The sudden arrival of GPT not only caught those who treated AI indifferently by surprise, but also led to an upsurge of discussion on "AIGC" among the whole people, making the public start to look forward to its arrival, and how much excitement it can add to this world
.... Looking forward, GPT is not humble either. With a swipe of a pen, it has been integrated into many fields such as collaborative office, painting, and IT programming. change.
Not long ago, Microsoft announced the integration of ChatGPT into the Power Platform Copilot low-code platform, and DingTalk also released a magic wand to integrate Alibaba Cloud's Tongyi Qianwen into the development platform. Including some time ago, Partner Cloud, a well-known low-code manufacturer in China, also announced that it has completed the integration of AIGC technology into product functions...
However, the unexpected encounter of two popular new stars, low-code and GPT, is destined to bring certain reconstruction and development. Disruption, but if discussed specifically, what can they ultimately refactor? Will the integration and co-creation between them be smooth sailing?
With these topics in mind, "ToB Industry Headlines" invited Partner Cloud CEO Dai Zhikang to discuss, hoping to draw some answers worthy of industry reference from his industry perspective and market cutting-edge observations.
01 Two supernovas, what can change the future?
"Software is engulfing the world now. In the future, all high-quality modern enterprises will become software enterprises, and application modernization has brought value to thousands of enterprises, and will become a modern enterprise with high-quality growth. " Executive Deputy Secretary of China Software Industry Association Minister Chen Baoguo said so recently.
The direction of reality is just like its speech. IDC predicts that by 2025, the digital economy will generate more than 500 million new applications and services, 90% of which will be cloud-native applications, and every enterprise in the future will be a software enterprise.
However, there are thresholds for traditional software development. For enterprises and institutions, if they choose to build themselves, not only the cost, energy, and construction speed are difficult to control. At the same time, the required talent pool is often not easy to achieve.
Of course, enterprises and institutions can choose to pay for the cost of purchasing third-party software service providers, but the problem is that most of the current software is not subtle enough to fit into the scenarios of thousands of smallest links and cannot meet the "last mile" service configuration.
These problems were once unavoidable but difficult to solve when enterprises are moving towards a software-led future development. Later, the low-code became popular, and it was resolved to a certain extent. To expand, low-code (or zero-code) is a low-threshold way to develop software. In theory, you only need to find non-technical personnel who are familiar with the business, and after training, you can quickly build customized applications that the enterprise needs.
The principle of this development mode is to integrate the source codes of many software functions to form modules of a certain standard system. By adopting and splicing corresponding functional modules, operators can create a simple software, which is not only convenient but also Very agile and fast.
However, although low-code is excellent, there are certain thresholds and limitations. Dai Zhikang said: "First, low-code development does have a lower threshold than traditional development, but low threshold does not mean no threshold. " He said that even if you choose to use low-code development, you still need users to understand software development.However, the use of zero code also requires corresponding personnel to conduct certain training and understand some simple data modeling and system building capabilities, which cannot achieve the effect of getting started. The threshold of the modeling link can even block about 80% of the business. personnel.
"Second, the software developed by low-code is composed of models and forms piled up and stitched together, and the models and forms are established products developed by the corresponding service providers and cannot be changed. Faced with the need for some integration of specific scenarios and industry requirements In the service link, the matching degree of software developed by low code is limited. " Dai Zhikang said.
However, with the development and popularity of AIGC technology, it has an excellent path to optimize these problems.
Dai Zhikang said: "First of all, the large model built by AIGC technology allows users to describe the application they want to create through natural language, and then build a simple program. And it can provide continuous improvement suggestions through natural language, Let it be adjusted automatically to eliminate the threshold." "
Secondly, low-code manufacturers can use the AIGC large model's own knowledge accumulation and the know-how cultivated by the enterprise to assist in functional modeling." Dai Zhikang said, "This is for It creates opportunities to adjust modules, forms, and output products with industry experience."
