Author |
Text | Ring says
Large-scale models are blooming everywhere in China, and those who have done or haven't done have to come and participate.
In the turbulent tide, it is inevitable that more people will start to pay attention to the first published Wen Xinyiyan.
Baidu, which was the first to release GPT large-scale model products among global technology companies, revealed some recent developments in Wenxin Yiyan at the Zhongguancun Forum just now:
"QPS query and inference response speed per second has increased by 10 times, and the inference cost has been reduced to one tenth of the original."
In the keynote speech "Large Models Change the World", Baidu CEO Robin Li looked forward to a new era of artificial intelligence centered on large models, and bluntly stated that large models will change the world.
Among them, it is not difficult to see that Robin Li is full of confidence in his own product Wenxinyiyan, and his vision for the future actually sorts out the context of how a large model can change the world with an industry attitude to some extent.
Starting from the internal working methods of the enterprise, to market competition, to international vision and human development, a roadmap to change the world with a large model step by step has surfaced.
With the change of working methods, the large model directly drives the transformation of enterprises and products
Within the enterprise, the large model is generating real value, and only if it can drive the transformation of the enterprise and products first, can the large model move towards commercialization step by step and change the world.
1. The interactive mode of "asking questions" has been changed, allowing corporate intellectual resources to focus more on cognitive upgrading
"In 10 years, 50% of the world's work will be prompt word engineering," and "natural language human-computer interaction will bring about a revolution in prompt words," predicted Robin Li.
After the large model brings artificial intelligence the evolution from discriminative to generative, AI can create independently to solve many problems.
In this case, the way of working within the enterprise will change dramatically.
Whether it is a manager or an ordinary employee, the ability to ask questions is more important than the ability to solve problems, because the latter has been solved by the big model.
Robin Li gave an example. It used to take half a day and a day for an assistant to check a specific business data. Now, as long as you put forward a request to AI, you can output the result in a second.
In this case, the quality of the performance often depends on whether the employees of the enterprise can properly ask the big model the questions that best meet the needs of the enterprise's development.
Its essence is to test the core cognition of all levels of the enterprise on the development needs of the enterprise.
Only when cognition is upgraded can fundamental changes occur in enterprise development.
This change in the way of interaction has returned to the high-quality enterprise management model in many academic theories, that is, everyone can concentrate on important, innovative and creative things, rather than being consumed in various tedious chores.
This kind of enterprise management model has been considered only in theory in the past.
No one has ever thought that the transformation of the interaction method brought about by the large model can be realized.
2. The service mode of "unlimited expansion" opens the ceiling of enterprise development
"Every customer can also have an exclusive 24/7 assistant who knows everything to serve him."
When talking about the redefinition of the big model to the specific business section of the enterprise, that is, marketing and customer service, Robin Li said.
With the generative capabilities of large models, the sleepless AI can communicate with customers with more human-like or even superhuman capabilities. In the past, the "knowledge map" mode seemed a bit silly. Smart customer service (can only interact within the map range ), will be completely upgraded.
The upper limit of a restaurant's business is stuck by the number of tables and the turnover rate; in the past, the company's marketing and customer service were also constrained by human capabilities.
For enterprises, there are not only high-quality services, but also "unlimited expansion". In some sections, they will no longer be restricted by physical conditions such as labor and places. High-quality services are enjoyed by customers with uniform standards, and the ceiling of enterprise development One side is opened.
And just as Li Yanhong said, "Whoever has the best way to communicate with customers will have this customer." The large model will help more companies obtain richer market opportunities.
3. The product upgrade of "native application" removes development obstacles and expands capacity space
DoNotPay helps people with lawsuits and writes legal documents, Jasper helps people write marketing and promotion copywriting, and Speak teaches people to learn foreign languages... Li Yanhong gave multiple examples in his speech to illustrate that big models are giving birth to AI native applications.
The so-called AI-native application refers to the application created entirely based on AI generative thinking in the era of large models, rather than the transformation and upgrading of old applications that "introduce" AI.
For enterprises, the emergence of such products essentially breaks an important product chain - "human-digital world-human".
In the past, the application products launched by many enterprises are often a kind of "digital media" existence, connecting users/customers at one end, and having real people to meet the needs at the other end.
