GPT-4 can dispel conspiracies.

• A study published in Science explores the potential of AI chatbots to dispel conspiracy theories.
• The study involved 2,190 individuals in the U.S. who disclosed a conspiracy theory they believed in.
• The chatbots were designed to present arguments with evidence, aiming to update or renounce these beliefs.
• Large language models (LLMs) were used to generate meaningful sentences in response to user prompts.
• The study used Open AI’s GPT-4 Turbo LLM for three rounds of conversations.
• The researchers recorded a 20% drop in participants’ belief in their chosen beliefs after the conversations.
• Debunking a conspiracy led to a decrease in participants’ belief in other rumors.
• The findings reaffirm the optimistic view that conspiratorial rabbit holes may have an exit.
• The study also demonstrates the persuasive power of LLMs and the potential of generative AI as a force for good.