OpenAI corrects major chart errors in GPT-5 demo after users spot glaring inaccuracies, with CEO Sam Altman admitting to a "mega chart screwup"

During OpenAI’s live demonstration of its upcoming GPT-5 model, several viewers noticed glaring inaccuracies in the charts displayed in the background. The errors sparked immediate attention online, with users pointing out that the visual representations did not match the data being presented. One chart compared GPT-5 and OpenAI’s o3 model on a metric called "coding deception," showing GPT-5 with a 50% deception rate and o3 at 47.4%. Despite the higher value, the bar for GPT-5 was significantly smaller than o3’s, creating a misleading visual. Another chart displayed percentages of 69.1%, 30.8%, and 52.8%, yet the bars were sized in a way that suggested the 52.8% figure was larger than both 69.1% and 30.8%, which was factually incorrect. OpenAI CEO Sam Altman acknowledged the mistake on Twitter, calling it a "mega chart screwup." A marketing employee at OpenAI followed up with an apology, referring to the error as an "unintentional chart crime" and confirming that the issue had been corrected in the official blog post announcing the demo. This incident is not the first time a major tech demo has been marred by technical or visual errors. In 2023, Microsoft’s launch of Bing’s AI chatbot included a notable blunder when it incorrectly listed "noisiness" and a "short cord" as drawbacks of a cordless vacuum. Similarly, Google’s AI chatbot Bard, now known as Gemini, gave an inaccurate response about the James Webb Space Telescope during a promotional video that same month. It remains unclear whether the flawed charts were generated using AI tools or created manually. OpenAI has not yet provided a detailed explanation or responded to inquiries about the production process behind the visuals. Despite the gaffe, the core demonstration of GPT-5’s capabilities proceeded as planned, though the chart errors drew significant scrutiny and highlighted the risks of high-stakes public presentations.