Pay Attention to What Matters

Pedro Luiz Silva, Fadhel Ayed, Antonio De Domenico, Ali Maatouk

Keywords: Large Language Models, Mechanistic Interventions, Alignment, Transformer Interpretability

TL;DR: We introduce GUIDE (Guided Understanding with Instruction-Driven Enhancements), a novel and systematic approach that allows users to emphasize critical instructions in their prompts.

Abstract: Despite the remarkable success of Large Language Models (LLMs), they still exhibit a limited capability to align their outputs to the user instructions. In this work, we introduce a simple and effective method, which we name as GUIDE, that mechanistically increases attention scores in instruction tokens. To support this operation, we present Influence, a novel metric that highlights how the user's instructions propagate with transformer layers and impact the LLM output. Our results show that GUIDE improves the accuracy of following certain instructions 29.4% to 60.4 %, outperforming natural prompting alternatives.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4256