Source: New Picking Business Review
Author: Tuo Luo Suan
Image source: Generated by Wujie AI
Do you all have the feeling that in the tech circle, if you don't talk about large models, you will appear to be out of date?
The continuous heat of AI has its reasons. In the past, we taught machines what to do, and they learned it. But now, with large models, machines can gradually learn knowledge that was not taught to them before.
Large models can be said to be a watershed for AI. Similar to leading the "Fourth Industrial Revolution," it drives global economic growth. It can reshape industries, has enormous commercial value and growth potential, and no major company would be willing to miss this opportunity.
Baidu was the first to propose that all industries are worth redoing with large models. Whether it's Alibaba, Tencent, or Huawei, they are also following suit. As early as March of this year when Baidu released the large model product Wenxin, it emphasized the need to be the "first company to redo all products," not integration, not access, but redoing and reconstructing.
The battle of large models is about to begin. In the past few days, the first batch of domestically filed large models has been approved, adding fuel to the transition of large models from virtual to reality.
AI is ushering in a golden decade, and large models are accelerating their application, while the road to productization of AGI is just beginning.
This is undoubtedly the main battlefield that entrepreneurs should strive for. According to Baidu's understanding: in the true AI era, new applications will be developed based on large models, and native AI applications are the best way to deeply integrate AI technology with various industries, as well as the best way to promote social progress.
Especially after Wenxin was fully opened to the public recently, the entrepreneurial boom of generative AI has attracted more attention. According to official data from Baidu, after Wenxin was fully opened, it replied to over 33.42 million questions from netizens within 24 hours. It can be said that the full opening of Wenxin is leading the trend of natural language processing technology, and more entrepreneurs and developers are deeply aware of the broad development prospects of generative AI.
Unprecedented large model entrepreneurship competition results are out
The impact of large models on humanity may be more intense than that of the Internet era.
How to understand this intensity? There is a comparison on the internet, roughly translated as:
Just as the Internet era did not only rely on the laying of network infrastructure to create value, but required the emergence of internet companies like Alibaba, Baidu, Tencent, and JD.com to bring real value to users.
Similarly, in the era of mobile internet, having only iOS and Android systems is not enough. The key is that smartphones are equipped with applications such as WeChat, Meituan, Didi, TikTok, and Taobao, making it more convenient for people in their daily lives and entertainment.
In the future AI era, various native applications will be created based on mature large models.
Reshaping various industries with AI is not just talk, but to truly transform this vision into reality, it requires the joint efforts of numerous developers and entrepreneurs.
With the mindset of "reconstructing product applications with AI native thinking," Baidu held an unprecedented "Wenxin Cup" competition. At the Baidu Cloud Intelligence Conference that was trending today, Baidu announced the final list of winners.
This means that the winning companies will not only receive one-stop AI resources support from Baidu in terms of technology and talent, but also have the opportunity to receive a portion of the investment fund worth up to tens of millions of dollars.
Such huge resources and investments demonstrate Baidu's long-term commitment and attention to AI. After a three-month competition, a group of AI unicorn companies have emerged in various industries.
Nolibox is one of them. Established in 2020, it is one of the earliest domestic companies to explore the application of generative AI technology in the design field. Currently, Nolibox has integrated several large models such as Stable Diffusion, GPT, and Baidu Wenxin, and based on the secondary development of relevant models, it has built two core products, "Hua Universe" and "Tu Universe," providing services for intelligent design needs in industries such as clothing, e-commerce, gaming, and education.
Initially, Nolibox focused on image design in e-commerce and marketing scenarios, which required switching between different types of work and tools. It wasn't until the emergence of large models that it greatly alleviated the shortage and high cost of materials in the design field.
From limited to unlimited, from high cost to low cost, production materials are being restructured by AIGC. Nolibox CEO Xu Zuobiao pointed out, using the Chinese corpus of Wenxin large model as an example: "In terms of the accuracy of copywriting and recommendation, Wenxin is still quite advanced, especially in some customized fine-grained scenarios."
Chen Lifu, the founder of the e-commerce shopping assistant Buysmart.ai, also has a similar feeling. Overall, he believes that "the use of Baidu's large model is quite smooth," and it was able to run smoothly without much modification when replacing GPT.
There is also a second-dimensional community application called Genie AI, which initially chose to use the GPT model. But after switching to Baidu's large model, "whether it's fine-tuning the model or the expression of key words, the performance of the two is similar."
