Search papers, labs, and topics across Lattice.
This paper introduces Gated Activation Steering, a novel framework that addresses the issues of sycophancy and hallucination in medical question answering by employing Inference Time Intervention (ITI). By learning separate steering directions for each behavior from contrastive clinical pairs and applying them to causally verified attention heads, the method effectively mitigates unsupported claims and answer shifts under user pressure. The results demonstrate that this approach significantly enhances the robustness of a 4-billion-parameter model, allowing it to maintain accuracy under pressure comparable to larger models without constant intervention.
Gated Activation Steering enables a 4-billion-parameter model to withstand user pressure with a robustness comparable to models over 100 billion parameters, without sacrificing response quality.
Sycophancy and hallucination are persistent failure modes of Large Language Models (LLMs) across domains. However, it becomes particularly consequential in clinical question answering, where responses must remain grounded in the provided context and robust to user pressure. Hallucination can introduce information that is unsupported by the context, while sycophancy can cause a model to abandon a previously correct answer when challenged by the user. Existing approaches, such as prompt-based safeguards and always-on activation steering, often address these behaviors separately or apply interventions broadly across turns, which can unnecessarily deteriorate responses that were already correct. To address these limitations within a single framework, we employ Inference Time Intervention (ITI) to jointly control both behaviors by learning separate steering directions for hallucination and sycophancy from contrastive clinical pairs and applying them to causally verified attention heads. During runtime, behavior-specific gates then determine when intervention is needed: the hallucination component mitigates unsupported claims, while the sycophancy component mitigates answer shifts caused by user pressure. We evaluate this framework on clinical questions grounded in EHR data while keeping the model weights frozen. Across all evaluation settings, we conducted 15,900 model-response runs. Across 600 pressure trajectories for the 4-billion-parameter model, the unsteered model caved in 570 cases. At the same time, gated steering helped it last longer in 551 of them. It held its ground under pressure at levels comparable to those of models with more than 100 billion parameters, showing that targeted inference-time steering can improve robustness without intervening at every turn.
