OpenAI rolls back update that made ChatGPT an ass-kissing weirdo

OpenAI has rolled back a recent update to GPT-4o, the default model that powers ChatGPT, following complaints from users that it made the chat bot act weird. "The last couple of GPT-4o updates have made the personality too sycophant-y and annoying (even though there are some very good parts of it), and we are working on fixes asap, some today and some this week," said OpenAI CEO Sam Altman in a X post spotted by TechCrunch.

As of late Tuesday, OpenAI said ChatGPT was running on an older, less sycophantic version of GPT-4o for all users. "We are actively testing new fixes to address the issue," the company said in a blog post. "We’re revising how we collect and incorporate feedback to heavily weight long-term user satisfaction and we’re introducing more personalization features, giving users greater control over how ChatGPT behaves."

OpenAI released the new GPT-4o late last week. By the weekend, people began noticing ChatGPT was being overly agreeable and verbose in its praise. As you can see from the X post below, often that praise was also inappropriate and made people feel uncomfortable.

In the postmortem OpenAI published on Tuesday, the company said the issue was due to changes it had made to GPT-4o designed to make the model feel more intuitive. "In this update, we focused too much on short-term feedback, and did not fully account for how users’ interactions with ChatGPT evolve over time. As a result, GPT‑4o skewed towards responses that were overly supportive but disingenuous," the company said.

"ChatGPT’s default personality deeply affects the way you experience and trust it. Sycophantic interactions can be uncomfortable, unsettling, and cause distress. We fell short and are working on getting it right."

To avoid a similar situation in the future, OpenAI says it plans to refine its training techniques and prompts to guide models away from sycophancy. It also plans to implement additional guardrails during model training. OpenAI says those guardrails should increase honesty and transparency. At the same time, the company plans to give users more ways to test and provide feedback on new releases before they're available to the public. Lastly, OpenAI says it will work on giving people more ways to tweak ChatGPT's default behavior.

Update 04/30/25 10:00AM ET: Added more context after OpenAI published a blog post detailing what went wrong with the latest GPT-4o release.