Now, the AI native application mentioned by Li Yanhong only needs a simple chain of "digital world-people", and most of the service content can be realized by the digital world itself.
As a result, many of the obstacles plaguing this type of product are resolved.
For example, foreign teacher products no longer need to compete for high-quality, high-quality foreign teachers, and work task products no longer need a large number of real people to meet customer needs for copywriting and design.
Moreover, compared with human services, the digital world will have more subtle insights into people, and it is more likely to provide truly personalized services.
From the above three points, it can be seen that the change of the large model to the enterprise is not superficial but fundamental, and it is a complete reengineering of concepts and models.
Therefore, enterprises should not just think about "introducing" large-scale models, but should actually try to use large-scale models to reconstruct business and become "large-scale model-native enterprises".
Baidu conceived it this way, and it did the same thing itself.
Li Yanhong revealed that Baidu is using AI native thinking to reconstruct all its products, services and work processes, and to be "the first company to redo all products."
For example, the Ruflow intelligent work platform allows every employee to have a work assistant with rich professional knowledge and real-time response. Among them, the ability to intelligently summarize chat records has surprised many employees, "the whole is stunned."
The transformational value of large models to enterprises has also been widely recognized.
When talking about the market performance of Wenxin Yiyan, Li Yanhong said, "The market demand is very strong, and the enthusiasm of Chinese people for embracing new technologies is unprecedented."
The large model is pushing the market competition to a new dimension and bringing a new pattern
Looking from the inside of the enterprise to the outside, the big model has also pushed some related market competition to a new dimension, and may even change the existing market structure in these fields in the past.
First of all, the competition rules of cloud computing that are most directly associated with large models need to change.
Robin Li said, "The rules of the cloud computing game have been completely changed."
A long time ago, there was a dispute between cloud computing 1.0 and 2.0. At that time, Baidu insisted on the 2.0 model, that is, cloud computing should be integrated with "capabilities" such as artificial intelligence and big data, rather than just providing "infrastructure" such as computing power and storage. "(that is, the 1.0 approach).
It must be admitted that in the period of rapid digital development, manufacturers who emphasize computing power and storage have indeed gained a lot of market by virtue of the cloud computing resource sharing model.
As the cloud computing market begins to enter a period of steady development, Baidu's original push for "capability on the cloud" has now become an industry consensus.
Now, with the wave of large models, whether there are large models, whether the models are good, and whether the framework is good or not have become more important criteria for customers to choose cloud vendors.
In other words, the 2.0 model of cloud computing has evolved again, and the "capability" brought by the big model has not only become a necessity, but also one of the key factors affecting customer decision-making.
New market changes may happen here.
Then, the competitive nature of many products and services on the market will also change.
"In the future, all applications will be developed based on big models. Every industry should have its own big models, and the big models will be deeply integrated into the real economy." Li Yanhong gave this judgment.
This is also a conclusion based on the value brought by the large model to the enterprise and the product.
On the other side, Li Yanhong showed a new "IT technology stack" in the era of large models.
Compared with the traditional three-layer IT technology stack of chips, operating systems, and applications, this new IT technology stack has four layers:
The bottom layer is still the chip layer, but the mainstream chip has changed from CPU to GPU (for example, Baidu has mass-produced tens of thousands of Kunlun chips);
Above the chip is the framework layer, that is, the deep learning framework (such as Baidu Flying Paddle, which ranks first in China);
On the top, it is the model layer. The well-known ChatGPT and Wenxin Yiyan belong to this layer. In fact, there will be other industry models;
At the top is the application layer, which is the various AI native applications mentioned above.
The combination of the two actually implies a deeper reality:
Since large models need to be widely used, and a new IT technology stack is required behind the application of large models, the competition of many products and services in the current market is not just their own competition, but transformed into this new IT technology stack. A competition of technology stack capabilities.
For example, large-scale model capabilities are beginning to be delivered to smart terminals, and smart speakers are expected to become the next refactored product. In this case, the future competition of smart speakers will get rid of "homogeneity" (the capabilities of smart speakers seem to be similar under the knowledge map capability), and test whether the IT technology stack behind them can provide more sufficient content generation capabilities .
Strong large model, natural interaction;
The big model is weak, it just can't understand, it just can't create content, and it can't imitate others.
