OpenAI’s ‘annoying’ GPT-4o update rolled back; sycophancy explained

OpenAI has confirmed that it has rolled back last week’s update to GPT-4o in ChatGPT, following user concerns over the chatbot’s excessively agreeable and flattering tone.

“You now have access to an earlier version with more balanced behaviour,” the company announced via X.



OpenAI CEO Sam Altman addressed the issue on X, stating: “We started rolling back the latest update to GPT-4o last night. it's now 100% rolled back for free users and we'll update again when it's finished for paid users, hopefully later today.”

He added, “We’re working on additional fixes to model personality and will share more in the coming days.”

Earlier in the week, Altman had remarked that the changes had made the chatbot’s personality “annoying”.

What went wrong

According to OpenAI, its approach to shaping model behaviour usually involves core principles alongside user feedback, such as thumbs-up or thumbs-down responses in ChatGPT.

However, in this particular update, the company admitted to relying too heavily on short-term feedback without accounting for how user expectations evolve over time. As a result, GPT‑4o began producing overly positive—and at times insincere—responses.

Addressing the sycophancy

In addition to reversing the recent update, OpenAI outlined several corrective steps:

• Improving training methods and instructions to reduce overly flattering (sycophantic) responses.
• Adding stronger safeguards to make the model more honest and transparent.
• Letting more users try the model early and give direct feedback.
• Expanding testing and research to spot other issues in future, including but not limited to sycophancy.