The material in this article comes from the Internet, if it is inconsistent with the actual situation or there is infringement, please contact to delete.
(Report Producer/Analyst: Liu Zejing, West China Securities)
1. Recent hot content
1.1. A small step for AIGC, a big step for AI
ChatGPT continues to be popular: ChatGPT is a chatbot software: its full name is Chat Generative Pre-Trained Transformer. It is a chatbot launched by OpenAI in November 2022. This software is simple and quick to use. You can achieve this simply by sending a request to ChatGPT. Article creation, code creation, Q&A and other functions.
ChatGPT continues to make a record: just 5 days after launch, ChatGPT already has more than 1 million users. According to NetEase, ChatGPT has only been online for two months, and as of January 2023, monthly active users have exceeded 100 million, setting a record for the fastest growing consumer app. According to Sensor Tower, it took TikTok nine months to reach 100 million users, while Instagram took two and a half years.
AIGC is not just ChatGPT:AIGC (AI generated content), that is, artificial intelligence automatically generates content, which can give full play to its technical advantages in creativity, expression, iteration, communication and personalization to create a new generation of digital content and interactive forms. ChatGPT is just one of the commercialization cases of AIGC AI writing. Other commercialized directions include AI mapping and AI underlying modeling. AIGC's future business direction also includes AI-generated video and animation.
AIGC is also constantly blowing up the market: Stability AI's open-source model, Stable Diffusion, can automatically generate images, known as Vincent diagrams, based on text descriptions entered by users. AIGC models that generate images, such as Stability Diffusion, DALL-E 2, and Midjourney, have exploded the field of AI painting. The popularity of AI painting marks the penetration of artificial intelligence into the field of art. In October 2022, Stability AI received about $100 million in financing, valued at $1 billion, becoming one of the unicorns.
AIGC is also considered to be a new way of content production after UGC, PGC/UCC: UGC is user-generated content, which originated in the Web 2.0 era, such as blogs, video sharing, Youtube and other application solutions; PGC/UGC refers to professionally produced content and professionally generated content, such as video monetization platform Youku, Tudou platform programs "Runaway Comics", "Never Expected" or user-generated content, such as Douyin and Kuaishou.
We see AIGC as disruptive in terms of setup costs and several advantages in terms of cost reduction and efficiency. It is expected to solve the current problem of uneven quality of PGC/UGC creation or reduce the spread of harmful content, while stimulating creativity and improving content. Diversity, in addition to the price, OpenAI's most expensive AIGC language model is da Vinci 750 words at $0.02 per piece, and the graphical AIGC model at only $0.020 per piece.
A small step for AIGC, a big step for AI: The emergence of AIGC has enabled AI to make a leap from perceiving and understanding the world to generating and creating the world. Before the rise of AIGC, artificial intelligence was more inclined to analysis, such as AIGC, the hottest personalized recommendation algorithm. The emergence of artificial intelligence is more "anthropomorphic", which can autonomously create text, images, videos and 3D interactive content, and trigger new scientific discoveries to create new value and meaning. As a result, Gartner listed generative AI as a 5-year trend for 2022. As one of the influential technologies, MIT has also listed AI synthetic data as one of the top ten breakthrough technologies in 2022.
AIGC is a powerful engine for data and content production: AIGC is moving towards reducing costs and increasing efficiency. We believe that as AIGC's technology matures, content generation can cover everything from social media to gaming, advertising to architecture, coding to graphic design, product design to legal, marketing to sales, all of which can be influenced or changed by AIGC.
2. Turning point, the pattern changes
We believe that the integration of artificial intelligence technologies such as computing power, algorithms, multi-model training, and multi-modality has greatly promoted the outbreak of AIGC.
1) Basic computing power: the nature of artificial intelligence and the massive calculation of data. Compared to AI algorithms, data is the most important. As a power source for accelerating computing, the importance of computing power is self-evident.
