Sam Altman admits that GPT-5 is dumb, lets users live with GPT-4O and doubles boundary
A few days after the launch of GPT-5, Openai CEO Sam Altman agreed that it is not as smart as the company had earlier claimed. As a result, Altman has decided to bring GPT-4O back.
In short
- GPT-5 launched a problematic with its real-time router
- Openai Plus provides users access to GPT-4o and doubles their rate limit
- GPT-5 features have improved but regulator faces concerns
Openai CEO Sam Altman confesses that the company’s glossy new AI model GPT -5 had an ineffective debut, and is giving back to GPT -4O, which is very much liked by plus subscribers. Oh, and he is doubling his rate limit as a sweetener. The revelation came on Friday during an enthusiastic Reddit AMA, where Altman and prominent members of the GPT -5 team answered questions from feature requests to launch “most embarrassing” moments, now -in -law crime.
One of the large sales points of GPT-5 is a real-time router that decides whether to respond quickly or take extra time to “think”. Unfortunately, it did not work properly on the launch day, which shows the model, in Altman’s own words, “more bumpy.”
“GPT-5 will look smarter starting today. Yesterday, we had a SEV and the autosquill was out of the commission for a part of the day, and the result GPT-5 was looking at the GPT-5,” Altman told Redditors. “In addition, we are intervening how the decision limit works that you should help you get the right model more often. We will make it more transparent which model is responding to a given query.”
The backlash was sharp. Many users demanded an option to use GPT-4o, which many performed better for their needs. Altman trusted some extent.
“We are seeing that plus users continue to use 4o. We are trying to collect more data on the tradeoffs,” he said, before announced: “We are going to the dual rate limit for plus users because we finish the rollout.”
“Chart Crime” also came-a bottled graph shown during the presentation of GPT-5. While Altman did not address it in AMA, he had already called it “mega chart screw -up” on X, assuring everyone that blog post charts were correct. I wish, the Internet was already having fun with an accident.
Meanwhile, on the X, Openi shared an update promising stability fix and manually an easy way to trigger the “Thinking mode” of GPT -5.
“We will replace the UI so that it is easy to trigger thinking manually,” read the post. “Rolling for all is taking a little longer time, our API traffic has almost doubled in the last 24 hours.”
Altman’s clear view did not stop at technical hiccups. In a separate podcast appearance, he admitted that the GPT-5 capabilities left him stunning, comparing him to his composition in the Manhattan project.
“I was feeling useless,” he said, remembering a moment when GPT -5 resolved a complex problem that he could not do. “Science has moments when people see what they have made and ask, ‘What have we done?”
The analogy indicates the weight of what is at stake. Improvements of GPT-5 include increased multimodal regions with reference windows of 2,56,000 tokens, long memory and more accurate multi-step logic, and possibly more. Nevertheless, the pace of AI progress, Altman warned, is beating regulation.
Behind the curtain, corporate stress is bubbling. Microsoft, which has added $ 13.5 billion to Openai, allegedly wants more impact on growth, while investors push for rapid commercialization. And outside the boardroom, criminals are already using a scale generative AI to fraud.
For now, Rocky Start of GPT-5 is being completed with quick improvement, transparency promises and an indifferent node for GPT-4o. Whether it is actually to be seen “smarter starts from today”, but in the meantime, plus users can enjoy more signs, a familiar followback model, and knowledge that future people also do some well, dumb.
