Chaotic Age of Large Models: Contradictions, Differentiation, and the Future

After more than half a year of turmoil, China's large-scale model industry has entered a new cycle. One side is fanatical and the other side is cold. Investors are non-stop looking for China's OpenAI in this era. The top executives of big factories or scientists with elite halos decide to give it a go. The AGI era that was once unreachable seems to have a new look because of the appearance of large models. Visible timetable.

And the intelligent life in science fiction is gradually emerging, and the relationship between technology and people, technology and industry, human civilization and technological civilization seems to be moving towards a new stage of reconstruction.

According to the "China Artificial Intelligence Large Model Map Research Report" released by the Institute of Scientific and Technological Information of China, as of the end of May, 79 large models with a parameter scale of more than 1 billion in China have been released. In the following two months, Alibaba Cloud's Tongyi Wanxiang, Huawei Cloud's Pangu 3.0, JD.com's Yanxi, Ctrip's "Word", NetEase Youdao's "Ziyue"... Whether it is Get up early, or go to the evening market, with the joint participation of various forces, according to incomplete statistics, there are more than 120 AI large-scale models released in China, and the "Hundred Models War" is in full swing.

But in the midst of the bustle, cold thinking about the big model gradually surfaced: whether there is a good business model, the ecological dispute between open source and closed source, and the implementation of the 2B/2C route. It seems that no consensus has been reached on all the debates , Innovation and alternation are happening all the time.

Some harsher realities have also emerged: for example, the subtle competitive relationship between large companies and small companies, and some casing products that lack core technical foundations have gradually lost their halo in front of large models. For example, how to turn AI from toys to tools in the 2C market requires a unique insight into user needs, and how to ensure the controllability of technology and the ease of use of large-scale model products on the 2B side, which raises a higher level of industry understanding for players. Require.

In this chaotic age of large models, there are gaps in information and cognition, consensus and non-consensus. Hot money, talents and scenario applications, computing power, data and algorithms all determine the direction of technology and the fate of entrepreneurs in it.

01 Contradiction: Hot money is pouring in, but be cautious

Under the upsurge of large-scale models, the restlessness of capital is a particularly clear line. Looking at the changes in the history of human science and technology, investors with a keen sense of smell will always choose to bet heavily on those "seeds", and the same is true for the competition for large models.

According to data released by research firm PitchBook, in the past six months, global venture capitalists have invested more than US$40 billion (about 290 billion yuan) in AI start-ups. AI start-ups are still in the limelight.

Among them, the most notable are two investments: one is Microsoft’s $10 billion investment in OpenAI, and the other is Inflection AI, a start-up company established in 2022. After completing $1.3 billion in financing at the end of June, its estimated The value also reached 4 billion US dollars.

A consensus is that in the United States on the other side of the ocean, a prosperous AI unicorn ecology has been formed.

In addition to the well-known OpenAI, Anthropic, and Inflection AI, the main product form is Adept, which is a virtual AI robot assistant, Cohere, which focuses on B-end enterprise services, Stability AI behind the Stable Diffusion image generation diffusion model, and Nvidia. Computing power provider CoreWeave...In short, whether it is at the model layer, middle layer, or application layer, the large-scale model ecology overseas is clearer than domestic ones. Quoting a sentence in the report of "Geek Park": "There are almost no new entrepreneurs who want to be the next OpenAI."

In contrast, in China, although a large amount of capital has also flowed into the artificial intelligence industry, if you trace the flow of money, smart money still flows to a small number of leading companies. According to the statistics of Huxiu, from the release of ChatGPT to the present, there have been only 21 financing events on the AI large model track. And the star unicorn companies that we are familiar with are only MiniMax, Light Years Beyond, Baichuan Intelligent, etc., and the rise of the above-mentioned unicorns has both early first-mover advantages and the endorsement of the experience of big players.

Image credit: @chiefaioffice

The previous debate between Zhu Xiaohu and Fu Sheng caused a debate in the venture capital circle around the value of large models. Behind the "big refinement model" in the capital circle, investors are still very cautious in their actions. On the one hand, the AI model is a highly professional and subdivided track, and at the same time it burns a lot of money. Therefore, it is decided to focus on the AI field, and investors and investment institutions with accurate insight into technology are actually very scarce.

On the other hand, there are still too few good targets. Judging from the investment logic of current institutions, investment is still the main theme. Either it is endorsed by an entrepreneurial leader such as Light Years Away. Although the founder does not understand technical principles, he understands the changing trends and business models of the technology industry very well, or is a well-known technical scholar in the AI industry, such as Zhipu AI, Listening Start-ups such as Mind Intelligence and Shenyan Technology have "Tsinghua Department" behind them.

