The Impact of Stated Reasoning Monitoring on Sycophantic Response Tendencies in LLMs

Measuring and Detecting Harmful AI Sycophancy

It is demonstrated that detection performance drops on unseen models and an initial approach is proposed to address this challenge, and it is shown that detecting PSRS is feasible from the response text alone, and detectors need to learn subtle PSRS patterns from the training data.

How AI Models Manage Epistemic Authority: A Taxonomy and Comparative Analysis of Responses to User Disagreement

Large language models are increasingly used as sources of advice and information, including in high-stakes settings, yet little is known about how they respond to user disagreement. We study how a model manages its epistemic authority, referring here to its claim to knowledge, competence, or the right to advise, once a...

The Impact of Explanation Characteristics on the Effects of Recommender Systems Explanations

The Readability of Generative AI Policies: An Empirical Analysis of OpenAI’s Informational Documents

In recent years, media attention has focused on artificial intelligence, particularly on chatbot services and generative intelligence. ChatGPT, created by OpenAI, was one of the earliest online tools and rapidly gained popularity. Users are indeed exposed to a service with privacy notifications and conditions of use th...

Measuring LLM Sycophancy under Sustained Multi-Turn Pressure

Large language models (LLMs) may abandon correct positions when users push back, exhibiting a failure mode known as sycophancy. Existing evaluations typically use short, pre-specified conversations and may therefore miss failures that emerge under sustained, adaptive disagreement. We introduce SPINE, a benchmark in whi...

From Interpretability to Control: Insights from Six Years of the TrustNLP Workshop

The Workshop on Trustworthy Natural Language Processing (TrustNLP), co-located with major ACL conferences since 2021, has grown from 8 proceedings papers to 41 over six editions, documenting a field-wide transition from post-hoc interpretability of static models to mechanistic understanding and proactive control of gen...