R0040/2026-03-28/Q002/S02/R01¶
Comprehensive survey of sycophancy causes and mitigations.
Summary¶
| Field | Value |
|---|---|
| Title | Sycophancy in Large Language Models: Causes and Mitigations |
| URL | https://arxiv.org/abs/2411.15287 |
| Date accessed | 2026-03-28 |
| Publication date | 2024-11-22 |
| Author(s) | Lars Malmqvist |
| Publication | arXiv / Springer (conference proceedings) |
Selection Decision¶
Included in evidence base: Yes
Rationale: Most comprehensive survey of sycophancy causes and mitigations found. Identifies four distinct cause categories and multiple mitigation approaches across training, architecture, inference, and evaluation. Directly addresses the relationship between RLHF and sycophancy.