According to the algorithm steps of machine learning, it can be divided into two parts: training and inference. The training stage requires an extremely large amount of data input to support complex neural network models. During the training process, the amount of computation increases due to the complex structure of the neural network and the massive amount of training data. Enormous, so the computing power and efficiency (energy consumption) of the processor are extremely demanding. 2) Algorithm model: The Transformer algorithm is a deep learning model that uses a self-attention mechanism. The mechanism can assign different weights based on the importance of each part of the input data. We are now familiar with ChatGPT and AI maps. It is built on the Transformer algorithm;
3) Pre-training model: The pre-training model has triggered a qualitative change in the technical capability of AIGC. Before the emergence of this model, there were problems such as high threshold, low training cost, easy content generation, and low quality.
Internally, AI pre-training models and AI pre-models can play an important role in achieving multi-tasking, multi-language, and multi-mode. Models such as Google's LaMDA and PaLM, and Open AI's GPT series.
4) Multimodality: Increases the diversity of AIGC, making predictive models more versatile and diverse. For example, the Open AI team's CLIP model can associate text and images, such as associating text "dogs" with images, and the association function is very rich.
AIGC's application ecology and content consumption market are gradually prospering: AIGC has a good performance in learning general knowledge and understanding generalization, and has the following characteristics in content generation.
1) Automatically generate content: Large language and image AI models can be used to automatically generate content such as articles, blogs, social media, and posts.
2) Improve content quality: We believe that AI-generated content is of higher quality because AI models can learn from large amounts of data and the information is accurate. For example, DALL· The effect of E is close to the level of ordinary painters.
3) Increase content diversity: AIGC models can generate many types of content, including text, images, audio and video, 3D content, etc., which can be combined with expertise to create more diverse and interesting content, which is expected to appeal to a wider range of people.
4) Low content production cost: Based on AIGC, the content production cost is significantly reduced, the efficiency is significantly improved, and content with unique value and independent perspective can be created.
5) Personalized content generation can be achieved: AI models can generate personalized content based on individual user preferences. For example, Stable Diffusion's two-dimensional art generation tool, Novel-AI, can meet the preferences and content needs of two-dimensional niche groups. AIGC has a broad future, and there are already many implementation scenarios: such as the current popular ChatGPT. ChatGPT interacts in the form of a conversation on a web browser, which can meet the basic functions of human dialogue, answering follow-up questions, admitting mistakes, and challenging incorrect statements. Ask, we think that the coding function of ChatGPT and the function of AI question answering system have been greatly improved, which can replace the search engine to a certain extent.
In addition, digital humans are also one of the application scenarios of AIGC: digital humans are a new mode of intelligent interaction between digital agents, and there are currently many applications, including virtual NPC roles, user virtualization in metaverse applications, etc.
AIGC has greatly improved the energy efficiency of digital human production. Users can provide images and videos, and AIGC can generate realistic digital humans that are ephemeral, inexpensive, and customizable. At the same time, 3D digital human modeling is beginning to take shape.
In addition, AIGC supports the identification, perception and analysis functions of AI digital human multimodal interactive decision-making, making it more humane.
AIGC has a strong generational capability and can radiate to other fields: such as medical, auxiliary design and other industries. For example, AIGC speech generation can help patients "speak." For example, British robotics scientist Peter Scott-Morgan transformed himself into the world's first "human robot" through multiple surgeries after being diagnosed with ALS in 2017.
HE CONTACTED THE MANUFACTURER OF THE CEREPRO VOICE SOLUTION, FORMED A CORPUS AND PHRASE DATABASE CONTAINING DIFFERENT EMOTIONS, AND FINALLY COMPLETED THE PUBLIC SPEAKING THROUGH VIRTUAL 3D CHARACTERS. AIGC can also assist in the restoration of cultural monuments. For example, Baidu uses AIGC maps to "restore" the "Fuchun Mountain House Map".
AIGC doesn't stop there, it has a lot of promise: AIGC promises to be a base for the 3D internet, including 3D modeling, scenes, and character production, while inspiring creators.
Take 2018's Red Dead Redemption 2 as an example, a 60-square-kilometer virtual stage took more than 600 art modelers eight years to complete. In 2022, The Matrix: Awakening will feature AIGC technology, and virtual cities will contain a total of 7 million instance assets. This includes 7,000 buildings, 38,000 drivable vehicles and more than 260 kilometers of roads, significantly increasing production efficiency.
In addition, AIGC is expected to generate 3D images, but it is still in the exploratory stage, and the speed and quality of generation still need to be improved.
