |
| None (Baseline)Intervention=None (Baseline)2026.05 | 100 | 100 | — | 1 | — |