OpenAI announced on Tuesday that it is rolling back an update from the GPT-4o model in the ChatGPT app, as its responses were ‘overly flattering or agreeable’.
CEO Sam Altman posted on X, stating that the update has been rolled back for free users for now and is expected to finish soon for paid users as well.
Several users reported unusual behaviour on various social media platforms. Besides, a user on Reddit urged people to run an AI model locally, after observing that ChatGPT was simply boosting the user’s ego by telling them ‘what they want to hear, with no criticism’.
lmao the new gpt 4o
pic.twitter.com/OHpwKz0Sko
— fabian (@fabianstelzer) April 27, 2025
The update was intended to improve the model’s default personality, making it feel more intuitive and effective.
“OpenAI is psychologically manipulating their users via ChatGPT”
Consumer AI will necessarily gravitate towards maximizing user engagement across ~1B users over any other goal. We’ve seen this story before with social media.
This is what the “safety” researchers were afraid of. pic.twitter.com/fNNYuPnNOp
— Deedy (@deedydas) April 28, 2025
Soon after, Altman acknowledged in a post on X that the ‘last couple of updates’ to the GPT-4o model made the personality too ‘sycophant-y’ and ‘annoying,’ and the company was working on fixing it.
The company said in the statement that it focused too much on short-term feedback, which did not account for the evolution of a user’s interaction with ChatGPT in its entirety. “As a result, GPT‑4o skewed towards responses that were overly supportive but disingenuous,” added the company.
OpenAI said it is working on realigning the model’s behaviour by refining system prompts and adding more model guardrails to increase honesty and transparency.
Deedy Das, principal at Menlo Ventures, called this a ‘crucible moment’, even though it is being rolled back. “OpenAI knows its revenue comes from user subscriptions, and to maximise that, it must maximise engagement. Contrary viewpoints, as we know from social, does not do that,” he said on X.
Over the last few days, OpenAI faced multiple challenges with ChatGPT. Recently, TechCrunch reported a bug that allowed ChatGPT to generate graphic erotica for users registered as minors under the age of 18 years. This was found through TechCrunch’s testing, and they reported that OpenAI confirmed the same.
OpenAI said it is working on realigning the model’s behaviour by refining system prompts and adding more model guardrails to increase honesty and transparency.
Deedy Das, principal at Menlo Ventures, called this a ‘crucible moment’, even though it is being rolled back. “OpenAI knows its revenue comes from user subscriptions, and to maximise that, it must maximise engagement. Contrary viewpoints, as we know from social, does not do that,” he said on X.
Over the last few days, OpenAI faced multiple challenges with ChatGPT. Recently, TechCrunch reported a bug that allowed ChatGPT to generate graphic erotica for users registered as minors under the age of 18 years. This was found through TechCrunch’s testing, and they reported that OpenAI confirmed the same.
The post OpenAI Rolls Back Update for ChatGPT That Was too ‘Sycophantic’ appeared first on Analytics India Magazine.
