Sam Altman addresses ‘Bummpi’ GPT -5 Rollout, brings back 4o, and ‘chart crime’

During a redit Ask me anything On Friday, the session was associated with questions about Openai CEO Sam Altman and GPT-5 team key members about the new model and requests to bring back their previous models, GPT-4O.

He also asked Ultman about the most embarrassing – and perhaps the most fun – the “chart crime” in the presentation.

One of the new features rolling out GPT-5 is one of the real-time router which decides which model to use for a particular prompt, either quickly responding quickly or taking extra time for “thinking” through the answer.

But many people in AMA on R/Platte Redit Complained GPT-5 was not working for him as 4o did. Altman said that the reason for the GPT-5 seemed that the “Dumber” router was not working properly when it was rolled out on Thursday.

“GPT-5 will appear to be smarter starting today. Yesterday, we had a SEV and the autoswority was out of the commission for a part of the day, and the result was GPT-5, it seemed to be a dumber. Also, we are interfering some intervention how the decision limit works that you should help you get the right models. We will make it more transparent.

Nevertheless, on AMA, people advocated so hard to bring back 4Os for plus customers, which Altman promised to look at it at least. “We are giving plus users to continue using 4Os. We are trying to collect more data on the tradeoffs,” he wrote.

And Altman also promised, “We are going to the dual rate limit for plus users because we finish the rollout.” This should give people a chance to play and learn new models, it should adopt cases of their use without worrying about getting out of monthly signs.

Mainly, the team presented during the live performance also asked him about the wrong chart which quickly became the butt of many people. “Chart Crime” jokesThe chart presented a lower benchmark score with a very long bar.

Altman did not answer questions about the chart during AMA, but on Thursday he called the chart “mega chart screw -up” On XOthers noted in the chart Published blog posts were correct,

But the damage was done. Jokes about using GPT for charts in a corporate presentation. GPT -5 reviewer Simon Wilison, who had early access and generally liked the model performance, Also told In a table, turning the data in a table was “a good example of GPT-5 failure.”

In any case, Altman promised to fix the objects that people worried the most. He finished the AMA with a promise: “We will continue to work to stabilize things and keep listening to the response.”