After reading the college entrance examination essays answered by the big models, I can relax my advocacy and fear of AI.

Text | Neighbor Chapter

The use of AI tools for literary creation has always been something that technology companies are willing to try. Especially at special time nodes such as the "College Entrance Examination" that attracts national attention, using AI to write college entrance examination essays is even more sought after by technology companies. .

This year, the overt and covert fights will be worse than before.

The core reason is, of course, that since last year Open AI’s generative AI application ChatGPT has almost qualitatively changed its capabilities, triggering a global frenzy for large generative AI models, making everyone recognize that AI technology has reached the “iPhone” moment. ‍

Domestic technology and Internet companies have also followed suit and launched their own generative AI large model products this year. According to incomplete statistics: from February this year to the present, domestic AI large model conferences have been held for nearly 20 years. In the market, companies such as Baidu Wenxinyiyan, Alibaba Tongyi Qianwen, iFlytek Spark Model, etc. almost all claim to the outside world that their product technology is leading.

Here, the "College Entrance Examination Essay", a comprehensive examination question that comprehensively tests candidates' material understanding ability, writing skills, critical thinking ability, and writing innovation ability, is undoubtedly a touchstone for testing the technological creation capabilities of various large-scale models.

It is precisely based on this that manufacturers who are committed to large-scale generative AI models are also gearing up to show off the literary creation capabilities of their own large-scale generative AI models.

For example, Zhou Hongyi shared an essay on this year’s National A-level College Entrance Examination written by 360 Intelligent Brain on Weibo.

But judging from the final results, the performance is good in basic skills but not enough in the critical level, and it is difficult to get high scores.

Wang Daji, a special teacher of Chinese language, commented on the AI ​​essay generated by 360 Intelligent Brain on Weibo: "The language of the essay is clear, the levels are clear, the basic writing skills are very good, and the performance is good at the basic level. The essay also talked about High technology, but under the conditions of scientific and technological development, why should we be the master of time, but why we have become the servant of time? We failed to grasp the key and did not think thoroughly enough. Therefore, we did not form an in-depth analysis. The sense of the times required by the title is relatively indifferent. In terms of development level, the performance is average."

In fact, this lack of speculative thinking at the level of literary creation is not limited to 360 Intelligence. In fact, it is almost a "common problem" in current large generative AI models.

In Sohu Technology, we used the National College Entrance Examination Essay Paper (A) as the topic, and analyzed five mainstream large language models: Baidu Wenxinyiyan, Alibaba Tongyi Qianwen, iFlytek Spark Cognitive Large Model, 360 Intelligent Brain, and ChatGPT The product was tested on the same topic, and five famous Chinese language teachers were invited to comment on the level of composition generated by each model. It was not difficult to find that: the five major models basically understood the composition concept of the National College Entrance Examination Essay Paper (A), and there were no Corresponding articles can also be generated in a short time for off-topic and off-topic phenomena.

But at the same time, the big models also have unclear understanding of the specific requirements for composition (such as the number of words in the articles they create does not meet the requirement of no less than 800 words), rigid writing, obvious sense of fragmentation, empty language, and "science and engineering" "flavor" is obvious; and more importantly, the articles they create generally suffer from the most fatal problems in literary creation, such as insufficient arguments and evidence and unoriginal viewpoints. This also leads to the college entrance examination essays created by these five large models. ChatGPT, which scored the highest, only scored 48 points.

Frankly speaking, ChatGPT’s 48 points and Baidu Wenxinyiyan’s 46.5 points are actually not low in the college entrance examination essay scores. They may even exceed the essay scores of many candidates, but they are not comparable to excellent essays. Not above, it can only be said to be at a relatively ordinary level.

The problems exposed by these large models in literary creation also allow us to see the shortcomings of current generative AI technology. It can even be said that those clamoring that generative AI technologies such as ChatGPT can now perform literary creation, The words that replace creative people are indeed a bit exaggerated and even alarmist.

Of course, we do not deny the potential of generative AI technologies such as ChatGPT to grow rapidly with continuous nourishment. However, to get rid of the current lack of creativity of large models in literary creation, it may not be possible for large models to What can be done overnight may take a long time.

Here, it can almost be said: After reading college entrance examination essays that use generative AI to answer questions, I think that whether it is advocating or fearing AI to replace literary creation, we can actually put it on hold for the time being.

Write at the end:

ChatGPT and other large model technologies have the capabilities of email & news & paper writing, video scripts, copywriting, translation, code writing, etc., allowing generative AI technology to greatly surpass past entertainment-oriented AI applications. Attributes have entered the stage of real productivity and even creativity.

As a result, many people in the industry exclaimed: Creative job opportunities may be replaced by GPTs first, and they began to fall into pessimistic expectations.

In fact, my view on the strength of generative AI technology has always been: there is no need to over-promote, and there is no need to be overly fearful. Treat the development and changes of technology with a normal heart, maybe will be better.

Standing in the face of the technological wave, what we should do more is to learn to make good use of these technologies as early as possible to help improve our work efficiency and quality of life.

Author: Lin Zhang [WeChat: ZLxgic, public account: TMT317], focusing on interpreting industry development trends and company products and services. Digitalization, AI, cloud computing, company financial reports, smart hardware, and consumer innovation are the focus.

The above is the detailed content of After reading the college entrance examination essays answered by the big models, I can relax my advocacy and fear of AI.. For more information, please follow other related articles on the PHP Chinese website!