Summary of Wildhallucinations: Evaluating Long-form Factuality in Llms with Real-world Entity Queries, by Wenting Zhao et al.
WildHallucinations: Evaluating Long-form Factuality in LLMs with Real-World Entity Queries
by Wenting Zhao, Tanya Goyal, Yu Ying Chiu, Liwei Jiang, Benjamin Newman, Abhilasha Ravichander, Khyathi Chandu, Ronan Le Bras, Claire Cardie, Yuntian Deng, Yejin Choi
First submitted to arxiv on: 24 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 The proposed WildHallucinations benchmark evaluates the factuality of large language models (LLMs) in generating information about real-world entities from user-chatbot conversations. The evaluation includes 118,785 generations from 15 LLMs on 7,919 entities, showcasing a significant gap between model performance and human expectations. Notably, half of these entities lack associated Wikipedia pages, making this benchmark essential for bridging the knowledge gap in various domains. Moreover, the study highlights varying hallucination rates across different domains, emphasizing the need to develop more effective LLMs capable of providing accurate information. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new benchmark, WildHallucinations, is introduced to test how well large language models (LLMs) can generate true information about real-world things people talk about. Right now, there’s a big problem: existing tests don’t cover the many topics users want answers about. To fix this, researchers created WildHallucinations by having LLMs generate text about chatbot conversations and then checking it against what we know from searching the internet. They found that some models are really good at making things up, especially when talking about things without Wikipedia pages. This study shows how important it is to have better language models that can give us accurate answers. |
Keywords
» Artificial intelligence » Hallucination
