A New Metric for Lowering Hallucinations in Large Language Models

This study looks into the issue of hallucinations and incorrect responses caused by language learning models (LLMs). A well-liked text production tool, ChatGPT, is utilized for many different Natural Language Processing (NLP) functions, including summarizing and generating conversations. A number of lengthy text summaries are produced by ChatGPT, which may lead to inaccurate or difficult responses. In this work, we quantitatively assessed the hallucination in the ChatGPT answers using the BERTScore mathematical measuring tool. The findings point to particular techniques that can raise reliability and precision. The goal of the research is to make language models more reliable, less prone to hallucinations, and able to explain why they provide incorrect responses.