Overviewing the evolutionary history of Generative Artificial Intelligence (AIGC), people have attempted to make machines generate content, engage in dialogue, giving rise to the earliest Turing test standards. For many years, the development of generative AI has been lukewarm. It wasn't until last year when applications like Midjourney, Stable Diffusion, and others ignited public enthusiasm for generative art, followed by the November emergence of ChatGPT, causing a phenomenal craze, bringing generative AI into the view of millions of users. This article combines international corporate practical experience, focusing on two perspectives: the construction mode of large models' ecology and the approach to industry application paths, outlining the industry application map of generative AI, sharing our preliminary observations and thoughts.

I. Ecology of Large Models

The emerging ecology of Model as a Service (MaaS) is accelerating.

Considering the current industry's three-layer structure of "infrastructure layer - model layer (MaaS) - application layer," we anticipate the potential development of a new Model as a Service (MaaS) ecosystem in the future. In descending order, the three-layer structure consists of:

• Infrastructure Layer: Algorithmic infrastructure such as GPU chips, AI chips, supercomputers, along with software like machine learning frameworks, and cloud operating systems.
• Model Layer (MaaS): Encompasses general AI large models and industry-specific models generated from training within vertical domains (intermediate layer). This involves the rapid fine-tuning or embedding of smaller models based on pre-trained AI large models, enabling the development of contextual, customized, and personalized small to medium-sized models. These models facilitate industrial pipeline deployment across different industries, vertical domains, and functional scenarios, exhibiting advantages in on-demand usage and efficient economics.
• Application Layer: Presents new functionalities, products, services, and applications for end-users through AIGC (AI-generated content) technology. Similar to the previous "Internet +," "AIGC+" or "AI+" is anticipated to deeply integrate with various industries, continuously introducing new application forms. In the future, existing apps could undergo a transformation using large models, while new AI-native apps are likely to emerge, resulting in the birth of more unicorns and even industry giants.

The deep layout of large models emphasizing both generality and verticality.

In the application layout of AIGC technology, there is a focus on both generality and verticality.

"Generality" refers to the lateral capability, applicable across various industries. Specifically, it includes aspects such as semantic multi-turn dialogue, knowledge base construction, intelligent search, enterprise-level RPA, multimodal content generation, and code generation:

• Semantic multi-turn dialogue: AIGC technology excels in semantic multi-turn conversations, especially in high-value sales and private communication scenarios. For instance, in live broadcast scripts, AIGC rapidly learns and masters best practices, such as various chatbots and IQ for sales.
• Knowledge base construction: AIGC technology has progressed from summarizing simple data to assisting users in comprehension, generating personalized databases, and introducing new search functionalities based on efficient information summarization. This enables AIGC to be applicable not only in enterprise-level AI but also as a personal assistant (e.g., Mem).
• Intelligent search: AIGC has shown significant achievements in the field of intelligent searches, such as New Bing and Perplexity. Systems that accumulate knowledge bases can leverage AIGC to implement semantic search functionalities.
• Enterprise-level RPA (Robotic Process Automation): Combining LLM (Large Language Models) technology with RPA, AIGC plays a crucial role in enterprise-level applications. Typical applications include Microsoft's Copilot and Salesforce in the CRM domain, tightly integrating with industry scenarios to empower users.
• Multimodal content generation: In fields like intelligent writing assistants, visual content creation for advertisements, intelligent NPCs, etc., AIGC technology has introduced innovative applications. Apart from text, it can generate images, videos, and even 3D digital content.
• Code generation: With code being a more standardized form of text, AIGC holds substantial potential in code generation. Tools like GitHub Copilot have shown impressive performance in this area. More code generation tools are emerging, expected to significantly enhance the productivity of developers and provide more convenient tools for non-technical individuals, substantially reducing the barriers to programming.

"Verticality" refers to the longitudinal capability where AIGC technology evolves into an industry expert by learning various industry-specific know-hows, significantly accelerating the digital transformation of various industries.

