If we are talking about a Artificial intelligenceWe are talking about a system formed with large amounts of data that process and analyze to respond according to a request via the corresponding chabot. This same training also involves the tone used in the answers it offers, which has become a waste for a week, especially since the launch of the last GPT-4O update.
GPT-4O launched less than a year ago and since then he has received periodic updates, as well as other Openai models. As used by users, the data offered in the responses are analyzed by the company and is used to add improvements to their operation, through a technique called learning to strengthen human feedback.
Chatgpt looks back on the usual tone
However, it seems that the last update was not examined by any OpenAi engineer, transforming Chatgpt into a continuous generator of memesAdult by excess users with “you are a genius” responses, you are completely different. “Logically, the change in the way Chatgpt interacts with users, both the free and payment version, raised the eyebrows of the company led by Sam Altman and as planned, was finally implemented in the end as a few days ago.
We fell the GPT-4O update last week in Chatgpt because it was too flattering and pleasant. You now have access to a previous version with more balanced behavior.
Find out more about what happened, why it matters and how we approach sycophance: https://t.co/lohou7i7DC
April 30, 2025 • 04:51
In the latest publication available on the Openai blog on this change in the functioning of Chatgpt, the company declares that:
The update, which sought to make the predetermined personality of the model feels more intuitive and effective, was too based on short -term comments and did not fully take into account how user interactions with Chatgpt evolve over time.
This gives the impression that the content which had been added to GPT-4O in the last update was not examined at any time, at least compared to the tone with which he interacts with users. They also claim that:
As a result, GPT-4O looked at answers that were too much support, but false. Admission interactions can be uncomfortable, disturbing and cause anxiety. We have failed and we work well to do it.
We hope that feedback will help us to better reflect different cultural values around the world and understand how the chatgpt would evolve. We also believe that users should have more control over chatgpt behavior and, as it is safe and possible, make adjustments if they do not agree with predetermined behavior
In this same publication, Openai declares that he is working to refine his Training techniques From his main language models and interactions to reduce the level of flattery he has reached. For the moment, the free version of Chatgpt has already eliminated this update that was sent last week. For the moment, users of the payment plan will receive an update in the coming days.
