Sam Altman addresses ‘uneven’ GPT-5-ORDER that brings 4o back, and ‘chart crime’

During a Reddit Ask-Me-Anywing session on Friday, Openai CEO Sam Altman and key members in the GPT-5 team were pepped with questions about the new model and requests to bring back his previous model, GPT-4o.

They also asked Altman about the most embarrassing – and perhaps funniest – snafu in the presentation, “Chart Crime.”

One of the new features that GPT-5 rolled out is a real-time router that decides which model to use for a particular prompt, either responds quickly or takes further time to “think” through answers.

But several people in AMA on R/Chatgpt Reddit complained that GPT-5 didn’t work as well for those as 4o did. Altman said the reason why the GPT-5 thought “Dumber” was the router did not work properly when it was rolled out on Thursday.

“GPT-5 will seem smarter that starts today. Yesterday we had a SEV and Autoswitcher was out of commission for part of the day, and the result was GPT-5 as a dumber. We also make some interventions to how the decision-making limit works to help you get the right model more often. We will make it more transparent about which model corresponds to a given request,”

Still lobbbled people at Ama so hard to bring 4o back to plus subscribers, as Altman promised at least to investigate it. “We are looking at letting Plus users continue to use 4o. We are trying to collect more data on trade -offs,” he wrote.

And Altman also promised, “We will double -rate limits for plus users when we’re done rolling out.” This should give people a chance to play and learn the new model, adopt it to their use cases without worrying about running out of monthly requests.

Predictably, he was also asked about the wildly inaccurate diagram that the team presented during the live presentation that quickly became the back of many “card crime” jokes. The diagram presented a lower benchmark score with a much higher bar.

Altman did not answer questions about the chart under AMA, but on Thursday he called the chart for a “Mega -Diagram screw” on X. Others noticed that the charts in the published blog post were correct.

But the damage was done. Jokes followed to use GPT for charts in a company presentation. GPT-5 reviewer Simon Willison, who had early access and generally liked the model’s performance, also pointed out that transforming data into a table was “good example of a GPT-5-Fiasko.”

In any case, Altman promised corrections to the things that seemed to worry people the most. He ended Ama with a promise: “We will continue to work to get things stable and will continue to listen to feedback.”