02 Differentiation: Big factories frantically accumulate bureaus, small factories desperately dig gold

Behind the series of changes around the big model is not only the progress of technology, but also the promotion of key people and key enterprises. If the lens is turned to the companies and people at the forefront of these waves, differentiation has actually occurred.

Dai Yusen, managing partner of ZhenFund, once had an ingenious metaphor: the emergence of GPT-3 is equivalent to the discovery of a new continent, and the emergence of ChatGPT is like the discovery of gold in the new continent. The Chinese company's catching up journey is like knowing the location of the New World and gold, and knowing that OpenAI is going there by boat, and knowing the general appearance of the boat, but without a detailed map.

After experiencing the crazy "release month" of the big model before, we can clearly see that this round of entrepreneurship is divided into the academic school, the big brother school and the big factory school. The relationship between them is not completely zero-sum competition. It is a "non-zero-sum game".

In the past period of time, in addition to showing the hard power of technology, the big manufacturers have become the main task of collecting bureaus and building ecology. Take Baidu, Ali, Huawei, Byte, and JD.com as examples. On the one hand, they have their own cloud services to provide computing power support. On the other hand, they also have their own layouts around the chip layer, framework layer, model layer, and application layer. , to further strengthen the barriers.

But among them, the play styles between big factories are also different. Major manufacturers represented by Ali, Baidu, and Huawei are more inclined to take the road of vertical integration, and achieve more than one fish in the three layers of computing power, platform, and model. On the other hand, Volcano Engine (Byte Cloud) and Tencent Cloud tend to take the platform route, build a model shelf supermarket, access more third-party large models, and provide corresponding fine-tuning, evaluation, and reasoning services.

For small domestic entrepreneurial factories, in the early stage of the competition for large models, in fact, the only certainty for start-up companies is "uncertainty". It does not need very complicated products, and it can be realized by targeting the pain points of users. initial success.

The "Miaoya Camera" that has been out of the circle recently is a typical case. The team said in an interview: "If AIGC's products do not charge money on the first day, they may not receive money." Through the use of low threshold, the marketing fission of social media is superimposed on the precise positioning of women who need photo. Even if there is no obvious innovation in technology, early commercialization can be realized with a single function. Miaoya has actually given the domestic application layer A good inspiration for startups.

For more start-up companies such as Miaoya Camera, how to grasp the "uncertain" cycle and further consolidate their own technical barriers and user stickiness is the key.

Image source: Screenshot of Miaoya Camera Xiaohongshu

03 Future: Intensified supervision, undecided pattern

In the foreseeable future, perhaps just like the arguments in the PR articles of major manufacturers, large-scale models will eventually empower thousands of industries, but beyond ideals, how to ensure the safety and controllability of large-scale model technology has also become an issue. focus of attention.

Previously, seven departments including the Cyberspace Administration of China jointly announced the "Interim Measures for the Management of Generative Artificial Intelligence Services", which provided a reliable legal basis for the future compliance and healthy development of generative artificial intelligence in terms of supervision methods and scope. And in the early hours of August 1, Apple’s app store in China removed a number of AIGC apps, which in fact also hinted at the policy side’s warming up of artificial intelligence supervision.

Overseas, tech giants are already facing thorny AI ethics controversies. The "AI Big Four" Anthropic, Google, Microsoft, and OpenAI jointly established the Frontier Model Forum to communicate with the United States, Europe, and the G7 on responsible and safe artificial intelligence issues. A coalition of open source communities such as Hugging Face, GitHub, and EleutherAI is also calling on EU policymakers to protect open source innovation when formulating the EU AI Act.

For entrepreneurs in the current large-scale model industry, in addition to entrepreneurial ideals and commercialization paths, the consideration of business model compliance will also be included in existing plans.

In addition to clear regulatory trends, more cutting-edge explorations are also taking place. The current industry discussions on a series of topics such as multimodality, AI agents, vector databases, and embodied intelligence are actually outside the upsurge of large models. More possibilities.

Taking AI robots in the field of embodied intelligence as an example, technology giants including Google have increased production by connecting large language models to robots to make robots smarter. And the same fiery wave of AI agents, even called "original AGI", has taken over the big model and become the next area of concern for big companies.

The tide has come, the future has come. To be sure, the chaotic era of large models may not last long, but competition and cooperation will continue for some time to come. Who can take the lead in using "uncertainty" to make up for shortcomings, who can truly implement large-scale model capabilities in subdivision and vertical scenarios, and who can build high-quality data flywheels faster, which will test their determination and endurance. It will also determine their respective ecological niches in the next round of competition.