Can Large Language Models Help Us Combat Online Misinformation?

The institute announces the publication of its first article, a study on automated fact-checking and the role of large language models (LLMs) like GPT-4 in combating misinformation, authored by Dorian Quelle and Prof. Alexandre Bovet.

As the quantity of misinformation online eclipses the capacity fact-checkers, LLMs may hold the promise of verifying content automatically. The study investigates into the performance of LLMs in fact-checking tasks, highlighting their ability to retrieve contextual information to asses the veracity of a statement even to substantiate their findings. The study calls for caution, showing that while GPT-4 is superior to GPT-3, substantial variations in accuracy exist depending on the nature of the queries and the truthfulness of the claims. The findings emphasise the potential of LLMs in fact-checking but caution against their variable accuracy, advocating for continued research to fully understand their capabilities and limitations.