MQuAKE: Assessing Knowledge Editing in Language Models via Multi-Hop Questions

By Zexuan Zhong, Zhengxuan Wu, Christopher D. Manning, Christopher Potts, Danqi Chen. Conference on Empirical Methods in Natural Language Processing, 2023

346 citations recorded by the source corpus.

The information stored in large language models (LLMs) falls out of date quickly, and retraining from scratch is often not an option. This has recently given rise to a range of techniques for injecting new facts through updating model weights. Current evaluation paradigms are extremely limited, mainly validating the recall of edited facts, but changing one fact should cause rippling changes to the model's related beliefs. If we edit the UK Prime Minister to now be Rishi Sunak, then we should get a different answer to Who is married to the British Prime Minister? In this work, we present a benchmark, MQuAKE (Multi-hop Question Answering for Knowledge Editing), comprising multi-hop questions that assess whether edited models correctly answer questions where the answer should change as an entailed consequence of edited facts. While we find that current knowledge-editing approaches can recall edited facts accurately, they fail catastrophically on the constructed multi-hop questions. We thus propose a simple memory-based approach, MeLLo, which stores all edited facts externally while prompting the language model iteratively to generate answers that are consistent with the edited facts. While MQuAKE remains challenging, we show that MeLLo scales well with LLMs (e.g., OpenAI GPT-3.5-turbo) and outperforms previous model editors by a large margin.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/56e952fd463accff09cf2e35432aaabd7c7c57f3

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Ori Yoran
2. Saku Sugawara
3. Ori Ram
4. Xanh Ho
5. Tomer Wolfson
6. Jonathan Berant
7. Tushar Khot
8. Ashish Sabharwal

Citations in the corpus

• Rethinking machine unlearning for large language models (2024, neutral): “This assertion is also supported by the existence of hard or even adversarial in-scope examples associated with unlearning targets and the challenge to remove their influence in LLMs (Zhong et al., 2023; Patil et al., 2024; Lynch et al., 2024).”
• Making Retrieval-Augmented Language Models Robust to Irrelevant Context (2023, background): “This can be partially explained by the ability of models to combine retrieval and their parametric knowledge (Talmor et al., 2020; Zhong et al., 2023; Cohen et al., 2023).”
• A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions (2023, neutral): “(Zhong et al., 2023b) proposes MQUAKE to evaluate the generalization of injected knowledge and finds that the post-edited model can successfully recall the edited facts but fails in complex multi-hop questions.”
• MoreHopQA: More Than Multi-hop Reasoning (2024, background): “For this reason, multi-hop QA has received much attention over the past few years, prompting the creation of several benchmark datasets such as HotpotQA (Yang et al., 2018), 2WikiMulti-hopQA (2Wiki; Ho et al., 2020), MuSiQue (Trivedi et al., 2022), MQuAKE (Zhong et al., 2023), MRKE (Wu et al., 2024)”
• Editing Large Language Models: Problems, Methods, and Opportunities (2023, neutral): “MemPrompt (Madaan et al., 2022),IKE (Zheng et al., 2023) and MeLLo (Zhong et al., 2023).”
• Knowledge Conflicts for LLMs: A Survey (2024, neutral): “In response to this, researchers explore knowledge editing techniques to efficiently modify a small scope of the knowledge in LLMs (Meng et al., 2022; Zhong et al., 2023).”