This article is reproduced from the WeChat public account "Living in the Information Age". The author lives in the information age. To reprint this article, please contact the Living in the Information Age public account.
In the past few days, the hottest thing in the field of artificial intelligence is ChatGPT, a conversational and chatting robot developed by OpenAI.
ChatGPT is a large-scale pre-trained language model that can generate human-like text responses in conversations. Its algorithm is based on the most popular Transformer architecture, which is a deep neural network that uses a self-attention mechanism to process input data. It is widely used in various natural language processing tasks. ChatGPT is trained based on a large amount of text dialogue data sets and uses a self-attention mechanism to learn the patterns and structures of human-like dialogue. This makes his answer very close to that of a real person. Some people even think that ChatGPT can completely replace search engines.
In Zhihu author DeFi's popular science article "Popular Science: What is ChatGPT?" 》, ChatGPT is introduced. As shown in the picture:
But at the end of the article, the author gave everyone an easter egg, indicating that the article itself was written by ChatGPT itself . For example, in the "Introduction to ChatGPT" section in the picture above, the author is asking ChatGPT: What is ChatGPT? The answers I got later, similar to the "Algorithm" section, were the answers the author got after asking "What is the algorithm behind ChatGPT?"
From the article, we can see that in ChatGPT’s answer in this scenario, it is almost difficult to tell whether it is a robot answering. It's no wonder that many people are amazed by its performance.
However, another group of people are not satisfied with the performance of ChatGPT. For example, the well-known programmer community Stackoverflow issued a temporary rule on December 4th: it is prohibited to use content generated by ChatGPT to answer questions on Stackoverflow. The reason is that the accuracy rate of the generated content is very low, and these specious content are harmful to the entire website and users seeking correct answers. The main problem here is that because the threshold for using ChatGPT is very low, many people have been using ChatGPT to answer questions raised by others in recent days. However, due to their own lack of professional knowledge, they do not have the ability to verify whether the answers generated by ChatGPT are correct. A lot of worthless and even misleading answers were produced.
Others tried some elementary school students’ questions, but ChatGPT’s answers were unsatisfactory. For example:
Behind the unsatisfactory answers to these simple questions lies our in-depth thinking about the development of the field of artificial intelligence.
In the field of deep learning, researchers often say: If you torture the data to a certain extent, it will confess everything.
This is a kind of self-deprecation. The current field of artificial intelligence mainly relies on a large amount of training data to train models. The success of a model is closely related to the amount of data it trains. So this will inevitably lead to a question: What if one day, after a super model is trained using the largest data set in the world, it still cannot get good enough results? After all, ordinary people do not need to learn all the knowledge in the world to have their own learning and judgment abilities.
And further speaking, will the data trained with all real data be better than the model trained with part of the real data? Consider that some real data has completely opposite answers to the same question. Just like there are always different people arguing about the same issue. These training sets are bound to have an impact on the training results of the neural network.
Perhaps, the real breakthrough in artificial intelligence will have to wait for a breakthrough in basic science. It's like Maxwell's equations brought people into the era of wireless signal transmission. In the space where we get along day and night, there may be deeper secrets hidden, waiting for people to discover.
The above is the detailed content of Behind the popularity of ChatGPT, where is the future direction of artificial intelligence development?. For more information, please follow other related articles on the PHP Chinese website!