Targeting AIGS: The Fourth Paradigm Release Model, Reconstructing Enterprise Software with Generative AI

ChatGPT craze has swept the world, creating a wave of exploration of large models. In April this year, major domestic manufacturers launched major models one after another.

Recently, 4Paradigm demonstrated its large-scale model product "Shishuo 3.0" to the public for the first time, and proposed theAIGS strategy (AI-Generated Software) for the first time: using generative AI Refactoring enterprise software.

Shishuo will be positioned as a new development platform based on multi-modal large models to improve the experience and development efficiency of enterprise software and realize "AIGS". So what exactly is AIGS, why is it positioned as AIGS, and how to implement AIGS? Fourth Paradigm gave a comprehensive answer at the Media Development Day.

Targeting AIGS: Transforming the entire software industry and helping business value achieve a qualitative leap

Dai Wenyuan, founder and CEO of Fourth Paradigm, believes that the current B-side software is extremely complex The interactive experience and the extremely low development efficiency caused by complexity leave enough room for reconstruction and transformation of generative AI.

In Dai Wenyuan’s view, in the past decade or so, C-side software products have been polished to a higher level, almost approaching the upper limit of user experience. In comparison, B-side software products often have more than ten levels of menus, and it is difficult to call the functions of enterprise software through natural language.

"Now when we have stronger semantic understanding and generation capabilities, coupled with GPT task translation, task distribution and reasoning capabilities, we can implement functions through better interactive methods To call, you no longer need to find a function located under more than ten levels of menu directories.” The user experience of enterprise software can undoubtedly achieve a disruptive improvement.

Furthermore, the original B-side enterprise software was highly customized and based on menu-based development. Basically, one function was upgraded at a time, and the product manager was required to draw the interface, design, and develop it. Wait, at least a month-level development time. Due to the emergence of new forms of interaction, functions and execution logic used to be arranged in the software interface, but now functions and logic are rewritten at the data, API and content levels, evolving into sky-level development efficiency.

"Just like ChatGPT is no longer a complex menu and a bunch of functions, you can do a lot of things with just a dialog box. And it is iterating behind it every day, but you You can’t feel any changes in its interface. In the past, the software was upgraded at the interface level, but in the future it will be upgraded at the data level."

The reason why Fourth Paradigm proposed the AIGS strategy is It is precisely based on this prediction: large models can bring about improvements in user experience and development efficiency, so they will definitely form a leap in the software industry. Large models are the new productivity. Taking large model infrastructure as the pilot project to transform the entire software industry, the business value and business model of the entire industry will leap forward.

On the day of the Open Day, 4Paradigm unplugged the network cable on site and demonstrated its ability to understand pictures, script writing, drawing, and writing code of Shishuo large models.

Realize AIGS: Copilot thinking chain CoT capabilities, forming a new paradigm of domain software interaction

In Dai Wenyuan’s view, to achieve AIGS, large models do not necessarily need to be knowledgeable , a generalist who is a decathlon champion, and more importantly, the model has the capabilities of Copilot (copilot) and CoT (chain of thoughts, multi-step reasoning).

"To transform enterprise software, large models cannot only have language capabilities. Shishuo 2.0 has added multimodality and Copilot, because the data in many enterprise software is multimodal. , and Copilot can translate human instructions into which API to call in the background."

It is reported that when Shishuo 2.0 is launched, employees can use voice, images, forms, videos, etc. The modal method initiates inquiries or instructions to Shishuo. After Shishuo understands it, the networked enterprise software calls up the relevant functions and outputs the answer in the required form.

However, employees will also face complex tasks when using enterprise software, which require people to perform functions one by one in sequence.

Dai Wenyuan explained specifically using image processing software: If you want to adjust brightness and contrast, these are the functions. But if the character P is made thinner, it cannot be realized by a single function. Behind this, people need to execute each function of the software based on experience to achieve it.

"When a person receives a complex job, he will reason out the subtasks to be performed step by step in his brain, and then execute them step by step. If it is replaced by a machine, if the same job , the machine has seen enough people in the past (accumulated data) to complete this work through steps one, two, three, and four. It can already summarize this routine and form a thinking chain."

Therefore, Shishuo 3.0 emphasizes that Copilot plus thinking chain CoT has stronger reasoning ability. After learning a large amount of data and "strategies", it can form intermediate logical reasoning steps, thereby achieving disassembly Divide and perform complex tasks.

In summary, the fourth paradigm summarizes the path of AIGS into three stages:

In the first stage, Copilot mobilizes different information , data, application, as an assistant to complete the user's instructions. It is equivalent to having a commander in all enterprise-level software systems. The commander listens to the user's instructions, such as "brighten the photo by 20%."

In the second stage, Copilot is based on the "knowledge base" of enterprise rules. AI can refer to the rules to do complex work, further enriching the capabilities of the "dialog box". For example, after AI queries the "portrait beautification" knowledge base, it can perform steps to repair the photo to look good.

The third stage, Copilot CoT (Thinking Chain). The usage behavior of the software system will eventually be learned by the large model, forming a thinking chain for AI in this field, which means that complex instructions such as "processing photos to make them look better" can be completed automatically by AI according to the steps.

The AIGS strategy of Fourth Paradigm refers to transforming enterprise software into a new interaction paradigm based on the Copilot COT capabilities behind the big model, and continuously learning on new interactions. The process of using software forms the "thinking chain" of domain software.

At the launch site of Shishuo 3.0, customer representatives from finance, aviation manufacturing, medical and other industries attended and gave a live demo of the product. In just two months since the emergence of ChatGPT, many enterprises and partners have already carried out in-depth cooperation with Paradigm in generative AI. As a new development platform based on large models, Shishuo looks forward to working with more partners and enterprise customers to explore opportunities to reconstruct enterprise software with generative AI, and jointly improve the usability and productivity of enterprise software.

The above is the detailed content of Targeting AIGS: The Fourth Paradigm Release Model, Reconstructing Enterprise Software with Generative AI. For more information, please follow other related articles on the PHP Chinese website!