With PPT entry, is ChatGPT a "good" track?

Recently, the more powerful GPT-4 was released, from a language model to a multimodal model, to achieve powerful image recognition ability, in the generation of lyrics, creative text, to achieve style changes and other leaps and bounds, in various professional tests and academic benchmarks performance comparable to human level, shocked the world.

At the same time, the domestic technology circle and venture capital circle, which have been "lying flat" for a long time, have stood up for a long time: Meituan Wang Xing announced that it will participate in the A round of investment of Wang Huiwen's ChatGPT startup "light years away"; Kai-Fu Lee announces the formation of the Chinese version of ChatGPT "Project AI 2.0"; 360 also eyed the new outlet, Zhou Hongyi appeared at the investor conference of a brokerage firm, personally popularized ChatGPT new opportunities; Baidu Robin Li re-entered the center of the stage with the help of "Wen Xin Yiyan".

For a while, ChatGPT has become the most beautiful "boy" and the hottest "track" in the technology circle and venture capital circle in 2023. The company has not yet been established, just draw a PPT, or even a dinner party, can attract a lot of money to "pay".

It seems that whoever misses ChatGPT misses the "BATJ" and "TMD" of the next decade. So, will ChatGPT domestic version be a good investment track? From a technical logical point of view, GPT can not only generate smooth and natural text, but also complete a series of NLP (natural language processing) tasks such as question and answer, translation, and novel creation, and initially have the "wisdom" of a certain person, replacing some simple and repetitive work; From the perspective of business logic, ChatGPT can replace a large number of repetitive and bottom-end labor with higher efficiency and lower cost, and has a wide range of application scenarios.

Even if the prospects look good, it is not easy to make ChatGPT domestic, here are a few aspects to pay attention to:

First, ChatGPT is an extremely expensive "burn money" game that requires an extremely large capital investment. At present, without seeing stable income, Microsoft's investment in Open AI alone has exceeded $10 billion, plus Musk and other investments, the financing scale may exceed 100 billion yuan. With such a large amount of capital to support an early-stage enterprise, even in the most frenetic era of VC/PE in the past decade, it is difficult to support in China;

Second, large model training requires more advanced GPU chips A100/H100, which are currently in a "disconnected" state. As a replacement and castrated version of the A800 chip, there is currently a shortage in China. No advanced chip support, training or big discounts.

Third, the quality of the Chinese Internet is worrying, the phenomenon of information islands is serious, and training is difficult. Chinese the Internet accounts for only 1.3% of the world's top 100 websites, while English accounts for 60%. In the past decade, this information has been closed in countless closed apps and several large platforms that are closed and opposed to each other, such as public accounts, headline accounts, Douyin accounts, Baijia accounts, NetEase accounts, etc.

If it is only less information and closed, it is difficult to say anything about the quality of information at present. Baidu, Ali, Tencent, Toutiao, the four giants have directly or indirectly blocked their own crawlers, only B station, Zhihu, Weibo and other smaller platforms allow other search engines to crawl relatively freely, which further leads to each search engine crawling can not crawl useful information.

Policy logic: There is still great uncertainty

From the perspective of business logic, although success is not easy, after all, things are man-made. But in the great new era, we judge whether a track is worth entering and whether it is worth investing, it is not enough to talk about technical logic and business logic, but also to talk about policy logic, that is:

Does this track contribute to the high-quality development of the new era? Is there a social benefit while achieving economic benefits? Is it in line with the needs of national strategic development? Is it in line with mainstream values?

According to the policy logic, the investment and entrepreneurship track can be divided into several categories:

The first category, encouraged tracks: scientific and technological hardware innovation, import substitution/neck technology, economic security, clean energy, high-end manufacturing, promoting common prosperity, artificial intelligence, big data, cloud computing, etc.;

The second category, restricted tracks: model software innovation, barbaric expansion of Internet platforms, disruptive innovation, and high-energy-consuming industries;

The third category, the intermediate track: between the encouraged class and the restricted class, the track dominated by business logic and technical logic.

A good track should not only have good business logic, but also conform to policy logic, that is, the best industrial development meets the needs of the times, and the bottom line is that it cannot be a restricted track. Otherwise, a policy "iron fist", no matter how deep your moat, no matter how high the technical barriers, how smooth the business logic, and how large the imagination space, you will be immediately beaten back to the original shape. Education, Internet, games... Over the past few years, vivid examples abound.

So, which category does ChatGPT fall into? Generally speaking, any track, a company, has multiple "labels", from one aspect, it belongs to the encouraged category, but from another perspective, it belongs to the restricted category. For example, Tencent is both an Internet platform, a game developer, and a cloud computing/big data giant. ChatGPT belongs to the authentic artificial intelligence and big data track, and from this point of view, it should be in the encouraged category.

But if you zoom in, ChatGPT also has many labels with certain and potential "original sin", which belong to the category of restrictions and "headwind" direction. This means that if you run wildly, the potential risk of ChatGPT being hammered at some stage is still not negligible:

First of all, it is not hard technology, nor is it lost to the stuck neck project, ChatGPT has the characteristics of soft mode innovation, will it be limited by this? At present, the innovation direction encouraged by the new era is in chips, in the field of hard technology innovation, import substitution and stuck neck. Software innovation, especially model innovation, is suppressed to a certain extent, and the channels for listing companies will be greatly narrowed in the future. In some ways, ChatGPT seems to be "lackluster" in terms of innovation, relying more on stacking massive chips to "make miracles", training, and not high technical content. The training results are to replace existing jobs and jobs with more efficient and low-cost solutions, which seems to be standard "model innovation" from this point of view.

