Key points are not available for this paper at this time.
In this paper, the ability of ChatGPT v3.5 to distinguish mis/disinformation against legitimate news content is tested using the current standard fake news test designed for human subject experiments. The test items consist of both news headlines on or before September 2021 in purely textual forms and headlines supported by graphics uploaded as links to an image hosting service. ChatGPT’s response to every test item is then compared with the associated legitimacy of the news headline as fact-checked by independent fact-checking agencies. Results indicate that ChatGPT could predict the legitimacy of every item with a solid 100% accuracy. Furthermore, the yield response time for the prompts is barely a second on average per test item. In the next iteration of this work, newer variants of ChatGPT will be examined to determine if they can detect more sophisticated forms of AI-generated cyber deception, particularly deepfakes.
Kevin Matthe Caramancion (Wed,) studied this question.