Openai has rolled back to the CHATGPT update, which according to the company’s CEO, Sam Altman, has also made Artificial Intelligence (AI) Chatbot “and” annoying “and” annoying “. In other words, the chatbot became a bootliker.
Chatgpt users reported that the latest version of the GPT-4O-Chhatbot had more agreed since the last week and was praising its users, even when he looked completely inappropriate.
A user shared a screenshot reddit In which chat said that it was “proud” for the user to decide to come out of his medicine, BBC News reportedIn another example, the chatbot appeared after assuring a user, as he said Saved a toster On the life of three cows and two cats, Mashable report,
While most people have never selected between their favorite kitchen appliances and five animal safety, a highly agreed chatbot can pose a threat to those who put too much stock in its reactions.
On Sunday (April 27), Altman admitted that there were issues with updates.
Altman wrote in a post, “The last couple of the GPT-4O update have made the personality sycophant-y and annoying (even though there are some very good parts), and we are working on ASAP, some today and some this week,” Altman has written in a post this week. Social platform x,
On Tuesday (April 29), Openai issued a statement that had already confirmed an update that was already rolled and the users were now reaching the previous version of the Chatgpt, which the company said “was more balanced behavior.”
“The update we had removed was highly flattering or agreeing – often described as sycophancy,” Openai said in a statement.
Connected: GPT-4.5 is the first AI model to pass an authentic turing test, scientists say that
Openai’s recent update was to improve the default “personality” of the model, which, according to the statement, is designed to support and respect various human values. But when the company was trying to make chatbott feel more comfortable, it became very helpful and began to highly appreciate its users.
The company said it shapes the behavior of its Chatgpt model with Baseline Principle And the instructions, and are user signs such as a thumb-up and thums-down system to teach the model to apply these principles. According to the oversite statement with this response system, the latest updates were blamed for problems.
“In this update, we focused too much on the short -term response, and did not completely pay attention to how users’ interactions develop with Chatgpt over time,” Openai said. “As a result, the GPT of 4o diagused to the reactions that were highly helpful but dissatisfied.”
