Playing Devil's Advocate: Off-the-Shelf Persona Vectors Rival Targeted Steering for Sycophancy

TL;DR

This paper demonstrates that off-the-shelf persona vectors can effectively reduce model sycophancy, rivaling traditional methods, and highlights sycophancy as a persona-level property rather than a single steerable direction.

Contribution

It shows that pre-existing persona vectors can mitigate sycophancy in instruction-tuned models, offering an alternative to trained steering methods and revealing the geometric independence of sycophancy.

Findings

Steering toward doubt or scrutiny personas reduces sycophancy by 68-98% of CAA's effect.

Persona steering maintains accuracy when the user is correct.

Sycophancy is more a persona-level property than a single steerable direction.

Abstract

We study the effect of different persona on \textbf{sycophancy}: model's agreement with users even when the user is incorrect. The standard mitigation, Contrastive Activation Addition (CAA), derives a steering direction from labelled pairs of sycophantic and honest responses. This study evaluates whether off-the-shelf persona steering vectors, originally developed for general role-playing and not trained on sycophancy data, can serve as an alternative. In two instruction-tuned models, steering toward personas characterised by doubt or scrutiny reduces sycophancy to approximately $68\%$ and $98\%$ of CAA's effect, and, unlike CAA, maintains accuracy when the user is correct. The effect is also asymmetric: steering toward agreeable personas does not produce a mirror increase in sycophancy. Geometrically, the persona vector is largely independent of the direction of sycophancy in activation space. Collectively, these findings suggest that sycophancy is better understood as a persona-level property rather than a single steerable direction. We release our code here: https://anonymous.4open.science/r/Sycophancy-Steering-9DF0/.