5 papers
Suppressing Pressure, Amplifying Evidence: Self-Guided Attention Steering to Mitigate Sycophancy and Stubbornness
Yinghao He, Mengyu Xu, Haixiang Sun +8
Reliable language models should resist unsupported user pressure while effectively using objective contextual information. However, models may exhibit sycophancy by yielding to uns…
Difference-in-Differences on a Censored Rating Scale Can Manufacture an Effect: Evidence from a Pre-Registered LLM-Judge Audit
Shuyi Fan, Boyuan Deng, Mengyu Xu +3
LLM-judge audits assess bias by comparing ratings across matched conditions. Difference-in-differences designs take the rating difference between two candidate responses to each it…
Equitable System-Prompt Selection via Constrained Mixed-Strategy GroupDRO
Mengyu Xu, Qiaoxin Yang, Zhihan Liu +4
Large language models are increasingly used for information seeking, yet semantically equivalent questions phrased in different ways can receive answers of considerably different q…
Decoupling Generation and Selection for Budget-Constrained Faithful Summarization
Zeyu Wang, Guanghua Wang, Meng Xu
Abstractive summarization models remain vulnerable to factual inconsistency, redundancy, and weak length control. We propose a modular generation-and-selection framework for senten…
Rethinking LLM-Judged Helpfulness as a Pedagogy Signal: A Pre-Registered Audit Across Tutor Models
Shuyi Fan, Boyuan Deng, Mengyu Xu +4
LLM tutoring poses a measurement problem: can a general-purpose helpfulness rubric distinguish direct answer-giving from pedagogical guidance? We audit this signal in a pre-registe…