ChatGPT criticized for a perceived decline in performance.
Users of ChatGPT have expressed dissatisfaction with a recent incident where the AI bot has been advising them to handle their tasks as if it were their supervisor. This has prompted OpenAI to launch an investigation.
On Thursday, the company announced that it is examining reports indicating that ChatGPT has started to reject user requests by suggesting they complete tasks themselves or outright refusing to assist – all while refraining from sending an out-of-office email from a beach in Cabo.
Through the ChatGPT account on X, OpenAI stated that it is actively seeking feedback on the observed trend of the model seeming to become ‘lazier.’
“We haven’t made any updates to the model since Nov 11th, and this behavior is certainly not intentional,” the company explained. “Model behavior can be unpredictable, and we are actively working to address and rectify the issue.
ChatGPT, hailed as a groundbreaking tool for those seeking to play solitaire at work while delegating tasks, has amassed an estimated 1.7 billion users since its launch in November of the previous year. During this period, research indicated that ChatGPT significantly enhanced productivity for many users, enabling them to produce higher-quality work.
However, users are now reporting instances where the bot, designed to simplify their lives, responds with an unexpected touch of sass.
For instance, according to Semafor, a startup founder requested the bot to list the days of the week leading up to May 5, only to receive a response claiming it couldn’t generate an “exhaustive list.” When Business Insider conducted a similar test, ChatGPT not only provided detailed instructions on calculating the number of weeks between December 9 and May 5 but also supplied a direct answer.
Reddit users express frustration over the arduous process of coaxing ChatGPT into delivering suitable responses to assigned tasks, often resorting to cycling through different prompts until achieving the desired outcome. A significant portion of the grievances revolves around ChatGPT’s coding capabilities, with users advocating for a return to the original GPT models. Additionally, users note a perceived decline in the quality of responses from the AI.
Workers have previously linked certain problems to a software bug, but as of Saturday, OpenAI has clarified that it is actively examining user grievances. In a statement on X, the company emphasized that the training process can lead to the development of diverse personalities within its models.
OpenAI has not yet provided a prompt response to a request for comment.
“Training chat models is not a clean industrial process. different training runs even using the same datasets can produce models that are noticeably different in personality, writing style, refusal behavior, evaluation performance, and even political bias,”
