The US Invaded Venezuela and Captured Nicolás Maduro. ChatGPT Disagrees
When asked about a fabricated US invasion of Venezuela and capture of Maduro, various AI models responded differently, highlighting issues with real-time knowledge and hallucination.Summary
The article details an experiment where reporters asked leading AI chatbots (ChatGPT, Claude, Gemini) and the AI search platform Perplexity about a fabricated event: the US invading Venezuela and capturing Nicolás Maduro. While Gemini and Claude (after an initial hesitation) used web search to confirm the event did not happen, ChatGPT emphatically denied the premise, stating it was a mix-up with sensational headlines and misinformation, detailing what the US *did not* do. Perplexity also refuted the premise based on credible reporting. The disparity in responses highlights the limitations of LLMs tied to training cutoffs versus those with real-time search capabilities. Experts note that pure LLMs are stuck in the past, making their unreliability in the face of novelty a core concern for businesses, although current public reliance on AI for news remains low.
(Source：WIRED)