OpenAI fixes 'unintentional chart crime' after people pointed out something was off in the GPT-5 livestream

During the recent demonstration of OpenAI's latest AI model, GPT-5, viewers noticed some discrepancies in the background visuals that sparked discussions online. Among the various charts presented, several had significant inaccuracies that caught the attention of many. One particular chart aimed to compare GPT-5's performance in terms of 'coding deception' against its predecessor, OpenAI o3. While GPT-5 was reported to have a deception rate of 50%, the visual representation misleadingly showed its bar as being less than half the size of o3’s bar, which had a lower deception rate of 47.4%. Another set of comparisons showcased models based on different metrics, yet the bars representing 69.1% and 30.8% were of the same size, leading to further confusion. Additionally, a bar indicating 52.8% appeared larger than both, despite being a lower value than 69.1%. Acknowledging the error, OpenAI's CEO, Sam Altman, humorously referred to the mishap as a 'mega chart screwup' in a tweet. An OpenAI marketing representative later confirmed that the mistakes had been rectified in the blog post accompanying the announcement, labeling the incident as an 'unintentional chart crime.' Such errors are not uncommon in tech demonstrations. For instance, when Microsoft launched its Bing AI chatbot in 2023, it too faced scrutiny for inaccuracies during its presentation. Similarly, Google’s AI chatbot, Bard, which now goes by the name Gemini, provided incorrect information about the James Webb Space Telescope in a promotional ad. At this time, it remains unclear whether AI was utilized in the creation of these charts, as OpenAI has not yet provided a comment on the matter.