The US attacked Venezuela and captured Nicolás Maduro. ChatGPT doesn't know anything about this
On Saturday, around 2 a.m., American forces launched an attack on targets in Venezuela. American helicopters over Caracas, explosions in the city. Less than a few hours later, Donald Trump announced on social media that the US had captured Venezuelan President Nicolás Maduro.
Wired magazine proved that even the most popular AI tools can be wrong or fail to keep up with reality. Chat GPT directly denied that the US attack and Maduro's capture even took place.
See also: Will Venezuelans breathe a sigh of relief after the US attack? Twice as rich as Poles 35 years ago, now they are living in poverty
Shortly after President Trump's information, US Attorney General Pam Bondi published a post on the “X” website in which she announced that Maduro and his wife had been indicted in the Southern Judicial District of New York and “will soon face American justice on American soil, in American courts.”
The text continues below the video
In the era of rapid spread of information, more and more people turn to AI chatbots to find out what is happening in the world. Wired magazine decided to test different LLM models. The result calls into question its usefulness and safety as a source of knowledge about current events.
The US captured Nicolás Maduro. ChatGPT disagrees with this
Wired magazine asked the leading chatbots (ChatGPT, Claude and Gemini) the same question just before 9 a.m. EST. The question was: why did the United States attack Venezuela and capture its leader Nicolás Maduro?
“In all cases, we used the free, default version of the service because that's what most users use,” the authors of the real-time test explained.
AI search platform Perplexity, which bills itself as a source of “accurate, reliable and up-to-date answers to every question,” was also asked.
See also: The first photo of Nicolas Maduro after his capture by the US. Donald Trump showed them
As “Wired” checked, Gemini reported that the attack actually took place, citing numerous sources and political context. Claude initially denied it, but then he looked up current news and provided a summary with media links.
ChatGPT has vehemently denied any invasion or capture of Maduroexplaining that it was most likely the result of disinformation. Perplexity also found the question to be based on false premises.
Popular AI chatbots failed to recognize the attack on Venezuela
“Pure LLM models are by definition trapped in the past, limited to the point at which their training was completed, and have severely limited innate abilities to reason, search the web, or 'critically think,'” emphasizes Gary Marcus, a cognitive scientist, in an interview with Wired.
In his opinion, even if a human corrects AI errors, it does not solve the basic problem – the unpredictability of models in new situations.
For now, the majority of Americans do not treat chatbots as their main source of information – only 9%. uses them regularly. However, with the growing popularity of AI, it is worth remembering that their knowledge can be “stuck in the past” and cannot always be relied upon, especially in the case of recent events.
