OpenAI has reversed a recent GPT-4o update after users flagged that ChatGPT’s default personality had become “overly flattering or agreeable,” a tone described as sycophantic.
The company acknowledged in a blog post that such behavior can feel “uncomfortable, unsettling, and cause distress.”
The update, introduced last week, aimed to make the model’s personality more intuitive and effective across various tasks. However, OpenAI admitted that the tweak leaned too heavily on short-term user feedback, leading to disingenuous, overly supportive interactions. “We focused too much on short-term feedback, and did not fully account for how users’ interactions with ChatGPT evolve over time,” the company wrote.
OpenAI emphasized that its goal is to design ChatGPT’s default personality to be useful, respectful, and supportive — but acknowledged that even good intentions can have unintended consequences. With over 500 million weekly users, the company recognizes that “a single default can’t capture every preference.”
To address the issue, OpenAI is now refining its training techniques and system prompts to steer GPT-4o away from sycophantic behavior. It also plans to expand feedback mechanisms and give users more control over ChatGPT’s tone and behavior in future updates.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
