During a recent Reddit AMA (Ask Me Anything) session, OpenAI CEO Sam Altman addressed users’ most pressing questions. At the heart of the discussion were the newly launched GPT-5 model, its initially “bumpy” performance, calls to bring back the older GPT-4o model, and the now-famous “chart crime” incident.
A technical glitch that made GPT-5 seem “Dumber”
One of GPT-5’s key innovations is its real-time router technology, which automatically determines the most suitable model to handle a user’s request — whether to respond quickly or take extra time for deeper reasoning.
However, on launch day, the router failed to function properly. As a result, many users felt GPT-5 was noticeably “dumber” than expected. Altman explained:
“Yesterday, the autoswitcher wasn’t working, which made GPT-5 seem much less capable. Starting today, it will feel smarter. We’re also going to make it more transparent which model is answering your query.”
The possible return of GPT-4o
Many users expressed nostalgia for GPT-4o’s style and performance. Altman responded by saying that Plus subscribers might regain access to the older model, pending further evaluation of trade-offs.
He also announced plans to double the prompt limits for Plus subscribers, allowing them to experiment with and adapt to GPT-5 without worrying about hitting monthly usage caps.
The “Chart Crime” that went viral
During the GPT-5 launch presentation, one slide showed a bar chart in which a smaller numerical value was represented by a taller bar — instantly earning the nickname “chart crime.”
Altman later acknowledged the mistake on X (formerly Twitter), calling it a “mega chart screwup.” While the official blog post contained the correct charts, the on-stage mishap had already sparked memes and jokes across social media.
AI reviewer Simon Willison, who had early access to GPT-5, praised its performance overall but pointed out that converting data into tables was still a weak spot for the model.
Final notes
Altman closed the AMA with a promise:
“We will continue to work to get things stable and will keep listening to feedback.”
For OpenAI, GPT-5 represents both a major technical leap and a public trust challenge. On one hand, features like the router showcase a forward-looking vision for adaptive AI. On the other, even a single glitch can have a significant impact on user experience, proving that innovation must be paired with stability.
