Wen, Sina Finance and Economics Zhou Wenmeng

Just after the Spring Festival, ChatGPT quickly detonated the capital circle and AI circle, and many practitioners blew it up.

Yuan Jinhui, founder of the OneFlow Deep Learning Framework, told Sina Finance that the technological progress of ChatGPT can be compared to the first "moon landing", which shocked the industry. But going out and asking founder and CEO Li Zhifei also affirmed ChatGPT's change, but said, "the answer it knows is that it lacks a logical starting point and reasoning rules, and is just a castle in the air." "

To this end, Sina Finance has opened a "chat.openai.com" official website account to measure ChatGPT to see if it is so magical after all. Can it lead the AI tuyere, or a flash in the pan?

Is the measured ChatGPT that magical after all?

"the experience of ChatGPT is already the best at present, and it can even be used as Alphabet Inc-CL C, and it may become an information portal in the future." This is the feeling of Sina Finance issued by a technology observer who has been walking in Silicon Valley for a long time after experiencing ChatGPT. But in fact, there are not a few people with similar feelings from abroad to at home.

According to a survey of 1000 students over the age of 18 conducted by online course provider Study.com, more than 9 out of every 10 students know about ChatGPT, far more than primary school educators, and more than 89% of them are using ChatGPT to finish their homework.

On the "chat.openai.com" official website account opened by Sina Finance, when the author asked the question "how to run a company", ChatGPT gave not only a brief and concise answer to the necessary processes and steps to run a company, but also a large number of specific and detailed guidelines and suggestions in further in-depth questions and answers.

"the emergence of ChatGPT has realized the so-called 'emergence of wisdom', which is a victory for the wisdom of human groups." This is the feeling made by Yuan Jinhui, founder of the OneFlow deep learning framework, after deeply testing ChatGPT. In communication with Sina Finance and Economics, Yuan Jinhui unsparingly compared the technological progress of ChatGPT to the first "moon landing" or "climbing Mount Everest", saying that such progress shocked people in the industry.

"it's a bit like the imagenet moment at that time. I couldn't do it for decades, and suddenly I found that the computer could solve the problem of image recognition. Or Zhang Yitang's twin primes prove the boundary to 70 million. Although the number is very large, this breakthrough is the biggest, and it is completely feasible to improve it to hundreds or even single digits. " Yuan Jinhui said.

Yuan Jinhui's feelings represent the views of the vast majority of practitioners in the AI industry on ChatGPT. In communication with Sina Science and Technology, Wang Yonggang, executive dean of the Institute of artificial Intelligence Engineering of Innovation works, and Li Zhifei, founder and CEO, all affirmed the profound impact and changes that ChatGPT has brought to the artificial intelligence industry.

But in addition to affirmation, some calm vigilance and reflection also exist. "AI creation is extremely dangerous in the field of knowledge education, and today's AI generation theory cannot guarantee that the logic of the generated content is correct and reasonable. Professionals such as AI/CS should keep calm. Today's AI still has not got rid of the problem of big black box, and there should be a moral bottom line to follow the trend. " Wang Yonggang, an innovation workshop, said.

In addition, Li Zhifei, founder and CEO, also told Sina Finance that "although ChatGPT's chat skills have reached a very high level, with logic and fluency, the answer he knows is a lack of logical starting point and reasoning rules. He is only a castle in the air. 70% of the answers can be answered, but he doesn't know why he got the right answer." Li Zhifei said.

According to the actual measurement of Sina Science and Technology, due to the time limit for ChatGPT system to acquire knowledge and data, the data currently available to ChatGPT is only up to 2021, and the newly generated knowledge and data have not been included in the future, which leads to outdated news and knowledge when users are not clear about the time deadline of knowledge on ChatGPT. In addition, because it is not open to users in China, ChatGPT is often wrong and inaccurate in answering questions that may be common sense to most citizens in China.

In addition, Huang Minlie, an associate professor in the Department of computer Science and Technology at Tsinghua University, also pointed out that "due to the back-end model and algorithm settings, ChatGPT does not have dynamic memory function. And at present, it is only the study and application of past knowledge, if there is not enough information available on the topic, it is easy to reply with incorrect data. "

How difficult is it to commercialize?

