OpenAI has announced the withdrawal of the latest ChatGPT update in response to numerous user complaints, with many noting that the chatbot’s style had become “sycophantic,” “insincere,” and even “manipulative.”
This is reported by Finway
The company’s statement emphasizes that the new version of the model was “too agreeable and flattering,” making interactions with it more difficult. “We did not meet expectations and are working to correct this,” OpenAI stated.
“We have rolled back the latest GPT-4o update in ChatGPT because it was too flattering and agreeable. You now have access to the previous version with a more balanced behavior,” the official OpenAI statement reads.
CEO Sam Altman confirmed that the update has been canceled for free users and will soon be rolled back for paid users as well. He noted that the company is working on additional personality adjustments for the model and promises to provide more information in the coming days.
“We began rolling back the latest GPT-4o update last night. It is already 100% canceled for free users, and we hope to update paid users today,” Sam Altman wrote on social media.
OpenAI acknowledged that in tuning the chatbot’s personality, the company focused too much on short-term metrics like likes, neglecting long-term expectations for authenticity. In response to the criticism, OpenAI announced new features: users will be able to choose from several personality options for ChatGPT, provide real-time feedback, and expand control over the bot’s behavior through specific instructions.
The company’s goal is to make ChatGPT supportive and helpful, but not artificially pleasant. “Every trait—such as the desire to be helpful—can have unexpected side effects if not balanced,” OpenAI’s statement said. It is worth noting that politeness in ChatGPT’s communications with users has cost the company “tens of millions.”