Openai published an autopsy on the default AI model that powers GPT-4O, forcing the company to update the model released last week.

Over the weekend, after the GPT-4O model update, users on social media noted that Chatgpt began responding in an over-verified and pleasant way. It quickly became a meme. Users posted screenshots of Chatgpt, applauding various problematic dangerous decisions and ideas.

In a post on Sunday X, CEO Sam Altman acknowledged the issue and said Openai will fix it “as soon as possible”. Two days later, Altman announced that the GPT-4O update was retreating and Openai was making "other fixes" for the model's personality.

According to Openai, the update aims to make the model's default personality "feel more intuitive and effective", which is for "short-term feedback" and "how it's not sufficiently explained how users' interactions with Chatgpt over time".

Last week we rolled back the GPT-4O update in Chatgpt because it was too flattering and enjoyable. You can now access older versions with more balanced behavior.

More about what happened, why it matters and how we solve sicophancy: https://t.co/lohou7i7dc

- Openai (@openai) April 30, 2025

"So, GPT-4O is biased towards oversupporting but unwise responses," Openai wrote in a blog post. "The interactions of prostitutes can be uncomfortable, disturbing and troubled. We're falling and trying to get it right."

Openai said it is implementing multiple fixes, including refining its core model training techniques and system tips, explicitly promoting GPT-4O away from adhesives. (The system prompt is an initial description that guides the model’s overall behavior and tone in its interaction.) The company is also building more security guardrails to “improve (model’s) honesty and transparency” and continue to expand its assessment to “help identify issues beyond candy.”

Openai also said it is trying to get users to provide “real-time feedback” to “directly influence their interaction with Chatgpt” and choose from multiple Chatgpt personalities.

"(w)E'Re explores new ways to incorporate broader democratic feedback into Chatgpt's default behavior," the company wrote in a blog post. "We also believe that users should have more control over how ChatGpt behaves and to some extent safe and feasible, adjusting if they disagree with the default behavior."