A Taxonomy of LLM Hallucinations скачать в хорошем качестве

A Taxonomy of LLM Hallucinations

In this AI Research Roundup episode, Alex discusses the paper: 'A comprehensive taxonomy of hallucinations in Large Language Models(2508.01781v1)' Large language models (LLMs) often generate plausible but factually incorrect content, a problem known as hallucination. This paper introduces a comprehensive taxonomy to classify these hallucinations, arguing they are an inherent part of current LLM designs. It explores key distinctions, such as intrinsic hallucinations (contradicting the input) versus extrinsic ones (contradicting reality), and factuality versus faithfulness. The research also analyzes the underlying causes, from data and model issues to prompting, and reviews methods for detection and mitigation. This framework provides a crucial tool for understanding and addressing one of the biggest challenges in LLMs today. Paper URL: https://arxiv.org/pdf/2508.01781 #AI #MachineLearning #DeepLearning #LLM #AIHallucination #AISafety #NLProc