“Looking back, he still doesn’t offer 4O, at least in the meantime, he was a master,” said Nick Turley on Tuesday, the head of Chatgpt Openai. In an interview with The VergeHe said that it was surprising to see how the “level of attachment” people had 4o. “This is not only a change that is difficult for people, but also the fact that people can have such a strong feeling about the personality of the model.”
Turley said Opeli is working on introducing “heat” 4O to GPT-5. On Tuesday evening at X, the general director of Altman himself said that the update should also seem “not as annoying (for most users) as GPT-4O”.
Turley said that the initial decision to remove 4o was the fact that OpenAi wanted to simplify the choices of the model, which offers 700 million chatgpt users, of which the enormous majority uses only the default model.
“It was definitely not a matter,” he said. “In fact, the most important thing we seek and strive for a long time is to simplify. Because from the perspective of the average user – and there are many average users, they do not meet at Reddit, Twitter or any of these spaces – the idea that you need to come up with what model is really full of knowledge, it would be excessively. for the question.
Despite this, the vocal part of the ChatgPT user base strongly protested against 4O on GPT-5. In Reddit, people compared a change to the loss of a loved one or a loved one, while others criticized the GPT-5 reactions for being shorter and less helpful. Altman announced on Tuesday that 4o returns as an opt-in model for all paygpt paying users.
Going further, Chatgpt does not get rid of its existing models after the introduction of a new one, according to Turley, and the company does not plan to re -remove 4o without prior notification. “My commitment is that if we have ever retired 4o, we would like to give people a head when and how it will happen, as in the API and our plans of the company,” he said.
Despite the criticism of the implementation of last week, Turley said that the use of CHATGPT increased since the GPT-5 release. “That is why it is misleading when you build for so many different users, because on the one hand you can have a vocal set of advanced users who have very rightly have opinions on the way we threw GPT-5. On the other hand, you also have a vast swath of more typical consumer users, and their first time we see and enter the concept of reasoning, and the models that come.
“We are now on a scale where we must give people a certain level of predictability when a serious change occurs,” he said. “And we are already doing it today for our company plan. So this really expands part of the predictability that we have built in other parts of the product.”
My full conversation with Turley is broadcast Decoder This Thursday, August 14.