Verbalizing LLMs' assumptions to explain and control sycophancy
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914462601052160 |
|---|---|
| author | Cheng, Myra Sieh, Isabel Zope, Humishka Yu, Sunny Ibrahim, Lujain Arora, Aryaman Moore, Jared Ong, Desmond Jurafsky, Dan Yang, Diyi |
| author_facet | Cheng, Myra Sieh, Isabel Zope, Humishka Yu, Sunny Ibrahim, Lujain Arora, Aryaman Moore, Jared Ong, Desmond Jurafsky, Dan Yang, Diyi |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_03058 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Verbalizing LLMs' assumptions to explain and control sycophancy Cheng, Myra Sieh, Isabel Zope, Humishka Yu, Sunny Ibrahim, Lujain Arora, Aryaman Moore, Jared Ong, Desmond Jurafsky, Dan Yang, Diyi Computation and Language Artificial Intelligence Computers and Society 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. |
| title | Verbalizing LLMs' assumptions to explain and control sycophancy |
| topic | Computation and Language Artificial Intelligence Computers and Society |
| url | https://arxiv.org/abs/2604.03058 |