Source: Baidu AI
AIGC is exploding in the field of artificial intelligence. Artificial intelligence technology leads intergenerational change. Generative AI is applied in multiple scenarios. Among them, the generative large language model (LLM) has the advantages of versatility, multi-round dialogue understanding, and reasoning tasks. The performance in the movie has amazed the world.
What is the current progress of AIGC’s landing application? How will AIGC lead the transformation of business models in the future? As the chief architect of Baidu’s commercial R&D, Li Shuanglong shared the current status of AIGC development and his own observations on the prospects of AIGC’s commercial applications from a technical perspective while the first Baidu Commercial AI Technology Innovation Competition was in full swing.
Different fields have their own advantages, and the application of AIGC is developing rapidly
At present, AIGC technology has been applied in some scenarios and has created real value. Li Shuanglong analyzed that in the three main AI application directions of NLP, multimodality and digital human, my country’s technology has developed very rapidly, and phased results have been achieved.
In the field of NLP, the overall technology paradigm worldwide has entered the stage of artificial general intelligence (AGI), and generative large language models have become the mainstream research direction of NLP. This year, many domestic companies released generative products related to large language models. Baidu also opened a test invitation for a new generation of knowledge-enhanced large language models “Wen Xin Yi Yan” in March this year, and iterated four times within one month of launch. The inference performance has been improved by nearly 10 times. With the rapid iteration of technologies such as instruction following and optimization capabilities, and feedback data tuning, the effects of generative products have made leaps and bounds.
In the field of multimodality, with the introduction of new technologies such as large-scale visual language pre-training technology (VLP) and diffusion model (Diffusion Model), the detail controllability and generation quality of Vincentian diagrams have been greatly improved. The Wensheng graph scene application represented by Baidu Wenxin Yige is currently very mature in the framework of the generation algorithm, and the usability of the generated images is relatively high. With the advantages of massive application scenarios, the development of domestic technology in the field of multi-modality is expected to reach the world’s leading level.
In the field of digital human beings, 2D avatar digital human beings have entered the field of vision of marketers. They have a series of advantages such as low production cost, realistic effects, and high technical maturity. Scenarios have also begun to be applied, greatly reducing the cost of manpower input for such scenarios.
“Although we started late, we are making rapid progress.” As a judge of the Baidu Commercial AI Technology Innovation Competition, Li Shuanglong believes that the competition has given him more hope. In the process of following the competition and giving lectures at major universities, we can see that many universities have set up majors related to the field of artificial intelligence, and have begun to cultivate talents for China’s artificial intelligence from the university stage. Baidu Business also hopes to use this event to give college students the opportunity to get in touch with cutting-edge technologies and constantly explore and innovate, to use competitions as a substitute for training and competitions to promote learning, and to jointly explore the forefront of AIGC technology.
Connecting more field capabilities, AIGC’s landing application will benefit thousands of industries
The explosion of generative AI technology has made the world realize that the intelligence level of AI has surpassed the imagination of countless people, but dialogue and image generation are only the tip of the iceberg of AIGC application scenarios. Li Shuanglong believes that the large-scale application of AIGC is bound to connect more fields and benefit thousands of industries.
At present, Baidu Commercial has implemented AIGC technology on a large scale in scenarios such as creative production and service consulting, and achieved certain customer value and business gains.
In terms of intelligent production of text creativity, relying on the Baidu Wenxin large language model, Baidu has developed a tens of billions-level creative copy generation large model that enhances customer marketing knowledge and user feedback behavior. Only a small number of models need to be defined. Creative content matching the scene will be automatically generated. Based on this, the creative production method in the marketing field will be completely rewritten, and the copywriting efficiency and advertising conversion rate will be significantly improved.
In terms of video content production, Baidu Commercial has created an end-to-end generation solution for 2D digital pop-up videos. Based on the AIGC large model, it realizes the automatic generation of video scripts and the automatic driving of digital pop-up broadcasts, which solves the problem of traditional live-action advertising video production. The problems of cumbersome process and difficulty in scale can greatly reduce the threshold and cost of short video production, and realize the double improvement of production efficiency and marketing efficiency.
For scenarios where there is a strong demand for clue acquisition but insufficient online customer service capabilities, Baidu Commercial has developed its own clue robot, which uses tens of billions of dialogue generation models to replace the manually configured dialogue template system, and provides merchants with multi-scenario automatic switching consulting and reception services. Realize low-cost, scenario-based, and personalized dialogue thread optimization capabilities. At the same time, the clue robot also has the advantage of 7*24 hours a day, which can follow up the clues in the first time and improve the response efficiency.
In addition to production and service consulting, the development of AIGC technology is bound to bring revolutionary changes to the commercial marketing ecology.
First of all, generative AI can greatly improve the efficiency of creative production and sales consulting, lower the marketing threshold, and improve marketing efficiency. Under the generative-based optimization mode, the production of marketing creative content will change from a single person sailing in the vast ocean to being supported by a complete set of “aircraft carrier fleet”. Users can focus on improving product and service quality, eliminate the gap in delivery capabilities between customers of different sizes, and promote the healthy development of the business ecosystem.
Secondly, for the entire commercial retrieval system, the monetization efficiency will be further improved. Relying on the strong expressive ability of large models, on the one hand, the system can provide small and medium customers with stronger traffic matching and reach capabilities, and improve marketing coverage; on the other hand, the traditional multi-layer search funnel will be greatly simplified and end-to-end Integration and optimization will greatly improve retrieval efficiency.
Finally, AIGC technology has gradually realized the transformation of Baidu search user product interaction mode. Under the new product interaction form, the user’s demand expression will be fully personalized, promote deep conversion, and achieve a significant increase in marketing efficiency.
Looking forward to more young blood to join and ignite AIGC’s prairie fire together
With the continuous development of AIGC technology, the refinement of algorithms and applications is bound to bring a better user experience. However, as the most cutting-edge field of artificial intelligence, what is the direction of AIGC technology iteration in the future?
Li Shuanglong believes that the direction of AIGC technology iteration will focus on engineering, algorithms, data and applications.
At the engineering level, improving the training production, reasoning efficiency, and stability of large models will be one of the core issues that need to be overcome in the large-scale AIGC model; at the algorithm level, the understanding and reasoning capabilities of multiple rounds of dialogue, long input sequences and The ability to generate long texts, detail images, and end-to-end long videos requires further technological innovation; at the data level, data quality will be more important than scale, and large models in the future The task is bound to require the construction of a better high-quality data production ecology; at the application level, how to better carry out in-depth joint optimization from the production end to the application end based on the scenario-based user feedback signal will be the effect of the large-scale model scenario Key points for improvement.
The two topics of the Baidu Business AI Technology Innovation Competition, “Business Transformation Behavior Prediction” and “AIGC Reasoning Performance Optimization”, are the core tasks and technical difficulties of Baidu’s business monetization scenario. Based on this, Li Shuanglong invites the young elites in the industry, hoping that the contestants and Baidu Business will challenge and overcome these technical difficulties together, so that AIGC can penetrate deeper into the business operation, bring about opportunities for change to thousands of industries, and provide hundreds of millions Users and tens of thousands of customers bring value gains!
“But these are not all of AIGC,” said Li Shuanglong. At present, AIGC technology has made great progress, but there are still many problems and challenges. He looks forward to and welcomes more partners and young talents who are interested in the forefront of the AIGC field to join Baidu and use AI to change the future together.