Once positioned as a low-tech model innovation, the channels for ChatGPT investment in domestic listing and exit may be greatly narrowed, and the development prospects will also be greatly impacted.

Second, will ChatGPT be a "disorderly expansion" and "leap" in the influence of Internet platforms? As mentioned earlier, the developer of ChatGPT, Open AI, is mainly supported by Internet giants such as Microsoft that burn nearly 100 billion yuan. After that, Internet giants such as Google, Facebook, and Twitter also announced their participation in the ChatGPT arms race. ChatGPT is a super money-burning game.

GPT-3 training costs about $1.4 million once, and for some larger LLMs (Large Language Models), training costs between $2 million and $12 million. Back in China, there are no more than 3 Chinese companies that can afford to play this game, and the probability is that these are the Internet giants with abundant cash flow.

According to an interview with Eleven Financial Eleven, at least 10,000 Nvidia's A100 chips are required to run ChatGPT. However, there are no more than 5 domestic GPU (graphics processing unit) chips with more than 10,000 pieces. Among them, there is at most only one company with 10,000 NVIDIA A100 GPUs. This shows that ChatGPT can only be a game between giants, and it will also strengthen the leading edge of giants.

Once the Internet giant ChatGPT is successfully cultivated, the moat, technical barriers and competitive advantage it brings to the company will be multiplied, which means that the Internet giants will once again achieve "barbaric expansion", and its subversion of traditional industries will be revolutionary, far from comparable in the past decade, and is this tolerated in the current environment?

Third, will ChatGPT cause a strong backlash while bringing productivity jumps, but also huge "creative destruction" to the existing social structure? There's no denying that ChatGPT and the AI revolution behind it have brought a huge productivity jump, but it will inevitably have an impact on existing social jobs.

According to a March 20 report from Open AI: "Researchers estimate that ChatGPT and future applications built using the program could affect approximately 19 percent of jobs in the U.S., and at least 50 percent of their job tasks." At the same time, 80% of the U.S. workforce will have at least 10% of their job tasks affected by ChatGPT in some way. ”

In the past few years, the community group buying burning war, the substitution of small traders and hawkers, has caused a strong backlash in society. The impact of ChatGPT on the existing social pattern is much more violent than the "barbaric expansion" and "model innovation" of the Internet in the past decade, which may aggravate the gap between the rich and the poor and the pressure on social employment in the short term.

At a time when employment pressures are enormous, can we withstand the "creative destruction" and "social shock" of productivity jumps? Will ChatGPT end up dying of its own "over-success"?

Finally, ChatGPT consumes a lot of resources, will it affect the "security", "dual carbon" and "stability" of the economy? As mentioned earlier, large model training requires at least 10,000 A800, and the procurement cost is only 1000000000000000000000000000000000000000000000000000 More importantly, advanced GPU chips are in the "stuck neck" state, which is already in short supply, and the basic domestic needs cannot be met, too much for large model training, is it too luxurious, will it affect the operation safety of the entire economy?

In addition, the daily power consumption of large model training is very staggering. It is estimated that ChatGPT runs for a month, which is equivalent to the electricity consumption of 175,000 people in the West. Considering that the power consumption and performance of the chips we use are poor for 1-2 generations, the power consumption will be even more amazing. At a time when the stability of power supply has become a challenge, excessive ChatGPT competition may affect the stability of the power grid and bring about an increase in energy consumption.

At that time, mainland digital currency mining once accounted for 80% of the world's total, but it was banned because power consumption affected economic security. ChatGPT certainly doesn't consume as much power as "mining" initially, but it could be a potential problem in the future as training progresses.

At present, the ChatGPT global competition has begun, are we actively entering? Or did you choose to "lie flat and watch" from the beginning? The emergence of new things will inevitably have an impact on our existing cognition, and will also help us to carry out cognitive iteration. As a new thing, whether ChatGPT can take root in the mainland depends not only on the ability of mainland entrepreneurs, entrepreneurs and technology experts, but also on the recognition and positioning of ChatGPT at a higher level. In some ways, the latter may be more important, in the following ways:

How to balance technological progress, productivity leaps and social stability? How to balance the traditional and digital economies? How to view the relationship between the "expansion" of Internet platforms and "participation in international competition"? How to view the relationship between "model innovation" and "hardware innovation"?

Recently, the central meeting has frequently blown warm winds and actively supported the healthy development of the Internet industry, especially in "leading development", "creating jobs" and "international competition". In particular, the digital economy has become a key force for reorganizing global factor resources, reshaping the global economic structure, and changing the global competition pattern.

This shows that the top-down tolerance and openness of "disruptive innovation" is much greater than expected; Continued support for the productivity leap brought about by the digital technology revolution. The environment faced by new technologies and models represented by ChatGPT has continued to improve substantially in recent years. ChatGPT localization deserves a better future!