OpenaiThe SPP-5 model was an update that changes worldwide to its very popular and early chat chat. But for some users, last Thursday’s launch was more like an exciting rebate, with the new Chatgpt with a diluted personality and made surprisingly dumb mistakes.
On Friday, Openai CEO SAM Altman took X to say that the company would maintain the previous model, GPT-4O, which was executed to more users. Altman said that a new feature designed to change perfectly between models according to the complexity of the consultation was broken on Thursday, “and the result was GPT-5. He promised to implement solutions to improve GPT-5 performance and the user’s general experience.
Given the hype around GPT-5, some level of disappointment seems inevitable. When Openai introduced GPP-4 in March 2023, he stunned the IA experts with his incredible skills. GPT-5, speculating the spikes, would probably be so fallen from the jaw.
Openai presented the model as a significant update with doctoral intelligence and virtuous coding skills. A system to automatically direct consultations to different models aimed to provide a softer user experience (it could also save the company by directing simple queries to cheaper models).
Shortly after the fall of GPT-5, a Chatgpt’s Reddit community was filled with complaints. Many users cried the loss of the old model.
“For some days now I have been trying GPT5. Even after customizing the instructions, he is not yet feeling the same. He is more technical, more widespread and honestly feels emotionally away,” a member of the community wrote in a thread entitled “Kill 4o is not innovation, it is to erase.”
“Sure, 5 is fine, if you hate nuances and feeling things,” said another Reddit user.
Other threads complained about slow answers, hallucinations and surprising errors.
Altman promised to address these problems by duplicating GPT-5 rate limits for Chatgpt Plus users, improving the system that changes between models and allowing users to specify when they want to trigger a “more weighted and capable mode of thought”. “We will continue to work to get things stable and we will continue to listen to comments,” wrote the CEO on X. “As we mentioned, we expected a little overload as we roll (Ed) so much at once. But it was a bit blurred than we expected!”
The errors published on social media do not necessarily indicate that the new model is less capable than its predecessors. They may simply suggest that the new model occurs in different cases of edge than the previous versions. Openai refused to comment specifically why GPT-5 sometimes seems to make simple blunders.
| Do you have a tip? |
|---|
| Are you a current or old Openai employee who wants to talk about what’s going on? We would like to know about you. Using a telephone or computer non -work, please contact the reporter with sign on Wak.01. |
The decline has caused a new debate on the psychological attachments that some users form with Chatbots formed to push their emotional buttons. Some Reddit users dismissed complaints about GPT-5 as evidence of an unhealthy dependence from a AI colleague.
In March, Openai published investigations that explored users of emotional bonds with their models. Shortly afterwards, the company issued an update to GPT-4O, after it became too symphantic.
“It seems that GPT-5 is less friendly, more” business “and less talk,” says Pattie Maes, a MIT teacher who worked in the study. “I personally think of this as a good thing because it is also what caused delusions, bias reinforcement, etc. but, unfortunately, many users like a model that tells them that they are smart and amazing, and this confirms their opinions and beliefs, even if (they are) wrong.”
Altman stated in another publication in X that this is something that the company fought in the GPT-5 construction.
“Many people actually use Chatgpt as a kind of therapist or life coach, even if they would not describe it,” Altman wrote. He added that some users may use Chatgpt in ways to help improve their lives, while others could “not know how far away from their long-term well-being.”
