in the wake ofartificial intelligence (AI)Technology continues to evolve and we rely more and more onChatGPTSuch intelligent assistants to help us solve our problems. However, a recent study has pointed out thatChatGPTThe fact that people seem to be getting "dumber" has led to a debate about why. A recent study from the University of California, Santa Cruz, offers a new explanation.
This study found that the large model performed significantly worse on tasks after the training data cutoff. This phenomenon, known as "task contamination," means that during the training phase of the large model, it has already seen a lot of task examples, which gives the impression that the large model has already been trained.AIthe illusion of having zero- or few-sample capabilities. However, in practice, when one proposes new tasks, the performance of the AI suffers because these tasks did not occur during the training period.
In addition, it has also been pointed out from another perspective that people's task requirements change over time. However, the parameters of large models are frozen, which means that they cannot adapt to such changes in the input distribution. If the models cannot continuously adapt to such changes, their performance will gradually degrade.
This phenomenon also reminds us that while AIs excel in some areas, they still have limitations. People often assume that if they ask an AI a question, it will be able to give an accurate answer. However, the reality is that the AI has already seen most of the common tasks during its training period. Over time, people start asking more new questions and tasks that were not covered during the training period. This resulted in the AI not performing as well as expected.
This study provides us with a new explanation for ChatGPT's "dumbing down": the world is changing while AI is stagnating. While AI excels in some areas, we must be aware of its limitations and constantly strive to improve and adapt to changes in technology. Only then will we be able to fully utilize the potential of AI to advance technology.
This article comes from users or anonymous contributions, does not represent the position of Mass Intelligence; all content (including images, videos, etc.) in this article are copyrighted by the original author. Please refer to this site for the relevant issues involvedstatement denying or limiting responsibilityPlease contact the operator of this website for any infringement of rights (Contact Us) We will handle this as stated. Link to this article: https://dzzn.com/en/2024/2528.html
