Hallucinating (or poorly fed) LLMs?

The problem of data accuracy

Abstract

Data scraping is crucial for large language models (LLMs) to gather substantial data for training. However, it raises concerns regarding accuracy. Web scraping systems lack filtering, leading to inaccurate and outdated information. Validating accuracy in large volumes is, however, technically demanding. Nevertheless, data accuracy is vital for output quality and user trust in LLMs. This presentation explores reconciling data scraping with accuracy, considering conflicting rights and interests at stake.