The potential for AI models to spread disinformation is being studied by researchers
Researchers at the University of Zurich recently discovered that the massive language model GPT-3, created by OpenAI, may be used to produce misinformation. According to the study written up in the journal Science Advances, GPT-3 could produce both accurate and convincing tweets—even when the tweets contained erroneous information. Additionally, the researchers discovered that it was difficult for people to tell the difference between tweets created by GPT-3 and those written by actual Twitter users. This finding raises questions about using AI models to disseminate false information. GPT-3, for instance, might be used to create phony news reports or social media posts to mislead readers.
The model was put to the test by the researchers, who had it write slanted reviews of products on TripAdvisor and Amazon and fabricate false news stories about political candidates. They discovered that the model could produce text that a person could not have written and that unwary readers would likely accept the erroneous information it made.
The study draws attention to the rising worry regarding using sophisticated language modeling to disseminate misinformation online. Many people turn to social media and other online platforms for news and information when confidence in traditional media outlets is at an all-time low.
Overall, the study underlines the necessity of closer examination of advanced language models and their possible influence on our capacity to discriminate between reality and fiction online. As these models are more extensively used, we must create efficient methods for preserving the accuracy of online information and safeguarding ourselves against the negative impacts of misinformation.
