CLOct 21, 2024

Steering Knowledge Selection Behaviours in LLMs via SAE-Based Representation Engineering

arXiv:2410.15999v361 citationsh-index: 17Has CodeNAACL
Originality Incremental advance
AI Analysis

This addresses the issue of LLMs relying on outdated or incorrect information due to knowledge conflicts, which is an incremental improvement for enhancing model reliability in applications like question-answering.

The paper tackles the problem of context-memory knowledge conflicts in LLMs, where parametric knowledge conflicts with contextual information, and proposes SpARE, a training-free representation engineering method using sparse auto-encoders to control knowledge selection behaviors, achieving improvements of +10% over existing representation engineering methods and +15% over contrastive decoding methods in open-domain question-answering tasks.

Large language models (LLMs) can store a significant amount of factual knowledge in their parameters. However, their parametric knowledge may conflict with the information provided in the context -- this phenomenon, known as \emph{context-memory knowledge conflicts}, can lead to undesirable model behaviour, such as reliance on outdated or incorrect information. Analysing the internal activations of LLMs, we find that they can internally register the signals of knowledge conflict at mid-layers. Such signals allow us to detect whether a knowledge conflict occurs and use \emph{inference-time} intervention strategies to resolve it. In this work, we propose \textsc{SpARE}, a \emph{training-free} representation engineering method that uses pre-trained sparse auto-encoders (SAEs) to control the knowledge selection behaviour of LLMs. \textsc{SpARE} identifies the functional features that control the knowledge selection behaviours and applies them to edit the internal activations of LLMs at inference time. Our experimental results show that \textsc{SpARE} can effectively control the usage of either knowledge source to resolve knowledge conflict in open-domain question-answering tasks, surpassing existing representation engineering methods ($+10\%$) as well as contrastive decoding methods ($+15\%$).

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes