ContextBench: Modifying Contexts for Targeted Latent Activation and Behaviour Elicitation

Robert Graham, Edward Stevinson, Leo Richter, Alexander Chia, Joseph Miller, Joseph Isaac Bloom

Keywords: AI Safety, Interpretability tooling and software, Sparse Autoencoders

Other Keywords: Prompt Optimisation, Elicitation, Feature Visualisation

TL;DR: This paper motivates the AI safety case for generating targeted, linguistically fluent inputs that activate specific latent features or elicit model behaviours, and introduces a benchmark for methods that do this task.

Abstract: Identifying inputs that trigger specific behaviours or latent features in language models could have a wide range of safety use cases. We investigate a class of methods capable of generating targeted, linguistically fluent inputs that activate specific latent features or elicit model behaviours. We formalise this approach as context modification and present ContextBench - a benchmark with tasks assessing core method capabilities and potential safety applications. Our evaluation framework measures both elicitation strength (activation of latent features or behaviours) and linguistic fluency, highlighting how current state-of-the-art methods struggle to balance these objectives. We enhance Evolutionary Prompt Optimisation (EPO) with LLM-assistance and diffusion model inpainting, and demonstrate that these variants achieve state-of-the-art performance in balancing elicitation effectiveness and fluency.

Submission Number: 61