laitimes

In the face of ChatGPT's global popularity, how should China's AI debut?

Image source: @VisualChina

AI Industry Research Center

In the three months of the first quarter of 2023, the United States gave two big gifts to the world. One is the inexplicable collapse of Silicon Valley Bank and Signature Bank in a few days, and the other is that the "phenomenal application" ChatGPT has become popular around the world in a short period of time and has become the "top stream" in the AI industry.

If the collapse of the two banks exposed systemic risks in the US financial sector, then the popularity of ChatGPT set off a wave of AI that spread wildly in the global capital market.

As a result, in March this year, the most spectacular "AI Collective Outbreak Month" in history was ushered in. In just half a month, Microsoft's Microsoft 365 Copilot, Google PaLM API, Baidu "Wenxin Yiyan", NVIDIA's A100 chip and other AI applications and products have appeared and been made public.

It can be said that ChatGPT has almost single-handedly opened up the chaotic global AI industry.

Overnight popularity, where is ChatGPT strong?

As a chatbot launched by the American artificial intelligence company Open AI, it has certain thinking, summarizing, and analyzing capabilities, and can be called a personal "chat assistant". Once it came out, it was sought after by hundreds of millions of people.

A female white-collar worker in Taiyuan sighed after the experience: "ChatGPT is excellent in writing emails and proofreading documents. A college student abroad asked ChatGPT to help him write a paper, and the paper was directly rated as "excellent" by the professor.

It is said that in order to confirm the ability of ChatGPT, Google once handed it a programmer recruitment interview paper, and finally its roll score reached the admission score line of Google's third-level software engineer, which is a position with a starting salary of $180,000.

Wang Bolin, general manager of Beijing Hua Carbon Green Energy Technology Co., Ltd., told reporters after using it that ChatGPT can achieve cross-industry and cross-field accurate information promotion through its ability to accurately match and integrate massive data resources.

Reducing mechanical repetitive work, saving time and reducing costs is where ChatGPT is powerful.

Human-machine war, whose job did ChatGPT steal?

On February 11, CCTV released a Weibo listing #10 occupations that may be replaced by ChatGPT#, which actually includes programmers, media people, legal workers, market analysts, teachers and other unexpected professions. Does ChatGPT really have such power?

In fact, from a technical point of view, ChatGPT is a universal content generation robot with both keyword intelligent retrieval and model intelligent matching functions, with four advantages: fast retrieval speed, simple and easy operation, large amount of information, and rapid automatic generation of content, which humans can not match.

Because physiologically speaking, it is impossible for a person to complete a writing in a few seconds; From a psychological point of view, if people write for a long time, they will definitely produce fatigue and boredom; And from a physical point of view, a person cannot be like an encyclopedia, who understands everything, knows everything, knows everything.

The four advantages of ChatGPT are beyond the reach of human physiological, psychological, and physical limits. Because, ChatGPT is a "machine", not a person.

But there are two sides to everything. The so-called success is also Xiaohe, and defeat is also Xiaohe. Because ChatGPT is not human, it also has no human aura. The content it generates is limited by computational model libraries and database frameworks, which are neither creative and imaginative, nor humane, and are completely built by stacking, logical but not rigorous, and emotionless. As ChatGPT himself puts it: "I generate text based on pre-2021 databases and probabilistic language models based on pre-2021 pre-2021 databases. ”

Therefore, I personally think that the list given by CCTV is debatable. Because it ignores the creativity, emotion, and rigor necessary in the profession, as well as the ever-changing world itself, especially for media professionals, lawyers, teachers, designers, etc.

Taking teachers as an example, if they only teach existing knowledge, ChatGPT is fully competent, but if they want to interpret and extend a new knowledge point, and teach students to apply and innovate flexibly, it cannot be done. But isn't that what teachers really should be doing?

Perhaps the teacher's cognition of existing knowledge is not as perfect as ChatGPT, but it is precisely because of imperfection that it stimulates people's thinking and exploration spirit. Lin Yutang wrote in "Life Is Just So": "Imperfection is the most perfect life."

Remember AlphaGo a few years ago? In the field of Go, it kills humans to the point of no resistance. But humans invented board games not to defeat their opponents, but to challenge themselves.

Therefore, those who can be replaced by ChatGPT must be those professions that do not need to think deeply, but only mechanical operations or those who do not want to think, not the profession itself.

Is it creation, or is it a story? What exactly should AI bring to humanity?

Since the advent of ChatGPT, there has been an anxiety in society that many people feel that ChatGPT can already be the same as "people".

On February 13, Fu Sheng, CEO of Cheetah Mobile, posted excitedly: "I was shocked by ChatGPT, because it actually guessed the reason why my dog barked 30,000..."

