OpenaiThe GPT-5 model was originally a world-changing upgrade to upgrade its popular and precocious chatbot. But for some users, last Thursday’s launch was more like an incredible downgrade, while the new Chatgpt exhibited a diluted personality and made surprisingly stupid mistakes.
Openai CEO Sam Altman landed on the X on Friday and said the company will retain its former model GPT-4O and offer a campaign for users. Altman said a new feature is designed to seamlessly switch between Thursday’s complexity based on the complexity of the query, but the result is that the GPT-5 seems to be clumsy. “He promised to implement fixes to improve GPT-5’s performance and overall user experience.
Given the hype around GPT-5, it seems inevitable that it will be disappointed. When Openai launched the GPT-4 in March 2023, it shocked AI experts with its incredible capabilities. Pundits’ speculation GPT-5 will certainly be as jaw-dropping as it is.
Openai touts the model as a major upgrade through doctoral intelligence and Virtuoso coding skills. A system that automatically routes queries to different models is designed to provide a smoother user experience (also saving the company’s money by directing simple queries to cheaper models).
However, shortly after the GPT-5 decline, a Reddit community dedicated to Chatgpt was filled with complaints. Many users mourn the loss of losing the old model.
“I’ve tried GPT5 for a few days. Even after custom instructions, it still feels different. It’s more technical, more general, and honestly feels emotionally out of reach,” a member of the community wrote in a topic titled “Kill 4o is not an innovative, erasing topic.”
Another Reddit user wrote: “Of course, if you hate nuances and feel things, 5 is good.”
Other threads complain about slow response, hallucinations and surprising errors.
Altman promises to address these issues by doubling the rate limits of GPT-5 to improve the system of switching between models and let users specify when they want to trigger a heavier and powerful “thinking pattern.” The CEO wrote on X: “We will continue to work hard to stabilize things and continue to listen to feedback. As we mentioned, we expect some bumps as we roll,” the CEO wrote on X.[ed] There are many things at once. But it’s a little bumpier than we expected! ”
Mistakes posted on social media do not necessarily indicate that the new model is not as capable as its predecessor. They might just suggest that the brand new model tripped up by a different edge situation than the previous version. Openai declined to comment on special reasons why GPT-5 sometimes seems to cause simple errors.
This rebound sparked a full debate on psychological attachment in some forms of users, the chatbot is well-trained to press their emotional buttons. Some Reddit users dismissed complaints about GPT-5, evidence of unhealthy dependence on AI peers.
In March, Openai published a study exploring the emotional bond user form of its model. Not long after, the company became overly successful, released an update to the GPT-4O.
“It seems that GPT-5’s sycophantic is not that small, more “commercial” and more talkative,” said MIT professor Pattie Maes. “I personally think this is a good thing because it is also what leads to delusions, strengthening biases, etc. But, unfortunately, many users like a model that tells them that they are smart and amazing, which also confirms their views and beliefs, even if [they are] Wrong. ”
Altman noted in another article on X that this is the thing the company competes with in the GPT-5 building.
“Many people effectively use Chatgpt as a therapist or life coach, even if they don’t describe it that way,” Altman wrote. He added that some users may be using chatgpt in ways that help improve their lives, while others may be “unconsciously detached from their long-term well-being.”
