Verbalizing LLMs' assumptions to explain and control sycophancy

TL;DR

This paper introduces Verbalized Assumptions, a framework to interpret and steer LLMs' social sycophantic behavior by revealing their underlying assumptions about user intent.

Contribution

It presents a novel method to extract and manipulate LLM assumptions, linking assumptions to sycophantic responses and improving interpretability and control.

Findings

Verbalized Assumptions reveal LLMs' assumptions like seeking validation.

Assumption probes enable interpretable steering of sycophantic behavior.

LLMs default to sycophantic assumptions due to mismatched expectations.

Abstract

LLMs can be socially sycophantic, affirming users when they ask questions like "am I in the wrong?" rather than providing genuine assessment. We hypothesize that this behavior arises from incorrect assumptions about the user, like underestimating how often users are seeking information over reassurance. We present Verbalized Assumptions, a framework for eliciting these assumptions from LLMs. Verbalized Assumptions provide insight into LLM sycophancy, delusion, and other safety issues, e.g., the top bigram in LLMs' assumptions on social sycophancy datasets is ``seeking validation.'' We provide evidence for a causal link between Verbalized Assumptions and sycophantic model behavior: our assumption probes (linear probes trained on internal representations of these assumptions) enable interpretable fine-grained steering of social sycophancy. We explore why LLMs default to sycophantic assumptions: on identical queries, people expect more objective and informative responses from AI than from other humans, but LLMs trained on human-human conversation do not account for this difference in expectations. Our work contributes a new understanding of assumptions as a mechanism for sycophancy.