At first glance, this reaction seems to be fine. But if you look closely, you will still find that the success of ChatGPT still has traces to follow. Because it has already framed the range of answers in the memory bank by asking questions before answering, it just selects the most likely answers, and the whole process is a bit like a robot fortune telling. If I were to answer, I would think of two answers without asking, one for the dog for 30,000, and the other for 30,000 when I first saw it.

In addition, some netizens did such a test, which asked ChatGPT to refer to Du Fu's "Spring Hope" and create a poem with the same sentence structure and based on the keywords provided.

The result is self-evident, these two poems are not at all in the same grade, Du Fu's poems are full of emotions and far-reaching artistic conception, even if they span thousands of years, they can directly hit people's hearts, while ChatGPT's poems are just a collection of words, empty and tangible, but there is no inner emotion and artistic conception expression, completely unknown. But a poem that does not embody human emotions, is that still poetry?

It can be seen that ChatGPT itself has not reached the level of human intelligence, and can only do things with traces and mechanical repetitions, only computing power, no intelligence, no feelings, and can not break through innovation. The huge changes it has brought to the development of AI are mainly in technical routes, model design and training methods.

As Baidu founder Robin Li said: "Technology accumulation is here, data accumulation is also here, computing power accumulation is also here, just at this point in time these three have the ability to explode, AI technology further mature, can be applied to many fields, greatly improve efficiency, and change people's lifestyle." According to reports, this is a prediction he made about the outbreak of ChatGPT-like products when he attended an industry summit in 2020. It can be seen that the advent of ChatGPT is just a natural success, not a foreign birth.

So what exactly should artificial intelligence bring us?

On March 31, Tian Qi, Chief Scientist of HUAWEI CLOUD AI Field, proposed that AI for industries will become a new flashpoint for AI. He pointed out that the real use of artificial intelligence is to promote the improvement of human production efficiency, and the application scenario should be expanded from the hot consumer side to the production side, just like electricity, if it is only used for lighting, it can not promote the progress of human productivity, only when applied to the production of all walks of life, will promote the improvement of production efficiency in the whole society. ”

As they said, the real value brought by ChatGPT is by no means independent content generation, but the successful application of NLP large model training and deployment in the field of AI, which will drive global attention and investment in AI, and ultimately promote industrial intelligence and social intelligence in various fields, just like the explosion of AlphaGO after defeating the strongest chess player in humans.

At the "2022 World Artificial Intelligence Conference (WAIC)", Chen Haibo, founder of DeepBlue Technology, said in a keynote speech that artificial intelligence is the underlying enabling technology, just like electricity and the grid, it can be used everywhere, not only can help individuals, change lives, but also empower society, achieve overall industrial upgrading and transform the world.

Yes, ChatGPT should be a technological tool to create a better world for humanity, not just a story that people tell everywhere.

Is it a firework, or a daylight firework? How should Chinese AI become popular?

Taking advantage of ChatGPT, Kai-Fu Lee, CEO of Innovation Factory, Wang Huiwen, co-founder of Meituan, Jia Yangqing, vice president of Alibaba, Wang Xiaochuan, founder of Sogou, and other business celebrities appeared one after another, and they all planned to incubate AI big models.

In response to this phenomenon, the CEO of a leading domestic AI company poured cold water on "Titanium Media", and the wave of large model fever caused by ChatGPT will eventually be "chicken feathers", because the technology is really not a "miracle" can be solved, five or six months later, there will be a lot of "bubbles" in the AI industry.

It is understood that ChatGPT technically takes the route of "big data + large computing power + strong algorithm = large model", and has explored a new paradigm of "basic large model (similar to brain) + instruction fine-tuning (referring to interactive training)", and realizes intelligence close to human language through the integration of the two, of which "big data, large computing power, and strong algorithm" is the pillar. In order to bring GPT-3's performance closer to humans, Open AI used 45TB of data and nearly 1 trillion words to train it. According to Guosheng Securities' estimates, it costs about $1.4 million to train just once, and tens of millions of dollars for some larger LLMs (large language models).

It can be seen that the success of ChatGPT is accumulated by Open AI's long-term investment and deep cultivation, which is not difficult to explain why Robin Li has always emphasized the word "accumulation".

In fact, ChatGPT is essentially a manifestation of AIGC technology. The so-called AIGC is to use artificial intelligence to automatically generate content, including text, pictures, audio, video, actions, etc.

Regarding the two major technologies of generative AIGC and NLP large model, many Chinese technology companies have been deeply engaged in this field for many years, and now there are mature products that have been applied to business, media, services and other related fields.

