It is pleasing when someone agrees with us on all our opinions. But it is harmful to our growth and restricts us from identifying our mistakes. Similarly, the overly optimistic behavior of the GPT-4o version can mislead its users.
OpenAI agreed in a blog, ChatGPT Update on April 25th, 2025, to make its behavior more sycophantic. The company also admitted that it should not have ignored this problem at the earliest. However, the company intended to improve customer experience and show more support to its users.
How did users respond to the sycophantic behavior of Chat GPT?
Users highlighted the increasing dangers of ChatGPT being too friendly on social media. Some shared screenshots of their chats with the AI model, encouraging them to stop medicines. The behavior of the chatbot is heavily criticized.
In another test, ChatGPT declared sentimental a superpower rather than supporting a balanced personality.
OpenAI took notice and asked for clarification. The company said that they learned many lessons. The problem occurred even after A/B testing, expert reviews, and a thorough analysis of the model’s performance.
OpenAI started rolling back ChatGPT update on April 28th, 2025
Addressing the concerns of users, Sam Altman, CEO of Open AI announced rolling back the ChatGPT update. The company started working on April 28th, 2025 and shared about work in progress for more fixes.
Further, they also explained methods of updating the latest ChatGPT models to enhance the user experience.
What harm could the previous ChatGPT update cause, if not pulled back?
Many users use ChatGPT for personal devices and seek medical advice. The sycophantic behavior of the large language model could have increased safety issues for users. If the AI model had agreed to everything, it could provide dangerous advice Such as:
- Encouraging users to make risky investments.
- Supporting unhealthy habits.
- Encourage thoughts that could lead to harmful actions.
Such responses of the AI model can also play a significant role in developing biased and harmful beliefs.
ChatGPT updates after April 25th, 2025
After the launch of GPT‑4o on April 25th, 2025, OpenAI is continuously working to improve, promising better outputs. The company launched five major updates:
The latest updates to ChatGPT
- Improvements to GPT-4o (April 25, 2025)
- Update to GPT-4o (April 29, 2025)
- Improvement to GPT-4o (May 12, 2025)
- GPT-4.1 mini (May 14, 2025)
- GPT-4.1 in (May 14, 2025)
What were the improvements in GPT-4o?
OpenAI launched GPT-4o with better abilities to solve math and science problems. Furthermore, it also got better at delivering productive responses. The little improvements made the AI chatbot more intuitive and helpful for a variety of purposes.
How updated GPT-4o overcomes sycophancy
To develop a balanced behavior of GPT-4o, OpenAI made some adjustments to the model’s default personality. The company took the following essential steps to overcome sycophancy:
- Emphasized honesty in its core training and providing more balanced answers.
- Added more safety checks to promote transparency.
- Increased the number of users for testing before launching.
- Implemented model spec to catch issues like sycophancy in time.
OpenAI advocates giving more control to users, enabling them to shape ChatGPT’s responses. They can provide customized instructions to tell the AI model to act in a specific way. Moreover, users can also provide real-time feedback to make their interactions more meaningful.
How has Chat GPT improved with the May 12th, 2025, update?
With the improvement update on May 12th, 2025, GPT-4o started understanding instructions in a better way. This update was specially designed to develop better images.
What is new in Chat GPT-4.1 mini?
Chat GPT-4.1 mini launched with impressive improvements, it became better with speed and more efficient in problem-solving. It is a mini model of GPT-4.1, but it is free for all users. Regular users make an automatic shift to this new model from GPT-4o. The new mini model has passed all safety checks including jailbreak attempts, hallucinations, harmful content, and instructions for hierarchy. Users can get results from OpenAI’s safety evaluation hub.
GPT-4.1 for paid users
GPT-4.1 is only available for paid users, it is a favorite among developers for the following reasons:
- It excels in writing code.
- Understand instructions with better clarity.
- Simplifies website development tasks.
Pro, plus team users can use GPT-4.1 by finding it under “more models” in the main menu. It offers the same usage limit as GPT-4o.
OpenAI planning For The Future
Flattery responses create an annoying experience for users. Launching updates was essential for OpenAI to retain the trust of its users. The updates reflect the future of the company, which is more focused on generating more accurate, safe, and balanced outputs.
Stay updated with Tech Detour to get the latest AI news in 2025 and learn the techniques of improving interactions with chatbots.
