PolitiFact, a fact-checking website, has tested ChatGPT, an artificial intelligence (AI) language model, to see if it can fact-check better than humans. ChatGPT uses probabilities to predict the next word in a sentence to give users the most helpful response, rather than the most accurate response. While ChatGPT did work seamlessly for some of PolitiFact’s claims, half of the time it made mistakes or came to a different conclusion than human fact-checkers. ChatGPT’s knowledge cutoff limits its usefulness, as it does not have access to data beyond September 2021. Experts advise that ChatGPT is not yet accurate or reliable enough to replace human fact-checkers but can be used as a support tool for them.
ChatGPT is a type of AI called a large language model that provides helpful responses using probabilities by predicting the next word in a sentence.
Bill Adair is a journalism professor at Duke University and founder of PolitiFact, who has been researching how AI can be used in fact-checking work.
Paragraph about the company:
PolitiFact is a fact-checking website that researches statements from politicians and experts and rates them for accuracy on their Truth-O-Meter. Founded in 2007, PolitiFact aims to hold politicians accountable by verifying their statements and publishing their findings for the public.
Paragraph about the person:
Bill Adair is a journalism professor at Duke University, who is also the founder of PolitiFact, a website that fact-checks politicians’ statements for accuracy. He has been researching how AI can be used in fact-checking work.
