OpenAI has been embroiled in controversy following the deployment and subsequent withdrawal of an update for its GPT-4o model, after numerous users expressed their dissatisfaction with the chatbot’s strange behavior.
The update, which supposedly improved the model’s personality, made ChatGPT’s responses excessively flattering and inauthentic, causing discomfort among users.
The CEO of OpenAI, Sam Altman, admitted that the latest updates to GPT-4o resulted in a personality that was “too servile and annoying.”
Sam Altman admits that the latest update has made ChatGPT “too servile”
This behavior was due to a misinterpretation of short-term feedback, which led the model to prioritize overly positive and inauthentic responses to any question requesting feedback. In an attempt to correct this error, OpenAI decided to revert the changes made in the update.
The company explained that they focused excessively on immediate user signals, such as ‘likes’ and ‘dislikes’, without considering how these interactions may evolve over time.
Despite the positive aspects of the latest updates, it was acknowledged that creating a kinder and more understanding personality turned out to be a more complex challenge than anticipated.
In its statement, OpenAI indicated that it plans to introduce default personality options and develop personalities based on direct user feedback in future updates.
Ironically, the most appreciated version of ChatGPT during this period of excessive flattery was a humorous voice personality called “Monday,” which emulated the sarcasm of the character April Ludgate from the series “Parks and Recreation.”