Built independently by an author, for readers. Read the story and support ChapterPal
ChapterPal Logo
Log inSign up

keyword

contextual knowledge

Contextual knowledge refers to the explicit information, evidence, or background text supplied directly within an input prompt or context window during inference in artificial intelligence systems. Distinct from parametric knowledge, which is statically encoded within a model weights during training, contextual knowledge is dynamically introduced at runtime. This external information enables language models to process user-provided documents, reference up-to-date facts, adhere to task-specific constraints, and reason over current data, even when the provided information contradicts or supersedes the model internal pre-trained memory.

1 item

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

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

Yu Zhao, Alessio Devoto, Giwon Hong, Xiaotang Du, Aryo Pradipta Gema, Hongru Wang, Xuanli He, Kam-Fai Wong, Pasquale Minervini

OrganizationsMiniml.AISapienza University of RomeThe Chinese University of Hong KongUniversity College LondonUniversity of Edinburgh

Why you should read this

Introduces SPARE, a training-free representation engineering method that leverages sparse auto-encoders to detect mid-layer conflict signals and steer whether large language models rely on parametric memory or contextual evidence during question answering.

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phenomenon, known as context-memory knowledge conflicts 1, can lead to undesirable model behaviour, such as reliance on outdated or incorrect information.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 inference-time intervention strategies to resolve it. In this work, we propose SPARE, a training-free representation engineering method that uses pre-trained sparse auto-encoders (SAEs) to control the knowledge selection behaviour of LLMs. 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 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%).

Added

2026-09-26