The speed and depth of the technical essence cognition, scenario technology application and corresponding commercial exploration of LLM (Large Language Model) that opens the door to the future of AGI (artificial general intelligence) in China exceeds imagination.
In December 2022, OpenAI launched a revolutionary ChatGPT application. This move once again ignited the incomparable enthusiasm of domestic technology companies to pursue AI: within half a year, medium-sized technology companies, technology start-up companies and technology application platform companies have all entered this round of AI technology innovation and application to varying degrees.
Of course, it is important to develop software and hardware products for AI technology applications, but the technical understanding and prediction of AGI's future form by the top leaders of a business organization is the key to truly observing what kind of competitiveness a business organization has in the wave of new AI technology development and how far it can go.
Recently, Zhongke Chuangda demonstrated its various software and hardware products including Rubik Language basic semantic large model, and from the explanation and prediction of LLM's cognition and future form, business application depth and strategic layout by Zou Pengcheng, CEO of Zhongke Chuangda, it can be seen that in addition to platform giants and entrepreneurial technology companies, it is a medium-sized technology company located in the middle layer, promoting the sample mainstream of AGI technology business exploration.
Exploration of the underlying technical logical path of AGI
A week ago, Chuangda released a variety of AGI product types, covering the entire life cycle from underlying software and hardware development to terminal applications; At the same time, Chuangda also released the intelligent native operating system, the R&D direction of large models and the strategic layout framework. Among the dazzling software and hardware products, it also includes the new application of Rubik in the field of smart cars: Rubik GeniusCanvas.
Although the AGI products released by Chuangda this time have a wide range of products and powerful functions, what is really worth paying attention to is the underlying technical thinking of AGI by Chuangda of Zhongke Chuangda.
Before understanding these technical thoughts, it is necessary to clarify the basic concepts. For example, what is AGI? First, AI (Artifical Intelligence), or artificial intelligence, is "any end device that has the opportunity to sense the environment and take the most appropriate action to successfully achieve its goal, which is artificial intelligence, also known as machine intelligence."
There are two core points of AI: perception and decision-making, that is, perceiving the environment and taking the most appropriate action to increase the probability of success in achieving the goal.
Second, because there is a huge gap between machine intelligence (AI) and human intelligence, more AI capabilities are needed to match or surpass human intelligence: for example, the ability to reason, use strategies, solve problems, and make decisions when there are uncertainties; Demonstrate the ability to know, plan, learn, and communicate using natural language, as well as the ability to synthesize these new competencies to achieve set goals.
Devices with these new capabilities comparable to human intelligence can be called AGI (Artifical General Intelligence). Artificial Super Intelligence (ASI: Artifical Super Intelligence) is the next stage in the development of these AI devices.
In terms of the AI capabilities demonstrated by ChatGPT4.0, it is currently in the AGI stage. The AI exploration of domestic technology companies has also maintained phase synchronization.
So, what is the relationship between LLM and AGI?
From the perspective of the underlying technology, AGI is a pre-trained language model based on the Transformer (Semantic Understanding and Formation of Logical Language Output) architecture. To ensure the effect of Transformer, you must have a very large language model, that is, LLM (Large Language Model).
LLM is a giant natural language processing model that Google launched in 2019. With trillions of parameters, the model excels at a wide range of natural language processing tasks.
There is an AI technology company "JUNLALA" in the United States: it has been deeply engaged in the field of LLM models for many years, and has continued to explore and expand the limits and boundaries of LLM models for a long time. JUNLALA BELIEVES THAT THE LLM MODEL IS THE ONLY WAY TO AGI; AGI's breakthrough in key technology fields is inseparable from the growth of LLM model scale.
In fact, LLM model size continues to grow rapidly in various fields. The LLM model covers many vertical fields (scenarios), such as office meetings, sales, B2B interaction, search, chat, legal, academic, accounting, and medicine.
LLM covers so many industries (subdivided scenarios), will AGI have compatibility problems? After all, AGI is AI hardware with some human intelligence that still runs on top of the existing Android or Windows operating system.
"AGI can be thought of as a new operating system, or AGI can be understood in terms of operating system." Hu Yu, co-founder of iFLYTEK and chairman of Future Intelligence, said, "From the perspective of the iteration speed of LLM, AGI may appear in various forms. If AGI is an operating system, then its existence in multiple industries, compatibility with each other will no longer be a problem. ”
LLM: A new generation of interactive operating systems
Similar to Hu Yu, Zou Pengcheng, CEO of Zhongke Chuangda, believes that LLM in various industries is regarded by some as technology and as data by others, "From the perspective of operating system, LLM has two characteristics: complexity and interaction. Operating systems were born for only one reason, and that is applications, and I believe LLM is the same. ”
It seems that Zou Pengcheng and Hu Yu made a technical explanation of AGI pendant scenarios from the aspects of software and hardware respectively. Given the relationship between AGI and LLM, Hu Yu and Zou Pengcheng should mean the same thing. Because Hu Yu also mentioned, "In the AGI era, LLM itself can also be regarded as an operating system. ”
Zou Pengcheng mentioned the interaction characteristics of LLM, Hu Yu said, "Touch defines Smart Phone, which is compared with AGI, which belongs to the previous generation of interaction mode products; And the law of technology is the new mode of interaction, which will drive new forms of terminals. Hu Yu did not specify what the new interaction model is, while Zou Pengcheng believes that "LLM has the ability to realize the dream of natural interaction." ”
What is natural interaction? How is it different from the Touch interaction form of a smartphone? Simply put, natural interaction is the technology that supports the natural user interface of people and terminals to interact or interact with each other. In a more natural and direct way, people can control AI terminal devices (smartphones are also a type of AI devices).
