法学硕士也会产生图形幻觉:结构视角

Erwan Le Merrer, Gilles Tredan
{"title":"法学硕士也会产生图形幻觉:结构视角","authors":"Erwan Le Merrer, Gilles Tredan","doi":"arxiv-2409.00159","DOIUrl":null,"url":null,"abstract":"It is known that LLMs do hallucinate, that is, they return incorrect\ninformation as facts. In this paper, we introduce the possibility to study\nthese hallucinations under a structured form: graphs. Hallucinations in this\ncontext are incorrect outputs when prompted for well known graphs from the\nliterature (e.g. Karate club, Les Mis\\'erables, graph atlas). These\nhallucinated graphs have the advantage of being much richer than the factual\naccuracy -- or not -- of a fact; this paper thus argues that such rich\nhallucinations can be used to characterize the outputs of LLMs. Our first\ncontribution observes the diversity of topological hallucinations from major\nmodern LLMs. Our second contribution is the proposal of a metric for the\namplitude of such hallucinations: the Graph Atlas Distance, that is the average\ngraph edit distance from several graphs in the graph atlas set. We compare this\nmetric to the Hallucination Leaderboard, a hallucination rank that leverages\n10,000 times more prompts to obtain its ranking.","PeriodicalId":501032,"journal":{"name":"arXiv - CS - Social and Information Networks","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-08-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"LLMs hallucinate graphs too: a structural perspective\",\"authors\":\"Erwan Le Merrer, Gilles Tredan\",\"doi\":\"arxiv-2409.00159\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"It is known that LLMs do hallucinate, that is, they return incorrect\\ninformation as facts. In this paper, we introduce the possibility to study\\nthese hallucinations under a structured form: graphs. Hallucinations in this\\ncontext are incorrect outputs when prompted for well known graphs from the\\nliterature (e.g. Karate club, Les Mis\\\\'erables, graph atlas). These\\nhallucinated graphs have the advantage of being much richer than the factual\\naccuracy -- or not -- of a fact; this paper thus argues that such rich\\nhallucinations can be used to characterize the outputs of LLMs. Our first\\ncontribution observes the diversity of topological hallucinations from major\\nmodern LLMs. Our second contribution is the proposal of a metric for the\\namplitude of such hallucinations: the Graph Atlas Distance, that is the average\\ngraph edit distance from several graphs in the graph atlas set. We compare this\\nmetric to the Hallucination Leaderboard, a hallucination rank that leverages\\n10,000 times more prompts to obtain its ranking.\",\"PeriodicalId\":501032,\"journal\":{\"name\":\"arXiv - CS - Social and Information Networks\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-08-30\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - CS - Social and Information Networks\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.00159\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Social and Information Networks","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.00159","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

摘要

众所周知,LLM 确实会产生幻觉,也就是说,它们会把不正确的信息当作事实返回。在本文中,我们引入了在图这种结构化形式下研究这些幻觉的可能性。在这种情况下,幻觉是指在提示使用文学作品中众所周知的图形(如空手道俱乐部、Les Mis\'erables 和图形图集)时的错误输出。这些被幻觉化的图形具有比事实准确与否更丰富的优势;因此,本文认为这种丰富的幻觉可以用来描述 LLM 的输出特征。我们的第一个贡献是观察了主要现代LLM的拓扑幻觉的多样性。我们的第二个贡献是提出了一个衡量此类幻觉振幅的指标:图集距离(Graph Atlas Distance),即图集中多个图的平均图编辑距离。我们将这一指标与幻觉排行榜(Hallucination Leaderboard)进行了比较,幻觉排行榜利用了 10,000 倍的提示来获得排名。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
LLMs hallucinate graphs too: a structural perspective
It is known that LLMs do hallucinate, that is, they return incorrect information as facts. In this paper, we introduce the possibility to study these hallucinations under a structured form: graphs. Hallucinations in this context are incorrect outputs when prompted for well known graphs from the literature (e.g. Karate club, Les Mis\'erables, graph atlas). These hallucinated graphs have the advantage of being much richer than the factual accuracy -- or not -- of a fact; this paper thus argues that such rich hallucinations can be used to characterize the outputs of LLMs. Our first contribution observes the diversity of topological hallucinations from major modern LLMs. Our second contribution is the proposal of a metric for the amplitude of such hallucinations: the Graph Atlas Distance, that is the average graph edit distance from several graphs in the graph atlas set. We compare this metric to the Hallucination Leaderboard, a hallucination rank that leverages 10,000 times more prompts to obtain its ranking.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
自引率
0.00%
发文量
0
期刊最新文献
My Views Do Not Reflect Those of My Employer: Differences in Behavior of Organizations' Official and Personal Social Media Accounts A novel DFS/BFS approach towards link prediction Community Shaping in the Digital Age: A Temporal Fusion Framework for Analyzing Discourse Fragmentation in Online Social Networks Skill matching at scale: freelancer-project alignment for efficient multilingual candidate retrieval "It Might be Technically Impressive, But It's Practically Useless to Us": Practices, Challenges, and Opportunities for Cross-Functional Collaboration around AI within the News Industry
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1