OpenAI Scrambles to Update GPT-5 After Users Revolt
OPENAIThe GPT-5 model was supposed to be an upgrade that changes the world to its extremely popular and early chatbot. But for some users, the version of last Thursday looked more like a heartbreaking degradation, the new chatgpt presenting a diluted personality and making surprisingly stupid errors.
Friday, the CEO of Openai, Sam Altman, went to X to say that the company would keep the previous model, GPT-4O, running for more users. A new feature designed to switch transparently between the models according to the complexity of the request that broke Thursday, said Altman, “and the result was GPT-5 seemed much more stupid.” He promised to implement fixes to improve GPT-5 performance and global user experience.
Given the media threshing around GPT-5, a certain level of disappointment seems inevitable. When Openai introduced the GPT-4 in March 2023, he amazed AI experts with his incredible capacities. The GPT-5, according to experts, would surely be breathtaking.
OPENAI presented the model as a significant upgrade with skills in intelligence at doctorate and virtuoso coding. A system to automatically transport queries to different models was intended to provide a more fluid user experience (this could also save money to the company by directing simple requests to cheaper models).
Shortly after the drop in GPT-5, a reddit community dedicated to Chatgpt filled with complaints. Many users have cried the loss of the old model.
“I have been trying GPT5 for a few days now. Even after the personalization of the instructions, it still does not feel the same thing. It is more technical, more generalized and feels honestly emotionally distant,” wrote a member of the community in a thread entitled “Kill 4o is not innovation, it is erasing.”
“Of course, 5, that’s good – if you hate nuances and feel things,” wrote another Reddit user.
Other wires complained about slow responses, hallucinations and surprising errors.
Altman has promised to solve these problems by doubling the GPT-5 rate limits for chatgpt users, the improvement of the system that rocks between models and allowing it to specify when they want to trigger a heavier and more capable “thinking”. “We will continue to work so that stable things and will continue to listen to the comments,” wrote the CEO on X. “As we mentioned, we expected a little bump[ed] Out so much at the same time. But it was a little more joyful than we hoped! »»
Errors published on social networks do not necessarily indicate that the new model is less capable than its predecessors. They can simply suggest that the brand new model is triggered by different cases of edge than the previous versions. Openai refused to comment specifically why the GPT-5 sometimes seems to make simple blunders.
The backlash has sparked a new debate on the psychological attachments that some users form with chatbots formed to push their emotional pimples. Some Reddit users have rejected complaints concerning GPT-5 as proof of unhealthy dependence on an AI companion.
In March, Openai published research exploring the emotional links that users are formed with its models. Shortly after, the company published an update of GPT-4O, after it became too sycophantic.
“It seems that the GPT-5 is less sycophetic, more” commercial “and less talkative”, explains Pattie Maes, professor at MIT who worked on the study. “I personally think of that as a good thing because it is also what led to delusions, to a reinforcement of biases, etc. But unfortunately, many users like a model that tells them that they are intelligent and incredible, and which confirm their opinions and beliefs, even if [they are] fake.”
Altman indicated in another article on X that it is something with which the company fought to build GPT-5.
“Many people effectively use Chatgpt as a kind of therapist or life coach, even if they would not describe this in this way,” wrote Altman. He added that some users can use Chatgpt in a way that helps improve their lives while others could be “unconsciously far from their longer-term well-being”.
