OpenAI acknowledges chart inaccuracies in GPT-5 presentation during livestream

OpenAI acknowledges chart inaccuracies in GPT-5 presentation during livestream

OpenAI CEO Acknowledges Significant Graph Error in GPT-5 Presentation

During OpenAI’s major GPT-5 livestream on August 7, 2025, CEO Sam Altman addressed a significant error within the presentation’s charts, referring to it as a “mega chart screwup.” This admission came after analysts highlighted discrepancies in how the new model’s performance was represented, raising concerns about the credibility of the showcased data, reports 24brussels.

The problematic graph, intended to illustrate GPT-5’s effectiveness in “deception evaluations across models,” exhibited inconsistent scales. For instance, the chart indicated a 50.0 percent deception rate for GPT-5, while the comparison model, OpenAI’s smaller o3, showed a contradictory 47.4 percent rate, with a larger visual representation. This inconsistency prompted the question of whether the data illustrated during the event accurately reflected the information later provided in OpenAI’s blog post, where GPT-5’s deception rate is reported as 16.5 percent.

As the controversy unfolded, Altman explicitly stated, “We have correct numbers in our blog post,” acknowledging the misrepresentation publicly during the presentation. Another OpenAI team member issued an apology, labeling the incident an “unintentional chart crime” and reiterated that the issue was rectified in the corresponding blog.

In the backdrop of OpenAI advocating for its new model as a significant advancement in reducing errors, this misstep on launch day raises questions about the internal processes used in verifying the model’s performance. While it remains unclear if GPT-5 assisted in generating these flawed charts, the company faced scrutiny on one of its most critical days.

The incident highlights the importance of accuracy in AI reporting, amplifying calls for transparency as organizations like OpenAI continue to influence the tech landscape. The implications of such errors could undermine user trust during a time when advanced AI models are under intense scrutiny for their potential impacts and capabilities, reports 24brussels.

Leave a Reply

Your email address will not be published.

Don't Miss

Mint Mobile launches affordable 5G home internet plan starting at $30 per month

Mint Mobile launches affordable 5G home internet plan starting at $30 per month

Mint Mobile Launches Affordable 5G Home Internet Service Mint Mobile has introduced
Google introduces recovery contacts feature for easier account access recovery

Google introduces recovery contacts feature for easier account access recovery

Google Introduces New Account Recovery Features to Enhance Security On October 15,