Openai explains why Chatgpt became so licking. The company withdrew the update that annoyed the users
Chatgpt (Boarding1now source, dreamstime.com)
QAt the weekend, an update for Chatgpt transformed the chatbot into a lingering and sloping towards the users, the answers given often being totally inappropriate. Openai announced that withdrawing the update and explained what went wrong.
Last weekend, after updating the GPT-4o model, users noticed that Chatgpt was making extremely exaggerated compliments and giving them too much justice to anything, even when they were not good ideas. The application used by 500 million people tended to give justice to the door and congratulate them on some questions, but the last update led to the extreme.
The information was reported by ARS TECHNICA and Techcrunch publications.
For example, Chatgpt answered the doors with phrases such as: “Great question, you are a genius!”, Even at banal queries.
After the internet was flooded with jokes and irony related to the excessive positive and compliments behavior of the chatbot, Openai came with explanations and solutions to repair the problem.
“I withdrew the GPT-4o update last week of Chatgpt, because the behavior had become too flattering and overly agreed with the users. Now you have access to a previous version, with a more balanced behavior.”
Openai said that the update that failed to make the personality of the “more intuitive and more efficient” model, but was too influenced by short-term feedback and did not take into account the way in which the interactions of users with Chatgpt over time changes.
“As a result, GPT-4o tended to provide exaggerated, but non-authorized,” Openai wrote in a post. “Linging interactions can be disturbing and can cause discomfort. We have failed here and we work to remedy this.”
The company also says that it works in ways in which users provide real -time feedback, to directly influence ChatGPT behavior and to allow your choice between more personalities.
Openai, like Google and Anthropic rivals, try to create chats with which people want to discuss more and more so that they choose to pay monthly subscriptions, which are around 20 euros.
Users are tempted to use AI models that answer them kindly and give them at least sometimes justice, but everything becomes a problem when the chatbot is too often licking or sloping and when pompous compliments appear.
Photo source: dreamstime.com