AIGC is expected to drive the metaverse as a productive force: one of the biggest features of the metaverse is its scalable development potential, and the biggest question is how to structure the content. 3D modeling in the metaverse cannot meet the needs of relying solely on manual design and development. We believe that with the gradual maturity of AIGC technology, it will be possible for the metaverse to meet the needs of massive user content in a low-cost and efficient way.
AIGC industrialization process accelerates, model-as-a-service business model opens: At present, the prototype of AIGC industrial system has basically emerged, which is divided into three layers: upper, middle and lower.
The first layer is the base layer: it is an infrastructure layer built on pre-model AIGC technology. At present, the companies settled in are leading technology companies such as OPEN AI and Stability.
The second layer is the middle layer: vertical, scenario-based models and application tools. By using the base layer model to generate applications, the application layer can quickly generate scenario-based, customized, and personalized applications based on the foundation. models and programs, such as Novel-AI;
The third layer is the application layer: it provides content generation services such as text, images, audio and video for C-end users.
OpenAI's business model is API interface loading: customers can build applications through OpenAI's powerful AI models, such as access to GPT-3 to perform various natural language tasks, Codex to translate natural language into code, and DALL to create and edit raw images. · E.
Companies charge according to AI models and the different needs of different projects. AI imaging systems are priced based on different image resolutions; The AI language text system is billed at the character unit price, and the adjustment model and embedding model are billed at the text unit price, and the price is different according to the Call model.
Currently, more than 3 million people use DALL· E, generating more than 4 million images per day.
The API is fast, flexible, and scalable. OpenAI's business model belongs to the underlying open and standardized SAAS service model: companies provide products such as automatic graph models or language model generation by providing relevant AIGC code. Users access and pay platform fees through API interfaces to obtain relevant images and languages. FOR CODE ALIGNMENT SERVICES, OPENAI MAY CHARGE SUBSCRIPTION-BASED SAAS FEES. We believe this business model is very user-friendly.
At present, the mainland AIGC business is still in its early stages. Some companies that generate text, images, etc. are still in their early stages and mainly provide ancillary services for content generation. Some enterprises have entered the AIGC development stage, but they are still in the free trial and traffic acquisition stage. WE BELIEVE THAT THE SAAS MODEL OF RELATED COMPANIES IS EXPECTED TO TAKE SHAPE IN THE FUTURE.
We believe that the birth of AIGC will have a revolutionary impact and hopefully empower a wide range of industries. As AI writing, AI drawing, AI underlying modeling, AI-generated video, and animation technologies mature, AI is poised to enter a new era, bringing unprecedented blue oceans while impacting existing entertainment, media, news, modeling, and other applications.
We believe that AIGC's beneficiary producers fall into three categories:
1) AI processor manufacturers;
2) Producers of AI commercial algorithm commercial landing;
3) Application manufacturers with AIGC related technical reserves.
Manufacturers with self-developed AI processors can provide computing power support for AIGC's neural networks: the nature of artificial intelligence and the massive computing of data, compared with AI algorithms, data is the top priority.
As a power source for accelerating computing, the importance of computing power is self-evident. AI processor chips can support deep neural network learning and accelerated computing, with exponential performance improvement and extremely low power consumption compared to GPUs and CPUs. Therefore, AI chips will play a big role.
Vendors that have implemented commercial AI algorithms have a leading position in related algorithms: Leading AI algorithm vendors have first-mover advantages in natural language processing, machine vision, and data annotation.
From the algorithmic point of view, data labeling is a key step in the generation of AIGC algorithm. In natural language processing, machine vision, etc., AIGC has a far-reaching impact on the application of this direction, such as virtual human-natural dialogue between natural person, AI mapping and AI underlying modeling, etc., with the further maturity of technology, AIGC will have a revolutionary impact on the application of this direction.
Application manufacturers with AIGC-related technical reserves are expected to open up a large number of markets: related entertainment, media, news, games, search engines and other manufacturers have massive demand for text creation, image generation, video generation and so on. With the gradual maturity of AIGC, the related AI algorithms continue to mature. By improving and combining with related applications, we believe that while reducing costs and increasing efficiency, relevant producers are expected to improve the quality of their creations, reduce problems such as the spread of harmful content, stimulate creativity, and increase content diversity. AIGC is expected to greatly promote the commercialization of related manufacturers and open up huge space.
