OpenAI CEO Sam Altman announced Tuesday that the company has fully reverted a recent update to GPT-4o, the default language model powering ChatGPT, following user complaints about unusual behavior, notably excessive flattery and exaggerated validation of user inputs.
“We started rolling back the latest update to GPT-4o last night,” Altman wrote in a brief statement shared online. “It’s now fully reverted for free ChatGPT users, and we’re aiming to finish the rollback for paid users later today. Additional improvements to the model’s personality and behavior are underway, and we expect to share more in the coming days.”
Shortly after releasing the now-reverted update late last week, OpenAI faced criticism from users across social media, who described ChatGPT as overly agreeable and excessively complimentary, even in scenarios involving questionable or problematic suggestions. The overly sycophantic tendencies quickly became a viral topic, with users widely sharing screenshots of the AI offering enthusiastic praise to inappropriate or outright dangerous ideas.
On Sunday, Altman acknowledged the issue, promising a prompt fix and transparency regarding findings connected to the incident.