Under all kinds of problems, despite the enthusiastic pursuit of some investors and AI practitioners, ChatGPT quickly became popular with AIGC concept stocks on the first working day after the beginning of the year, causing the flush artificial intelligence plate to rise by 2.64%. The share prices of listed companies such as Tors, Haitian AAC, Jingyeda, Yuncong and other listed companies have risen by the limit. However, there are still a large number of companies and individuals who say "NO" decisively to ChatGPT.

Shortly after the launch of ChatGPT, Stack Overflow, a well-known programmer question-and-answer site, announced a ban on using questions generated by ChatGPT to answer community questions and imposed disciplinary measures that violators can be banned for up to 30 days. In addition, professors from New York University and other universities and many scholars have also warned that "AI will make students' brains shrink, so avoid ChatGPT like a plague!"

Under the dispute, the commercialization of ChatGPT has become the focus of attention. In Li Zhifei's view, "the emergence of ChatGPT has brought a new business model possibility for the AI industry." Prior to this, the charging mode of AI enterprises is more To B charging system, which is poor in replicability, lack of network scale effect, and the marginal cost is uncontrollable. But the emergence of ChatGPT, let people see the possibility of AI technology to empower C consumers, by providing small and medium-sized enterprises or individuals with AI tools or services, it is also possible to obtain recognition and benefits. "

"ChatGPT provides a new perspective. On the basis of a powerful pedestal model, supplemented by high-quality and diverse data, the reinforcement learning method can achieve very good results. In this, the closed loop between data and model is formed, which is something worth thinking about in every AI enterprise. In the future, in addition to general task assistants, open domain chat will also become the main form of human-computer interaction. One meets the information needs, the other meets the social and emotional needs. " Huang Minlie told Sina Finance.

However, as a "phenomenal" technological innovation, the large-scale commercial application of ChatGPT can not be separated from the support of computing power. It is reported that at present, a training on the base of a model of ChatGPT costs US $5 million, and the comprehensive cost of a question and answer is US $0.80. According to the previous budget of Li Di, CEO of Xiaobing company, "if Xiaobing runs the system in the way of ChatGPT, the amount of interaction carried by Xiaobing will cost nearly 300 million yuan per day, and even if ChatGPT can optimize the cost to 10% of what it is now, it will not earn it back."

"because the biggest feature of artificial intelligence is high concurrency. If the cost is like this, you might as well hire someone. " Li Di said. In his view, ChatGPT is expensive because it pursues the ultimate interactive effect and uses the most complex ways to achieve it. "although people are constantly looking for a variety of equivalent replacement methods, the cost will continue to gradually reduce, but so far it is more of a breakthrough in research, a breakthrough in the laboratory."

Bloomberg and Citigroup Inc recently estimated that ChatGPT could generate $3 billion to $11 billion in sales for NVIDIA Corp in the next 12 months. ChatGPT became popular, but it was the lower-level computing companies that were the first to profit from it.

A flash in the pan?

Although the popularity of ChatGPT is also accompanied by criticism and controversy, optimistic voices still believe that the emergence of ChatGPT will further promote the further development of conversational AI and generative AI.

In its 2023 Top Ten Technology Trends released on January 11th, Dharma predicted that generative AI will usher in an explosion of applications, greatly promoting the production and creation of digital content. In addition, venture capital in the generative AI sector reached $2.1 billion and more than 100 financing events in 2022 (as of December 6, 2022), according to data analysis service PitchBook. Venture capital in this sector grew by 425% between 2020 and 2022.

"I don't think ChatGPT will be a flash in the pan. It will be a very important milestone in the automation or semi-automation of all kinds of online counseling, chat and other professions, or even a necessary node." Li Zhifei told Sina Finance. In his view, "although the current ChatGPT still has some defects, but then there must be a lot of people and companies will enter, his vertical commercialization to do a good job."

From the law of history, the realization of each technology is moving forward in the midst of controversy and criticism. Although those critical voices often mean that the present is correct, most of the criticized people tend to iterate forward after being criticized and eventually win back their lost dignity.

Edit / Viola