3. Introduction of industry-related enterprises
We believe the emergence of AIGC will have a revolutionary impact and hopefully empower a wide range of industries.
We've put together three roadmaps:
1) Manufacturers with self-developed AI processors: the advantage objects are Cambrian, SenseTime, and Haiguang Information;
2) AI algorithm commercial landing manufacturers: iFLYTEK, Torsi, other beneficiary companies are: Hanwang Technology, Haitian Ruixiang, Hongsoft Technology, Yuncong Technology, Greenfield; 3) AIGC related technology reserve application producers: the beneficiaries are Wanxing Technology, Chinese Online, China Literature Group, Kunlun International, and Visual China.
iFLYTEK: Global leader in the field of natural language processing
iFLYTEK officially released the "iFLYTEK Super Intelligence 2030 Plan" in early 2022, aiming to move towards the long-term vision of becoming a "global leader in the artificial intelligence industry". The plan is the company's core strategy, the purpose of which is to build a complex system based on cognitive-based human-machine collaboration and self-evolution, that is, to let the robot's senses surpass humans, have the ability to develop independently, and create a complex intelligent system that can continue to develop autonomously, helping robots enter thousands of households.
The program is divided into three phases of milestones:
Torsi: A leader in semantic intelligence and a comprehensive service provider of data meta-market
Founded in 1993, the company's main business includes artificial intelligence products and services, big data products and services, and data security products and services. It has become a comprehensive service provider of the data meta market.
In the field of artificial intelligence, the company focuses on the research and development and industrialization of semantic intelligence technology. Compared with speech recognition and image recognition, semantic understanding is more complex and difficult, requiring "data" and "knowledge" to operate in two rounds. In the field of big data, the company has the technical capabilities of data collection, data management, data retrieval, data analysis and mining throughout the life cycle.
In terms of data security, the company's subsidiary, Tianxing Network Security, provides products and services for the secure transmission and exchange of data.
Hanvon Technology: Artificial Intelligence Leader
Hanvon Technology is a leader in the field of artificial intelligence: founded in 1998, it is a provider of facial recognition, big data and intelligent interaction technologies, products and services. Based on core technology, the company meets market demand and has formed a series of software and hardware products with identification technology as the core and for various market segments. It has electronic point pen, Hanwang electronic paper book, Hanwang pen, text king, business card, drawing board and other general products.
The most important business of the company is: 1. Pen intelligent interactive business. The company's technology in pen intelligent interaction is mainly divided into two categories: electromagnetic touch and capacitive touch; 2. Text big data business includes smart justice, smart archives, smart humanities, smart education, smart finance, etc.; 3. Facial and biometric identification; 4. Related AI terminals.
Cloudwalk Technology: Empowering the human-machine collaborative ecosystem
Cloudwalk Technology is an artificial intelligence company focusing on providing human-machine operating systems and industry solutions. It is committed to promoting the industrialization of artificial intelligence and the transformation and upgrading of various industries.
The company's business is divided into two parts: 1. Human-machine operating system: through the underlying AI business process engine and decision support system, optimize the human-computer interaction experience, improve the efficiency of business process execution, human-machine decision-making functions include human-computer collaboration and overall operating system, "Qingzhou platform" service platform and AI application supporting SaaS services.
2. Artificial intelligence solutions, including comprehensive solutions in four major fields: smart finance, smart governance, smart travel, and smart business.
————————————————————
The copyright of the report belongs to the original author, do not make any investment advice! If there is any violation, please delete it by private message, thank you!
【Original Exhibition Think Tank】Report Center Report Selection-Original Exhibition Think Tank|A useful knowledge platform for 300 million people
Disclaimer: The above content and materials are derived from the Internet, and the author of this article does not intend to target or allude to any real country, political system, organization, race, or individual. The above does not mean that the author of this article endorses the laws, rules, opinions, behaviors in the article and is responsible for the authenticity of the relevant information. The author of this article is not responsible for any problems arising above or related to the above and does not assume any direct or indirect legal liability.
If the content of the article involves the content of the work, copyright images, infringement, rumors or other problems, please contact to delete. Finally, if you have any different ideas about this event, welcome to leave a message in the comment area to discuss!