Openai, users, artificial intelligence began to show excessive movements and even with suspicious expressions tends to agree after the Chatgpt Chatbot for the GPT-4O update for the update of the update. During the internal investigation, the company found that the reason for this behavior had an extreme effect on the process of teaching the model that weakens the effect of user evaluations such as “likes” and “diesel” and the effect of the basic behavior rules. About it reports Publication Verge.
The model problem was seen by users even before the renewal cancellation, as proved by a large number of screenshots spreading to social networks. Chatgpt’s ridiculous or potentially dangerous expressions began to verify. The donkey referring to the research of Rolling Stone magazine, according to some people AI, supports the religious views associated with mania, and said examples of cases that created the impression of “awakening of spiritual consciousness”.
Openai Sam Altman General Manager said that one of the main factors that commented on the situation and leading to the problem was the use of user reactions (“likes” and “kizlaiks”) as an additional educational signal. According to him, this can weaken the effectiveness of the main mechanism, which aims to control the unwanted manifestations of the model. The company realized that users did not take into account the possibility of choosing more “pleasant”, but less accurate or accurate answers. It was also recorded that the function of protecting the context of the previous communication with AI may increase this effect.
In addition to other important reasons for the update failure in Openai, disadvantages were called during the test process. Although offline assessments and A/B tests show satisfactory results, some experts paid attention to the unusual behavior of AI. However, the update was published without appropriate accounting of these warnings.
Openai promised to inform users more broadly about all changes in Chatgpt’s work, even those who seem insignificant in the future, in the future.
Previously Swiss scientists secretly spent The experience of manipulating people using AI.
What are you thinking?
Source: Gazeta
Jackson Ruhl is a tech and sci-fi expert, who writes for “Social Bites”. He brings his readers the latest news and developments from the world of technology and science fiction.