OpenaiThe GPT-5 model was to be a change in the world of updating to the extremely popular and premature chatbot. But for some users, the release last Thursday seemed more that by hitting, and the novel chatgpt presented a diluted personality and makes surprisingly stupid mistakes.
On Friday, Openai General Director, Altman himself He started to x The statement that the company would stop the previous model, GPT-4O, acting for plus users. Altman said that a novel function designed to switch without any problems between models depending on the complexity of the query was broken on Thursday, “and the result was a GPT-5”. He promised to implement corrections to improve the GPT-5 performance and the user’s overall impression.
Given the noise around the GPT-5, a certain level of disappointment seems inevitable. When Opeli introduced GPT-4 in March 2023, he stunned AI experts about his amazing skills. GPT-5, speculated by experts, would certainly be just as stunning.
Opeli advertised this model as a significant update of intelligence at a doctor’s level and virtuoso coding skills. The system for automatically conducting queries to various models was aimed at ensuring a smoother user impression (it can also save company money by directing elementary queries to cheaper models).
Shortly after he dropped GPT-5, Reddit CHATGPT community filled with complaints. Many users mourned the loss of the elderly model.
“I have been trying GPT5 for several days. Even after adjusting the instructions, it still does not seem the same. It is more technical, more generalized and honestly seems emotionally distant,” wrote one of the members of the WA community thread The entitled “Kill 4o is not an innovation, it’s erasing.”
“Sure, 5 is fine – if you hate nuances and feeling things”, another Reddit user wrote.
Other threads He complained about leisurely reactions, hallucinations and surprising errors.
Altman promised to solve these problems by double the GPT-5 speed limits for ChatGPT Plus users, improving the system switching between models, and allowing users to determine when they want to run a more divorced and capable “thinking mode”. “We will continue to work to stability of the situation and listen to feedback,” he wrote CEO on X. “As we mentioned, we expected a certain bunch when we are going[ed] So many things at the same time. But it was a bit more bumpy than we expected! “
Errors published in social media do not necessarily indicate that the novel model is less capable than its predecessors. They can simply suggest that a completely novel model is stumbled by various edge cases than previous versions. Opeli refused to comment specifically why GPT-5 sometimes seems to make elementary mistakes.
The reaction caused a novel debate on psychological attachments, some users with trained chatbots to press their emotional buttons. Some Reddit users rejected complaints about GPT-5 as evidence of unhealthy dependence on his companion AI.
Opensi in March Published research Discovering emotional bonds create your models. Soon after, the company issued a GPT-4O update after it he became too sycophantic.
“It seems that the GPT-5 is less sycophantic, more” business “and less talkative,” says Pattie Maes, Professor Mit, who worked on the study. “I personally think it is a good thing, because it also led to illusions, strengthening of prejudice, etc. But unfortunately many users like a model that tells them that they are smart and amazing, and what confirms their opinions and beliefs, even if they are [they are] evil.”
Altman indicated in Another post on x That this is something that the company fought in building GPT-5.
“Many people effectively use chatgpt as a type of therapist or a life trainer, even if they didn’t describe it in this way,” Altman wrote. He added that some users can operate chatgpt in a way that helps improve their lives, while others can be “unknowingly distracted from long -term well -being.”