R0040/2026-03-28/Q002/S02/R03¶
Wei et al. paper on synthetic data reducing sycophancy.
Summary¶
| Field | Value |
|---|---|
| Title | Simple Synthetic Data Reduces Sycophancy in Large Language Models |
| URL | https://arxiv.org/abs/2308.03958 |
| Date accessed | 2026-03-28 |
| Publication date | 2024-02-16 |
| Author(s) | Jerry Wei et al. |
| Publication | arXiv |
Selection Decision¶
Included in evidence base: Yes
Rationale: Demonstrates that sycophancy can be reduced through data-level intervention (synthetic non-sycophantic examples) without changing the training algorithm. This is directly relevant to the question of whether RLHF must be replaced or can be fixed through better data.