Summary of Kg-fpq: Evaluating Factuality Hallucination in Llms with Knowledge Graph-based False Premise Questions, by Yanxu Zhu et al.
KG-FPQ: Evaluating Factuality Hallucination in LLMs with Knowledge Graph-based False Premise Questions
by Yanxu Zhu, Jinlin Xiao, Yuhang Wang, Jitao Sang
First submitted to arxiv on: 8 Jul 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper addresses the issue of large language models (LLMs) being misled by false premise questions (FPQs), which can lead to factual knowledge errors. Existing benchmarks for assessing this vulnerability are limited in scale and scalability due to manual construction. To address this, the authors introduce an automated pipeline for creating FPQs based on knowledge graphs (KGs). The pipeline modifies true triplets from KGs into false premises and uses GPTs to generate semantically rich FPQs. The authors present a comprehensive benchmark, the Knowledge Graph-based False Premise Questions (KG-FPQ), which contains approximately 178k FPQs across three domains and six levels of confusability. They conduct extensive evaluations on several representative LLMs and provide valuable insights. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper talks about a problem with big language models called false premise questions. These questions can trick the models into giving wrong answers. Right now, we don’t have a good way to test these models for this kind of error. The authors came up with a new way to make lots and lots of fake questions based on special diagrams that show how things are related. They used a powerful computer program to make these questions sound really realistic. The authors then made a big collection of these fake questions, which they call the Knowledge Graph-based False Premise Questions (KG-FPQ). This will help us understand how well language models can handle tricky questions. |
Keywords
» Artificial intelligence » Knowledge graph
