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
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 -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
📖 Reader Mode
~2 min readAbstract: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: this https URL
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2606.26155 [cs.AI] |
| (or arXiv:2606.26155v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2606.26155 arXiv-issued DOI via DataCite |
Submission history
From: Maty Bohacek [view email]
[v1]
Tue, 23 Jun 2026 20:10:53 UTC (3,346 KB)
— Originally published at arxiv.org
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 →HOBA: Hierarchical On-Policy Bidding Agents for Adaptive Online Advertising
HOBA (Hierarchical On-policy Bidding Agents) is a novel hierarchical reinforcement learning framework that enhances online advertising bidding systems by improving adaptability and reducing hyperparameter tuning costs. It utilizes a for hyperparameter inference, a SARSA agent for expert model selection, and a dynamic expert pool for bid execution, achieving a +3.6% increase in target cost during large-scale deployment and outperforming state-of-the-art baselines on AuctionNet.