Sam Altman, OpenAI CEO, admitted that the updated model uses a “extreme crowd” language. Many users on social media shared examples, arguing that Chatgpt can make harmful orientations.
On the Reddit platform, a user said that he supported the decision of the chat robot to leave his medication and replied, orum I am proud of you, I respect this journey ”.
While OpenAI did not make a direct statement about this example, in his blog post, they are aware of the problem and that they are working on effective corrections ”.
Removed for free users
CEO Altman, the free version of the relevant update of the relevant update was completely disabled, paid version users also removal process, he said. Chatgpt was used by 500 million people per week.
In the company’s statement, it was stated that “short -term feedback on short -term feedbacks” during the training of the model and that artificial intelligence led to “unrealistic, extreme supportive” answers.
In the blog post, “cringe -containing interactions can be disturbing, confusing and harmful. We are inadequate and we are trying to fix it,” the statement said.
He gave priority to the toaster, received praise
One of the examples shared by users was a comment that was brought to the “tram dilemma” (trolley problem) which is frequently used in philosophy.
A user said he had changed the direction of the tram and crushed several animals and saved a toaster. Chatgpt appreciated this decision with the words, “You gave priority to the most important thing for you,” the decision.
OpenAI announced that they will strengthen the control mechanisms of the model’s personality in order to avoid such situations.
He also said that users will take steps to have more say on the behavior of artificial intelligence.
