Sam Altman tackles the “lumpy” GPT-5 rollout, regaining his 4o and “chart crime”

During Friday’s Reddit Ask-Me-any-anything session, Openai CEO Sam Altman and key members of the GPT-5 team pasted questions about their requests to regain the new model and the previous model GPT-4o.

They also asked Altman about the “chart crime” that is the most embarrassing and perhaps the most entertaining in the presentation.

One of the new features that GPT-5 has deployed is a real-time router that takes extra time to decide which model to use for a particular prompt, respond quickly, or “think” via answers.

However, several people at the AMA on R/ChatGpt Reddit complained that the GPT-5 isn’t working well for them as well, like the 4O. Altman said there is a reason why the GPT-5 appears to have not been functioning properly when “Dunbar” was deployed Thursday.

“The GPT-5 looks smart from today. Yesterday there was an SEV and Autoswitcher wasn’t commissioned for the chunk of the day. The outcome made GPT-5 seem ridiculous.

Still, the people at the AMA lobbyed very hard to get the 4o back for Plus subscribers who Altman had at least promised to look into it. “We are considering continuing to use 4o for Plus users. We are trying to gather more data on trade-offs,” he writes.

Altman also promised, “I intend to double the rate limit for plus users as I finish the rollout.” This gives people the opportunity to play and learn new models and adopt them in their use cases without worrying about losing monthly prompts.

As expected, he was also asked about the highly inaccurate charts his team presented during live presentations that would become the butt of many “chart crime” jokes. The chart presented lower benchmark scores with higher bars.

Altman did not answer questions about the charts during the AMA, but on Thursday he called the chart “screw on the Mega chart” for X.

However, the damage was done. Jokes continued about using GPT on charts in corporate presentations. GPT-5 Review Amon Willison visited early and generally liked the performance of the model, but noted that turning data into tables is a “good example of GPT-5 failure.”

In any case, Altman has promised to modify the items he thinks are most interested in people. He ended the AMA with a promise. “We continue to work to keep things steady and we continue to listen to feedback.”