In addition to expressing recognition of Wenxin's strength, Genie AI CEO Peng Kangwei added, "Considering security and Chinese compatibility, Wenxin's large model will be better."
Undoubtedly, Baidu has provided a solid foundation for startup companies, and the Wenxin Cup entrepreneurship competition has become a hub for the most cutting-edge information in the large model field. AIGC products are showing a trend of diversification, and the 2C and 2B fields are mutually infiltrating. In this competition, potential projects cover productivity tools, marketing/sales, medical, and other fields, and span multiple vertical fields such as AI assistants, e-commerce, and law.
For example, ChatDOC, a professional knowledge question and answer assistant and reading platform based on document intelligence, and the AI-driven new material discovery platform Huanliang Technology.
In this way, unlike other domestic companies, Baidu is focusing on high-value industries and application layer projects, and is committed to building a prosperous, powerful, and vibrant AI ecosystem.
AI native applications are the main battlefield of the large model era
Many startup companies have embarked on the "acceleration" of large model entrepreneurship based on the development of AI applications using the Wenxin large model 3.5.
It is important to point out that choosing to embrace large models no longer puts one at a disadvantage. AI native applications pay more attention to the feasibility of technology and commercial prospects.
Due to resource and time constraints, many companies that initially wanted to develop the best large models need to rethink their entrepreneurial ecosystem positioning in order to develop solutions that are suitable for themselves on the basis of mature large models.
Just like the tech giant Microsoft, after the emergence of ChatGPT, they did not develop large models on their own, but chose to cooperate with OpenAI:
First, they launched the latest version of the Bing search engine and Edge browser with technical support from OpenAI, then fully integrated GPT-4 into Office, and finally added an AI assistant called Copilot in Windows 11…
Despite deep cooperation with OpenAI in the early days, Microsoft did not develop large models on its own, nor did it aggressively acquire OpenAI.
Fully utilizing other large models, developing based on other large models, and incubating native applications that play a role in specific scenarios in various industries is an inevitable trend.
Developing large models itself is not the goal; only applications have value. Especially against the backdrop of industrial upgrading, accelerating the practical application of large models from the "showroom" to AI native applications can truly provide the necessary support for people's production and life, and promote innovation and development of the Chinese economy.
However, it is worth noting that compared to the prosperity of large models, the number and scale of native applications in China are still limited. In contrast, almost every day, new AI native applications that deeply integrate into production and life are emerging abroad.
For example, the "hottest product in the document-to-image application track," Midjourney. Because of the factory workers who screw screws, the master Yoda who takes photos with villagers, and the space opera theater, these highly realistic photos have become popular.
There is also DoNotPay, which claims to be the "world's first robot lawyer." Founded by Joshua Browder, a computer scientist from Stanford University, the company developed based on the GPT-3 API, and in just 5 years, it handled over 3 million cases, helping users cancel fines worth $80 million.
Tech entrepreneurship is transitioning from "Internet+" to "large model+," and many companies have found a business model.
DoNotPay's current valuation has reached $210 million; Jasper, which focuses on AI text generation, can write TikTok video scripts. Founded 18 months ago, its valuation has reached $1.5 billion.
In addition, the online language learning platform Speak, even received a $20 million investment from OpenAI, the parent company of ChatGPT.
After Microsoft released Copilot in March, it sparked a discussion on "What is the Killer App of the AI era" a few days ago.
Five months later, at the "Xixi Forum" closed-door meeting, Kai-Fu Lee made a comparison: "WeChat gave up compatibility," and concluded that "Copilot is not an all-in large model product."
This conclusion has once again ignited curiosity about AI native applications. Many people support this view: if Copilot is removed, the Office software will still be Office, which could possibly become the most successful product of the large model era.
If AI is just the icing on the cake, then AI native is still far from ideal in Kai-Fu Lee's eyes. However, in any case, AI native applications will become the most worthwhile main battlefield of the large model era.
Consciously cultivating AI native thinking has become an urgent priority, so it is not difficult to notice that Robin Li has emphasized more than once in a conversation:
"The key point of the new international competition strategy is not how many large models a country has, but how many native AI applications are on your large models, and to what extent these applications improve productivity."
