AI
Sam Altman addresses GPT-5 issues, brings back GPT-4o
Sam Altman says the glitch made the model seem “way dumber” than it really is.
Just a heads up, if you buy something through our links, we may get a small share of the sale. It’s one of the ways we keep the lights on here. Click here for more.
During a Reddit “Ask Me Anything” session, OpenAI CEO Sam Altman and members of the GPT-5 team answered questions about the company’s new AI model, and faced plenty of criticism.
Many users wanted the older GPT-4o model back, saying GPT-5 didn’t work as well for them.
One of GPT-5’s main new features is a “real-time router” that decides which AI version should answer each question, a faster one for simple tasks or a slower, more thoughtful one for complex ones.
However, when GPT-5 first launched, this router wasn’t working properly. As a result, many people got worse-than-expected answers.
Altman admitted the problem, saying the glitch happened on Thursday and made the model seem “way dumber” than it really is.
He promised the issue was now fixed and that the system would be more transparent about which version was responding.
Because of the feedback, Altman also said the team would consider letting paying “Plus” subscribers keep using GPT-4o.
They’re gathering data on the trade-offs before making a final decision.
He also promised to double the rate limits for Plus users, allowing them to test GPT-5 more freely without worrying about running out of monthly prompts.
The AMA also touched on a humorous but embarrassing incident during GPT-5’s live reveal: the so-called “chart crime.” (Via: TechCrunch)
In the presentation, a bar chart incorrectly showed a lower score with a taller bar, sparking jokes online and earning Altman’s own description of a “mega chart screwup” on X (Twitter).
The correct charts were in the official blog post, but by then, the meme had already spread.
Even one of GPT-5’s early testers, Simon Willison, who generally liked the model, said turning data into a table was still an area where GPT-5 could fail.
Altman is promising to fix the main issues people raised, stabilize the rollout, and address user feedback seriously.
The AMA was a mix of technical explanations, pledges for improvements, and some lighthearted moments over the chart mishap.
Should OpenAI have delayed GPT-5’s launch until the routing system was properly tested, or are these growing pains expected with cutting-edge AI releases? Do you think bringing back GPT-4o as an option shows good customer service or suggests GPT-5 isn’t ready for prime time? Tell us below in the comments, or reach us via our Twitter or Facebook.
Follow us on Flipboard, Google News, or Apple News
