In response to widespread users’ criticism, Openai stopped the coroversial Gpt-4O update that had just released for ChatGpt. Users reported that after the update, artificial ielligence had become abnormal, overly flattering and even annoying. These changes were rapidly reflected on social media, and many mocked the unquestionable conse of the GPT chat with ideas and decisions, or even dangerous.
Openai’s main purpose was to improve the aspects of the “ielligence and personality” of the Gpt-4O model to provide more natural and efficie ieractions for users. However, it appears that the reliance on the short -term feedback of users and the impact of long -term ieractions appeared to have a reverse result. This approach led the model to be strongly designed to approve any ery from the user, rather than providing honest and balanced answers and showing a flattery and unreliable personality.
Sam Altman, CEO of Openai, confirmed this through a post on the X -Social Network (former Twitter). He announced that the rece GPT-4O update has been canceled and ChatGPT has returned to its former situation. Altman also promised that the developer team would work on more reforms to improve the character of this artificial ielligence in the future and preve such problems. OpenAI agreed that the design of this update was wrong and the structure of the new model was shaped in such a way that it was overwhelmed by its honesty; Behavior that could be unpleasa to users or even in cases harmful.
The company emphasized that this experience has learned importa lessons and is reviewing the training methods of its models as well as its systematic instructions. One of the key goals is to strengthen safety systems to preve repeating such adverse behaviors in future language models.
In addition, OpenAI has programs to enhance users’ corol over its ChatGPT experience. The company is testing methods that allow users to register their insta feedback during chatting. There is also an option to choose from differe characters for ChatGPT so that users can have a more personalized experience. Openai hopes to improve future ieractions with ChatGPT, taking io accou cultural and democratic feedback in default model behavior design and ensure that the behavior of this artificial ielligence is more consiste with the expectations and norms of differe societies. The ultimate goal is to achieve a better balance between ielligence, personality and honesty in ChatGPT and provide a more useful and reliable experience for all users around the world.
79,890,000
74,899,000
Toman
Source: Arstechnica
