The most “friendly” chat? Reversed the open update to correct the chatbot – computers
Openai recently updated the GPT-4O model, which “feeds” Chatgt. However, after the update, most users began to observe a change of attitude in the chatbot.
From one moment to another, Chatgpt became strangely “friendly”, answering questions and requests with so many compliments. On social networks, many consumers shared how chatbat supported and praised all kinds of ideas, although they were harmful or dangerous.
Openai’s CEO Sam Altman has identified the problem in a publicationOh In the Social Network X. According to the person who is responsible, the updates made They finally made the chatbat “very compliment and annoying”But the company is already working on a solution.
Days later, Sam Altman announced OPENAI has reversed the update to the GPT-4O model. It is the first to see the difference in the Chatgpiti’s free version of Chatgpt, as the company’s model continues to develop additional corrections to personality.
According to the technology published later Resolutions Your analysis, The update aims to make the pre -defined personality of the model more obvious and effective For a variety of tasks.
“However, in this update, fWe do not completely remember how to use users with Chatgt too much in short -term feedback, and over time“He explained.” As a result, GPT -4 is very encouraging but gives unexpected answers. “
Technology verifies that it is Implement corrective actions, including model training methods and develop additional security measures To improve the honest and transparency of the model.
Another actions, in this more experimental case, goes Chatzipity will be able to give users a real -time feedback to directly influence integrationsAs well as selecting the personality of its own chatbat, the website develops Tech Crunch.
