paper recommendation for Part 8: Explainability

[zepingyu0512]

Hi, thanks for this great repository!

I’d like to recommend our recent paper for inclusion:

Title: Back Attention: Understanding and Enhancing Multi-Hop Reasoning in Large Language Models

Link: https://arxiv.org/pdf/2502.10835

Abstract:
In this work, we design and use mechanistic interpretability techniques to analyze why LLMs cannot perform latent multi-hop reasoning well. To achieve this problem, we propose Back Attention, a mechanism that enables large language models to explicitly revisit prior intermediate steps when conducting multi-hop reasoning.

This work might be a good fit under the Part 8: Explainability Section

Thanks for considering!