As a result, the competition of smart speakers is likely to open up a new situation. Many products and services may have to go through this process.
Finally, Dimension focuses on the competition of large model manufacturers, and the "follow the trend" style of play is no longer feasible.
The value of the aforementioned large model makes the large model itself a fiercely competitive market with deep commercial value space.
In the past period of time, a large number of manufacturers have entered the game, giving people the illusion that anyone can make large-scale model products.
However, those who follow the trend and enter the market may face the fate of being eliminated immediately, and the market structure will change from everyone's participation to a few capable ones occupying a dominant position.
The reason is that the competition of large models has actually become the competition of enterprise background. There are no dark horses here, only white horses accumulated through long-term innovation and practice.
When Li Yanhong analyzed the new IT technology stack, he showed Baidu's current deployment situation:
This full-stack system can only come from Baidu's long-term accumulation. It is impossible to see the big model and join it immediately.
Baidu recognizes that it has long-term investment and accumulation in AI technology, which is a prerequisite for the development of large models.
And more importantly, Baidu's "foundation" is not only for large-scale models, but also has successful practical applications in areas such as urban smart transportation. Li Yanhong gave an example of how Baidu's AI information control system has greatly improved Beijing Yizhuang area. traffic conditions.
In other words, the large model and its new IT technology stack are one of the many achievements on the soil of Baidu's AI technology, and are the product of the unique advantages of Baidu's AI technology, not even deliberately.
Competitors who follow suit may face the end of being eliminated after the large-scale model competition gradually deepens, and the reshuffle of the industry may be doomed.
Large-scale model risk control under good prospects: go to the poker table first, then set the rules
Finally, since the value of the large model is so obvious, will there be "side effects"?
Whether machines will replace people and cause massive unemployment, and whether AI will get out of control and cause moral crises for humans, these questions are often asked.
The large model undoubtedly brings the pace of general artificial intelligence (AGI) closer. Robin Li calls it "intelligent emergence" (the ability of the large model does not necessarily need to be taught by people, but can be "understood" by oneself). But before changing the world, large-scale model manufacturers and the industry also need to answer these questions, make preparations, and formulate relevant rules.
However, this is a matter that requires the joint efforts of global large-scale model manufacturers and the industry.
For China, it is necessary to worry about these things and to participate in the formulation of rules, but the premise is that it must first go to the poker table before the rules can be set.
In his speech, Li Yanhong mentioned that Wenxin Yiyan has achieved controllable data, framework and models, which "can reflect the high level of technological self-reliance and self-improvement in international competition."
With the performance of products such as Wenxinyiyan, Chinese large-scale models can already enter the global poker table with an independent attitude.
Regarding the follow-up risk control, Robin Li said, "To prevent loss of control, countries with advanced AI technology need to work together to formulate rules from the perspective of a community with a shared future for mankind. There are tickets to the global competition."
As for the next step, whether it can "help China's economy create the next golden 30 years" as Li Yanhong said, while ensuring that the risks are controllable, it still requires the joint efforts of domestic manufacturers including Baidu.
*The pictures in this article are all from the Internet
*This content is original by 【铃铃说】, without authorization, anyone may not use it in any way, including reprinting, excerpting, duplicating or mirroring.
#铃铃说Focusing on digitalization of enterprises and upgrading of industrial intelligence, this is an in-depth interpretation of new technologies, new models, and new ecology NO.343 focusing on all new technologies, new models, and new ecology related to entrepreneurship, industry and business
【over】
had rang the bell
1Top 10 authors of the year for venture capital and technology websites such as Titanium Media, Pintu, and Everyone is a product manager;
2 Tiger Roar Award judges;
3 Authors: Authors of best-selling books such as [Mobile Internet + Business Opportunities under the New Normal];
4 Special commentator for nearly ten newspapers and magazines such as "China Business News", "Business Circle", "Business Circle Review", "Sales and Market";
5 Nearly 80 columnists including Titanium Media, 36kr, Tiger Sniff, Jiemian, The Paper;
6. The creator of the concept of "brain artist" (brain craftsman), which has evolved into "We Media" and has become an industry;
7 Honorary Mentor of Tencent All-Media School, Communication Consultant of Several Technology Intelligence Companies.