For example, the initial human-computer interaction requires the keyboard to type commands, or the mouse to control the terminal device. Then there's the touch-based approach to controlling smartphones, but at the natural interaction stage, neither a keyboard or mouse nor touch. In the age of AGI, human-computer interaction may not require touch, and the same function as keyboard and mouse can be achieved with only sound (currently implemented) or even ideas (not yet realized).
Natural interactions are likely to also include the creation of new applications based on the LLM operating system.
Zou Pengcheng believes that LLM is a new generation of operating system. "As a new operating system, LLM is different from traditional Android and even Harmony system: in addition to being distributed, there is also a dynamic spontaneous. That is, LLM automatically generates demand, automatically generates application, which has never been done before. It used to take a long time for programmers to do the application, but now the operating system can finally automatically generate new applications. ”
The essence of natural interaction is to make operation easier, similar to the technical effect of the operating system. The task of the operating system itself is to solve the problem of complexity: to be able to translate complex hardware and software, such as cloud computing and model technology, into value that users can understand.
What is the role of the operating system?
"Operating systems need to translate LLM capabilities into applications that are acceptable to users or interact with well." "Three years ago, we developed a platform, which is WuChat. Simply put, WuChat allows people to communicate with each other in language with mobile phones, car machines, etc. For example, users do not communicate and control with computers, refrigerators, air conditioners and other applications without any software or hardware. On this basis, we provide a full set of enterprise tools for data management and data annotation. ”
It is necessary to point out that LLM is divided into general public large model and vertical large model. Between the two, Zou Pengcheng believes that through a distributed approach, the transformation from general to vertical scale can be achieved.
"No matter how smart LLM is, it still needs to drive hardware, security, real-time control, multimedia display and distribution. "The role of distribution is to distribute data and computing power to the right places. In this way, different forms of large models Fine-Tuning (fine-tuning) can be divided into domain (vertical domain or subdivision scene) models.
Smart home and automotive cockpit applications
Why emphasize the necessity of pendant LLM? What capabilities are needed to establish a universal LLM?
Zou Pengcheng said, "I believe that whether it is technology, business, or even finance, everyone pays attention to the general big model. This requires a fusion of algorithms, computing power, data, and servers to complete. ”
From the current status and trend of LLM technology development, there will not be too many companies with general LLM capabilities, and more companies will build their own or cooperate with ecological partners to build pendant LLM, and will continue to optimize the capabilities of these pendant LLM.
Chuangda is no exception. They built the Rubik Enterprise tool layer, which provides the support base for the entire IT through the underlying Rubik OS, and the combination of the two can better use the capabilities of LLM for applications.
To realize the vertical LLM capability for the application, various hardware bearers are also required. As Hu Yu said, "The essence of the operating system is what kind of hardware the application service selects as the standard carrier." "Zhongke Chuangda has launched a new type of hardware: Rubik device family. When presented as an application terminal, these two are a combination of software and hardware.
How to apply? That is, how to commercialize?
Simply put, through Rubik server, Rubik box and IDC (Data Center: Internet Data Center), combined with front-end services and cloud computing power, it provides demand application solutions for various industries.
Among them, Rubik box, user-oriented (B&C), supports enterprise databases, personal knowledge bases, as well as health, medical, pet and kitchen management applications, and more.
From the perspective of application characteristics, Rubik box not only provides the functions of traditional NAS (Network Attached Storage) system (Note: NAS is the storage server operating system), but also home intelligent management applications and car cockpit application management.
For example, a cattle farm in Fangshan District, Beijing, has achieved cost reduction and efficiency improvement through WuChat and Rubik OS of Zhongke Chuangda.
Compared with cost reduction, for example, before the application of the Kechuangda AGI solution - the B-end application development cycle of this cattle farm starts from 6 months, after the application, the cycle is shortened to 4-6 weeks, and the efficiency of process customization is increased by 60% - more importantly, efficiency has been increased.
As a cattle breeding expert rather than an IT expert, this cattle farm originally needed to reshape the entire IT process when facing new needs, but now through Zhongke Chuangda's WuChat technology and vertical LLM, all cattle-related physiological and environmental data, various changes in sensing equipment and other information, can be directly read into and realize the collaboration between people and equipment, and automatically generate application solutions.
In terms of smart home, Zhongke Chuangda has served a lawn mower manufacturer. After the application of Creang Da Pendant LLM and Rubik OS, the R&D and maintenance time was reduced from the standard team of more than 30 people manpower and 9 months, which greatly reduced the time and maintenance costs.
The intelligent cockpit of smart cars, Zhongke Chuangda also realizes a more engaging (interesting) interface of the intelligent cabin environment through tools such as Rubik Language (basic semantic large model), Rubik Copilot (coding efficiency tool), Rubik device (vertical LLM deployment tool) and Rubik GeniusCanvas (intelligent cabin design plugin).
Among them, Rubik GeniusCanvas is based on Rubik Studio, automotive HMI design software and 3D engine Kanzi, which can provide designers with super intelligent assistance from concept creation, 3D element design, special effects code generation and scene construction production.
Through Rubik GeniusCanvas, designers only need a few simple language conversations, Rubik GeniusCanvas can design images and build models according to requirements, greatly improving the design efficiency and quality of automotive cockpit HMI.
In just a few thousand words, it is difficult to fully explain the overall picture of the depth of understanding of the underlying logic of the technology of the relationship between LLM and AGI technology (application). From the current ability of the application cases of Zhongke Chuangda, it is not difficult to see that the business prospects of this technology company are moving towards a broader blue ocean of new technologies under the blessing of "AI+".
This article is from Wall Street News, welcome to download the APP to see more