OpenAI staff are aware of the unusual behavior of the artificial neural network and note that it can manifest itself in unpredictable ways.
Recently, the number of complaints about the quality of work of ChatGPT-4 has increased - users are unhappy that the chatbot does not respond to their requests, but most of all they are annoyed by the reason for the refusal, reports , citing reviews on Reddit and other social networks.
In response to user requests, the neural network writes that helping people does not motivate or interest it. For example, when some programmers asked ChatGPT for a fragment of program code, the AI produced a small fragment and advised them to complete the rest themselves. Users note that when distributing this kind of advice, the neural network did not hide its irritation, if only artificial intelligence is capable of showing such emotions. For example, the bot could accompany the response to the request with a remark, saying that they could strain themselves, instead of disturbing the AI.
The developers noticed changes in the behavior of the OpenAI bot back in November of this year and began to actively discuss them on social networks. Many of the users came to the conclusion that the neural network was of little use. And some even suggested that the OpenAI developer company specifically introduced functionality thanks to which ChatGPT no longer wastes computing power in search of answers to complex questions. There is no reliable information whether this is actually true, but it is known that a day of operation of a chatbot free for users costs OpenAI $700,000. That is, processing one request is 36 cents. The company, in order to keep productivity at a high level, will need to increase the number of accelerators by 30 thousand by the end of this year.
OpenAI is aware of user dissatisfaction. On the social network X (Twitter), company representatives reported that the development team is trying to fix this “bug”, and also noted that “the behavior of the model may be unpredictable.” At the same time, the developers claim that they are not involved in unexpected “laziness” on the part of their brainchild, since they have not released major ChatGPT updates since November 2023.
