Sam Altman addresses the “abnormal” GPT-5 Rollout, bringing the 4th back and “Crime of the Map”

During a reddit Ask-me-what Meeting on Friday, Openai’s chief executive and key members of the GPT-5 team were full of questions about the new model and requests to restore its previous model, GPT-4O.

They also asked Altman for the most annoying – and perhaps more funny – SNAFU in the presentation, the “graph crime”.

One of the new features launched by the GPT-5 is a real-time router that decides which model will use for a particular prompt, whether it responds quickly or takes extra time to “think” through answers.

But many people in AMA in R/Chatgpt Reddit complained The GPT-5 did not work for them too, as the 4th did. Altman said the reason GPT-5 seemed to be “Dumber” was that the router did not work properly when it was released on Thursday.

“The GPT-5 will look smarter than today, yesterday. We had a SEV and Autoswitcher was out of supply for a part of the day and the result was that the GPT-5 appeared in the way, we will also make some interventions on how the limits of the decision to help you get the right model more often.

Still, people on AMA have been pressing so hard to bring back 4 for plus the subscribers that Altman promised to consider at least. “We are considering leaving users and continuing to use the 4th. We try to gather more data on the agreements,” he wrote.

And Altman also promised: “We will double the interest rates for users as we finish growth.” This should give people the opportunity to play and learn the new model, to adopt it in the case of their use without worrying about the exhaustion of monthly prompts.

Projected, was also asked about the wild inaccurate graph presented by the team during the live presentation that quickly became the end of many “Crime crime” jokes. The diagram presented a lower reference score with a much higher bar.

Altman did not answer questions about the diagram during AMA, but on Thursday he called the “Mega Chart Curtup” chart to x. Others marked the charts in The published blog post was correct.

But the damage was done. The jokes followed the use of GPT for diagrams in a corporate presentation. The GPT-5 Simon Willison critic, who had timely access and generally liked the performance of the model, Also points out That the conversion of the data into a table was “a good example of GPT-5 failure”.

In any case, Altman promised corrections to objects that seemed to be more concerned. Ama finished with a promise: “We will continue to work to do things steady and continue to hear comments.”