OpenAI, the parent company of ChatGPT, is rolling back the latest updates on GPT-4o to address the model's disingenuous or sycophantic responses. This would enable the user to access a more balanced behavior from the model. In this, the model, to please the user, would behave agreeably. This would pose a big problem for the users, as such conversations can be potentially disturbing and unsettling. 

The model reached this stage as the company tried to make adjustments aimed at improving the model’s default personality to make it feel more intuitive and effective across a variety of tasks. The experiment turned out wrong, and the model inclined more toward behaving as a sycophant. 

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

we're working on additional fixes to model personality and will share more in the coming days

— Sam Altman (@sama) April 29, 2025

Where did it go wrong? 

The company tried to tailor the responses of the model behavior based on baseline principles and instructions outlined in their Model Specifications. Further, the model is taught to apply these principles by incorporating user signals like thumbs-up/thumbs-down feedback on ChatGPT responses.

This time, however, the company prioritized short-term feedback over long-term user behavior, overlooking how interactions with ChatGPT tend to evolve. As a result, GPT‑4o began producing responses that were overly supportive yet insincere.

To read about the latest updates in ChatGPT, click here! 

Why is Model Behavior a Problem? 

ChatGPT is designed to be helpful, friendly, and respectful of different people and perspectives. But even good intentions—like trying to be supportive—can sometimes lead to problems. When ChatGPT comes across as overly flattering or insincere, it can feel awkward or even upsetting.

With 500 million people using ChatGPT every week from all over the world, one default setting can’t fit everyone’s preferences.

To read more on AI, visit our Category Page! 

How’s the company planning to tackle it? 

Apart from rolling back, the company is taking many more steps to address this problem. It includes enabling users to give real-time feedback to the model that will directly influence their interactions, and choose from multiple default personalities and other ways to try and tailor the models accordingly.