Low code is a way that enterprises need to pay attention to in the process of softwareization. AIGC technology is the key path for enterprises to enter the intelligent stage. However, low-code is not popular enough and flexible enough. AIGC needs an actual carrier to present intelligence. The two complement each other, allowing them to form a partnership of "racing driver and navigator" and work together towards a better future.
02 The vision is beautiful, but is the reality not rough?
The mountain of ignorance, the valley of despair, the slope of enlightenment. It seems that the development and application of any technology must go through these three stages. This also makes many people start to question that low-code and AIGC, which are both emerging trends, may have to go through a storm and accept some challenges in the short term after they are integrated.
In this regard, Dai Zhikang said that the two will definitely face some ups and downs, but not now.
"The capability advantages of large models such as ChatGPT and LLaMA are in September 2022. I have observed that compliance with the inherent sensitivity of technology companies and the pain points faced by current low-code and zero-code platforms."
Dai Zhikang said, "We started to make some feasible attempts on the partner cloud zero-code platform based on the large model early on. In the process, we determined that the integration between the two will go through a long development boom. period, and this is determined by the characteristics of the times. "
Indeed, from the perspective of technology-driven progress, although intelligence has been mentioned long ago, it has been unsatisfactory to the public in terms of application, because of its ability to understand and interact. It did not meet people's expectations. AIGC appeared in a new form of intelligence, which can cut into people's lives, but to achieve this effect, a medium carrier is needed.
As far as enterprise development is concerned, softwareization is an irreversible trend and the best starting point for digitalization and intelligence. Therefore, when the software becomes the best carrier for AIGC applications, it is close enough to people's life and work applications, and can give a good interactive experience and intelligent effects. It will inevitably promote its development at an extremely fast speed. Before it is fully popularized, it is difficult to meet development. Obstructed.
This is a bit similar to the new energy vehicles that have been advocated all over the world in recent years. They are the darling of the times and need to quickly enter the existence of comprehensive popularization.
Of course, it is not easy to integrate AIGC and low-code. Functional adaptation, data adaptation, etc. are obstacles, which also need to be carefully polished. Dai Zhikang and his partner cloud team also understand this, but in the face of the needs of the times, they still choose to forge ahead and polish in depth.
Dai Zhikang introduced that the generative low-code AI large model is different from other software products. Its pre-training "data" is the various "headers" of collaborative forms, process forms, and analysis dashboards.
For example, in the sales process of a new energy vehicle store, the "header" of the collaboration form is preset to be product name, product code, store product data, stage sales data, stock vehicle code, current remaining stock vehicles, etc., and the default process form is application Personnel-sales volume-existing vehicles-administrator approval-more detailed store data...
Through different sales links of different auto manufacturers, specific "data (header)" integration, learning, and training must be done. When a certain auto manufacturer Smart decisions can be made quickly when the need arises.
Based on this technical path, Dai Zhikang and his technical team spent more than half a year on "data" training, and built a generative low-code platform for Partner Cloud - AI Construction.
Dai Zhikang introduced that AI construction is a complex and multi-collaborative process compared to the past, which allows business personnel to directly find the "AI construction" entry on the partner cloud zero code platform, enter natural language in the dialog box, and describe what they want. , so as to get a pre-built application, and at the same time see the thoughts and inspirations built by other users.
Through continuous adjustment after reference and continuous refinement of pre-built application functions, a basically satisfactory product can be obtained. In this link, "AI construction" is not the executive role of "one-sentence generation application" as it is generally believed, but an auxiliary role of "assisting in building application robots".
Auxiliary roles have been needed by people for a long time, especially in the current era of talking about intelligence, which is destined to have a smooth development path.
Dai Zhikang said that the current "AIGC + low-code" model focuses on the ability of low-code platforms to combine AI, increase the upper limit of platform applications, and lower the lower limit of platform thresholds. Its appearance will not replace anyone, it will only help anyone, and it is destined to open its arms to anyone, and there will be no crisis before it is completely popular.