Openai has implemented several modifications to its GPT-4O model, following an incident where Chatgpt has become too pleasant and validating. Users noted that the GPT-4O model had caused a Chatppt cat to excessively applaud problematic ideas and validated doubts, arousing concerns between users. CEO Sam Altman recognized the problem on X, promising immediate corrective.
He wrote,,
“The last two GPT-4O updates have made the personality too sycophan and boring (even if there are very good games), and we are working on fixes as soon as possible, some today and some this week. Corrects the personality of the model and will share more in the coming days. »»
Register For TEKEDIA Mini-MBA Edition 17 (June 9 – September 6, 2025)) Today for early reductions. An annual for access to Blurara.com.
Tekedia Ai in Masterclass Business open registration.
Join Tekedia Capital Syndicate and co-INivest in large world startups.
Register become a better CEO or director with CEO program and director of Tekedia.
Recognizing the problem, Openai launched a decline, restaurants a previous and more balanced version of GPT-4O. Last week, the organization shared initial information on the accident, explaining why it occurred and its plans to remedy it. They recognized that the question had not been detected before deployment and undertook to explain the surveillance, learned lessons and improvements to their processes.
In a blog article, Openai wrote,
“On April 25, we deployed an update of GPT-4O in the Chatppt which made the model significantly more sycophantic. Excessive emotional reliance or risky behavior.
“We Began Rolling that update back on April 28th, and users now access to an earlier version of GPT-4O with more balanced Responses. Earlier this week, we shared initial details about this Issue. Why It was a miss, and what we intend to do about it. We didn't Launch, and we want to explain why, what we've learned, and what well improve.
The changes arise as the basis of chatgpt users is increasing, with 60% of American adults using it for advice, according to a recent Express Legal financing survey. This dependence increases problems of problems such as sycophance and hallucinations. OPENAI plans to allow user comments in real time, refine the behavior of the model to reduce sycophance, offer several model personalities, strengthen safety railings and develop evaluations to catch broader problems.
Openai noted a change in the way users are looking for deeply personal chatgpt advice, a less important trend a year ago. “As the AI and the company have co-evolved, it has become clear that we must deal with this case of use with great care,” said society, committing to prioritize this in its security efforts.
Improvements of future versions of the Openai model
In response to the incident, Openai implements several key changes:
- Explicit behavior review: Future updates will officially assess behavioral problems such as sycophance, hallucinations and inconsistency as potential launch blockers, even if they are difficult to quantify.
- Alpha testing phase: A new opt-in test phase will allow selected users to provide detailed comments before public launches.
- Increased accent on qualitative tests: Occasional verifications and practical evaluations will be high in importance, in particular when quantitative signals are ambiguous.
- Best offline assessments and A / B: The evaluation frames will be extended to capture nuanced behavior models.
- Improvement of membership in the model specification: OPENAI will strengthen its capacity to measure how models respect the defined behavioral ideals.
- Transparent communication: The company will now proactively announce all updates of the model – major or subtle – with known limitations, to promote the understanding and confidence of users.
The GPT-4O update incident highlighted the critical importance of the behavior of the model as a central component of IA security and reliability. Openai has recognized that even with robust A / B tests, offline assessments and internal journals, important behavior problems can still be missed. Consequently, the behavior of the model will now be treated as seriously as traditional security risks in future deployments.
Thanks to this episode, Openai has readjusted to developing models which are not only intelligent and useful, but also aligned with user well-being, transparency and security.
