AI chatbots have become significantly worse at spreading misinformation, according to a recent analysis, with platforms like Inflection, Perplexity, and Meta standing out.
NewsGuard researchers studied the world’s 10 leading generative AI tools. They found that chatbots spread false claims about 35% of the time when asked controversial questions, which is almost double the 18% rate last year.
The worst performer was Inflection, which made false claims 57% of the time when asked news questions. Perplexity’s rate was 47%, while Meta and ChatGPT’s were 40%.
McKenzie Sadeghi, NewsGuard’s AI editor, believes the reason lies in the changes in the AI tools’ training methods. While they previously analyzed data limitations or avoided sensitive topics, large language models are now leveraging real-time web searches, which are often contaminated with disinformation from malicious actors, including Russian networks.