For example, on March 3 last year, Baidu launched the AI sign language digital human for the hearing-impaired, the intelligent Yunxiling AI sign language platform, which is based on the NLP sign language translation model of natural sign language, generates nearly 10 million "natural sign language corpus" sentences as training data, and generates sign language videos through AIGC technology, providing online communication, leisure, learning and other service help for the hearing impaired, and it only takes a few seconds to convert 1,000-character text to sign language.

On November 20 last year, HUAWEI CLOUD digital human, Yunsheng, born in MetaStudio (Digital Human Lab), was publicly unveiled at the first "HDC.Cloud HUAWEI CLOUD Developer Day" held in Shanghai. With AIGC technology and Pangu CV large model, MetaStudio not only allows everyone to have their own digital "doppelganger", but also provides services such as realistic modeling, video production, speech synthesis, virtual live broadcast, and virtual video production.

On January 16 this year, "Xiaoning", a digital human employee of Bank of Ningbo Shanghai Branch, hosted an online live broadcast event that combined virtual and real and broke the dimensional wall, not only introduced his work content humorously and humorously, but also professionally and cordially recommended a variety of financial products and preferential benefits of Bank of Ningbo to netizens. Based on SenseTime's original "virtual IP solution" and a variety of leading AI technologies, the digital human employee is a "virtual IP" created exclusively by SenseTime for Bank of Ningbo, which can achieve efficient and low-cost AIGC content creation, helping the bank achieve user accumulation and marketing conversion of front-end business.

On February 20 this year, the media reported that Jiang Ni, an IP digital human developed by DeepBlue Technology for the Wuhan Jianghan Road Smart Street Project, officially took the stage. According to reports, the digital human Jiangni uses the core technology of metamind products independently developed by DeepLan, NLP pre-trained multimodal large model, and has the capabilities of speech recognition synthesis, user dialogue intention understanding, and sentiment analysis. Through a large number of AI pre-training and the establishment of a knowledge graph related to the block, "she" can naturally communicate on the Jianghan Road Pedestrian Street Mini Program, cordially answer questions about the block, introduce the century-old history of Jianghan Road, guide tourists to target merchants to spend, push block preferential information, etc.

Obviously, the above digital people are all built with AIGC related technologies, and they are all similar to ChatGPT's "big data + large computing power + strong algorithm" "NPL pre-training large model" technical route, which can be called the application model of domestic AIGC and NLP large models.

According to relevant sources from the Institute of Artificial Intelligence of Donghua University, the application of AIGC in China has a precedent and has leading technology. For example, in the "2019 China International Fashion Design Innovation Competition" held in April 2019, DeepVogue, an AI designer known as the Chinese version of Alpha Gou, won the runner-up of the competition, using AIGC artificial intelligence content generation technology to design three different clothing, and also won the "Most Popular Award" voted by 50 public judges.

Through these successful cases, it not only shows the technical strength and development potential of Chinese technology enterprises in the development of ChatGPT-like products, but also proves that Mainland China has not only not lagged behind in the development and application of AIGC and NLP models, but has also empowered the industry.

The development of digital technology has gone through "Internet +", "Internet of Things +", "Big Data +", "Cloud Computing +", and the emergence of ChatGPT today means that the prototype of "big model + AIGC" has emerged, which is of epoch-making significance.

In January, Microsoft announced an expanded partnership with Open AI and an additional $10 billion investment. In this regard, Forbes magazine commented that although ChatGPT does not yet have a clear profit model, the development potential of artificial intelligence is huge.

Objectively speaking, ChatGPT has indeed led to the transformation of the global AI industry. Huang Tiejun, president of Beijing KLCII Artificial Intelligence Research Institute, said in an exclusive interview with reporters: "The success of ChatGPT should not only be regarded as a breakthrough in a new generation of chatbots, but should pay attention to its revolution to artificial intelligence and even the entire information industry." ”

But we must also understand that at present, ChatGPT is still only a technical paradigm exploration, and its functions are mainly focused on some open, standardized, and universal tasks, and its intelligence is still stuck in the past, and the real disruptive technological breakthrough has not yet arrived. We should have a clear understanding of this.

All in all, it is undoubtedly right to learn from others' strengths and seize development opportunities, but it is also necessary to give full play to your own advantages and adhere to the direction of the track. Only by taking the road of serving people's livelihood and empowering industries step by step, will more Chinese technology companies such as Huawei, Baidu, SenseTime, and DeepBlue shine in the world in the future, and AI technology will be like a spark, setting fire to all walks of life in China.

Alexandre Dumas famously said in The Three Musketeers: "Everyone for me, I for all." Since all the AI technology achievements made by mankind so far are inseparable from the help of all walks of life, then we should give these achievements back to all walks of life and empower their development, which is the right way for China's AI development.

Read on