By combining international corporate investment and financing information, here are some typical examples from the medical, financial, retail, and manufacturing sectors:

• Medical (Zebra Medical Vision, Aidoc, etc., utilizing AGI for medical image analysis).
• Finance (Bloomberg introducing Terminal AI large models for financial intelligence services).
• Retail (Companies like Stitch Fix employing generative AI for personalized shopping experiences, inventory management, and demand forecasting).
• Manufacturing (Companies like General Electric leveraging generative AI to optimize production processes, predictive maintenance, and supply chain management).

Five typical modes of collaboration within the large model ecosystem are gradually unfolding, and we have outlined the primary five.

Among these, API calls and plugins are the main methods through which most companies access the capabilities of large models.

• API Calls: The primary method used by most companies to leverage large models.

In the current technological landscape, API calls are widely employed in knowledge bases and customer service domains.

Companies utilize ChatGPT's open Fine Tuning API by uploading their proprietary knowledge base (including product documents, FAQs, historical customer service conversations, etc.) into ChatGPT, thereby establishing a private model.

It's noteworthy that even with this approach, startup companies cannot own the entirety of this model.

Through this method, users of ChatGPT can possess an exclusive Chat Bot. This bot not only possesses conventional conversational functionalities but also utilizes the company's proprietary, sometimes undisclosed training data to provide highly targeted services.

• Plugin Mode: Plugins herald the potential for large models to become a new OS.

Apart from API calls, on March 24, 2023, OpenAI made a significant announcement: ChatGPT now supports integration with third-party plugins and immediately launched 11 plugins.

Through these plugins, users can use ChatGPT to make purchases, book hotels, flights, access specialized data searches, and more. This significantly enhances ChatGPT's productivity and brings forth numerous possibilities for its development.

This plugin mode offers diverse development prospects for potential plugin applications and their impact.

In summary, whether functioning as an operating system or an application store, the turning point for AI has arrived.

II. 6 Categories of Entry Paths in Industry Applications

When categorized by user type, Generative AI (AIGC) presents two paths in the C-end and B-end markets. The C-end has reached a critical point of usability and even excellent user experience, while the B-end is set to expand from high-value pilot fields towards the Model-as-a-Service (MaaS) ecosystem.

• Critical Usability Achieved in the C-end Market

  C-end applications encompass the next-generation of efficiency tools, new forms and methods for future gaming, and open up new spaces for cross-domain entities such as digital beings, metaverse, and robotics. Additionally, AIGC has sparked a significant outbreak in the content creation domain.

• B-End Market: Expanding from High-Value Pilot Fields to the MaaS Ecosystem
• Efficiency-driven Approach of Generative AI: Enhancing writing and productivity. For example, applications in meetings, recruitment, various office assistants (document assistants, programming assistants), and similar scenarios.
• Initiating in High-Value Fields such Marketing, Financial Education, etc.

  In the financial sector, AIGC applications are exploring areas like intelligent customer service, robo-advisors, merchant onboarding, fraud detection, intelligent marketing, etc. However, significant challenges still exist in algorithmic risk, privacy protection, information security, among others.

• Knowledge-Intensive Fields Hold Enormous Potential for Generative AI Applications

  In the medical field, GPT-4, as a medical AI chatbot, potentially aids in medical records, medical knowledge, and medical consultations. It can assist healthcare professionals in enhancing work efficiency and patient consultation experience. However, limitations and potential risks also exist.

  In these domains, AIGC technology continuously expands application scenarios, providing unprecedented efficiency enhancements and innovative opportunities across various industries.

However, it's imperative to note the challenges and risks, ensuring that while the technology realizes its potential, it also safeguards user interests and security.

When one person can function as an entire team, it could potentially give birth to B-end products with To C (business-to-consumer) experiences. With the acceleration of industrial transformation, profound changes are expected at the organizational level in the future. The model of individuals and small teams might become a significant form, further blurring the boundaries between To B and To C software...

Of course, this is just our preliminary thinking. Under the influence of generative AI technology, the future is bound to continue overturning our understanding, continuously pushing the boundaries of imagination, and bringing forth a new revolution in productivity. Regardless, the future is already upon us.