Baidu has already revealed its trump card as the default base model for AI developers
Whether it's the hosting of the Baidu Wenxin Cup or the face-to-face closed-door exchange meeting led by Zhang Yong, they all point to the same goal:
As a leader in the long-term deep cultivation of the AI venture capital circle, Baidu is playing the most powerful role in driving the development of startups, fully penetrating various industries, and empowering the full lifecycle of AI native applications.
In the future, China will definitely have its own large model ecosystem, and the Baidu Wenxin large model will become the default base model for AI developers. Why?
With Baidu's accumulated years of equity investment experience, a better understanding of how to tap the value of AI companies, and by going deep and practical, Baidu has taken the lead in proposing the landing approach of industry large models.
"Accelerating the use of the Wenxin large model and Wenxin's speech to build AI native applications, reconstructing Baidu's original product line and business system, and promoting long-term growth." This is an important battle for Baidu in the field of AI.
Through multimodal content generation, Baidu's AIGC commercial creative platform "Qingduo" can generate 100 creative copies in 2 minutes, create a digital human model in 3 minutes, and produce a complete digital human population video in 5 minutes; in addition, based on the capabilities of large models, Baidu has created the AI assistant "Super Assistant," which can compress tasks that would take hours to minutes or even seconds…
Not only within the company, but Baidu has also launched AI native applications in multiple fields such as healthcare, finance, law, education, and transportation, and has achieved significant results.
For example, Xiaodu Doctor can automatically generate diagnosis results and treatment recommendations based on user input symptoms or questions; Xiaodu Finance can automatically generate investment plans and profit forecasts based on user input assets or goals; Xiaodu Teacher can automatically generate learning content and answers based on user input courses or questions.
Baidu's AI native applications not only provide users with more intelligent, convenient, and efficient services, but also bring more commercial value and competitive advantage to Baidu.
According to financial report data, Baidu Intelligent Cloud's second-quarter revenue increased by 5% year-on-year to reach 4.5 billion yuan. Intelligent Cloud achieved consecutive quarters of profitability (non-GAAP), and Baidu insists on building a new engine around generative AI and large language models to drive long-term growth.
AI native thinking indicates the potential that Baidu can unleash in the era of large models, and the investment and accumulation in large model technology will directly determine how far Baidu can go.
According to the latest IDC report "AI Large Model Technology Capability Assessment, 2023," the Baidu Wenxin large model received the only full score in industry coverage and has achieved extensive business layout and landing scene exploration in fields such as energy, finance, education, and healthcare.
As early as 2013, Baidu began to lay out AI-related technologies and has accumulated research and development investment of over 140 billion yuan. As of April 2022, Baidu's global AI patent applications have exceeded 22,000.
The strong research and development investment has enabled Baidu to make breakthroughs. Today, it has independently developed all four layers of the essential architecture for large model development: chips, frameworks, models, and applications.
By playing a "combination" for these four core architecture layers, Baidu has further formed its own technological advantages, and the cost reduction of Wenxin's speech is proof:
Since the start of the internal testing in March, in less than two months, Wenxin's speech has completed four iterations, with a 10-fold improvement in inference performance and a 10-fold reduction in inference costs. As of May 23, after two months of internal testing, through the end-to-end optimization of the Baidu Intelligent Cloud Qianfan large model platform's four-layer architecture (chips-framework-model-application), Wenxin's speech has achieved a 50-fold improvement in inference performance in high-frequency scenarios.
As one of the earliest players in the field of generative AI in China, Baidu has proven its strong evolutionary capabilities.
Having a product means having a market, and having competitors means needing to compete. The situation of the "battle of large models" is largely to alleviate concerns about the "semiconductor hardware bottleneck."
To make domestic large models go global, it is urgent to find breakthrough methods. Leading technological breakthroughs proactively, represented by tech giants such as Baidu, is not only a show of strength but also a responsibility.
In fact, most companies neither have the ability nor the need to develop large models. Combining the capabilities of large models with enterprise business effectively to create stronger value will be an opportunity for companies to overtake others on the bend.
From this perspective, since the emergence of ChatGPT, which gave birth to the AI large model craze, Baidu seems to have entered a new era.
Market valuation largely proves this. According to third-party data, as of the close of September 5, 2023, Baidu's total market value on the Nasdaq market was 399 billion Hong Kong dollars, ranking second among Chinese internet companies and fourth among global internet companies.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
