Detecting and Controlling Sycophancy with Cascading Linear Features
Quick Answer
This study introduces an iterative data generation pipeline for isolating cascading linear features to detect and control sycophancy in language models.
Quick Take
This study introduces an iterative data generation pipeline for isolating cascading linear features to detect and control sycophancy in language models. By moving beyond binary sample pairs, the method enhances interpretability and performance, outperforming existing baselines in detection and steering with lower computational costs. The findings suggest that sycophancy features form linearly separable subspaces, improving model activation selection.
Key Points
- Introduces a new data generation pipeline for detecting sycophancy in language models.
- Focuses on cascading linear features for better interpretability and behavior control.
- Demonstrates improved performance over LLM-as-a-judge and system prompting baselines.
- Achieves lower computational demand while enhancing detection and steering capabilities.
- Code and data available at the provided GitHub link.
Paper Resources
Article Content
From source RSS / original summaryarXiv:2606. 26155v1 Announce Type: new Abstract: Interpreting and controlling model behaviors through activation steering methods requires many pairs of contrastive samples that clearly exhibit desired or undesired behavior. These data pairs determine the degree to which interpretability frameworks can reliably detect model features responsible for a behavior, and therefore the ability to steer models toward or away from such behavior.
In this work, we present an iterative data generation pipeline that isolates cascading linear features responsible for a behavior. Specifically, we show how moving beyond simple binary pairs of samples, and instead isolating samples that show degrees of features that scale linearly with behavior, allows for better disentanglement of features. We focus on detecting and steering away from sycophancy -- the tendency of language models to prioritize user validation.
We demonstrate that sycophancy features discovered through cascading samples form linearly separable subspaces, and allow for selection of model activations that more clearly correspond to the desired behavior than baseline approaches. We also evaluate their ability to enable detection, deterministic scoring, and robust steering, and see that they either match or outperform LLM-as-a-judge and system prompting baselines while providing lower computational demand and more interpretability guarantees.
Code & Data: https://cascading-feats. github. io/
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →How Do Tool-Augmented LLM Agents Perform on Real-World Energy Analytics Tasks?
This study evaluates tool-augmented LLM agents on 243 energy market analytics tasks, revealing significant performance differences between closed-source and open-source models. The tasks cover market data retrieval, knowledge interpretation, and quantitative modeling, highlighting the need for real-time data and specialized tools in energy analytics.