SAIF: A Sparse Autoencoder Framework for Interpreting and Steering Instruction Following of Language Models
This work addresses the challenge of understanding and controlling instruction following in LLMs for practical applications, representing a novel method for a known bottleneck.
The paper tackles the problem of interpreting how large language models follow instructions by using sparse autoencoders to identify specific latents responsible for this behavior, and shows these latents can steer model outputs to align with instructions, with the method scaling across different model sizes.
The ability of large language models (LLMs) to follow instructions is crucial for their practical applications, yet the underlying mechanisms remain poorly understood. This paper presents a novel framework that leverages sparse autoencoders (SAE) to interpret how instruction following works in these models. We demonstrate how the features we identify can effectively steer model outputs to align with given instructions. Through analysis of SAE latent activations, we identify specific latents responsible for instruction following behavior. Our findings reveal that instruction following capabilities are encoded by a distinct set of instruction-relevant SAE latents. These latents both show semantic proximity to relevant instructions and demonstrate causal effects on model behavior. Our research highlights several crucial factors for achieving effective steering performance: precise feature identification, the role of final layer, and optimal instruction positioning. Additionally, we demonstrate that our methodology scales effectively across SAEs and LLMs of varying sizes.