According to the media training camp, "the cumulative user scale of Wenxin Yiyan has reached 300 million, and the number of daily calls has also reached 500 million. On June 28, Wang Haifeng, Chief Technology Officer of Baidu and Director of the National Engineering Research Center for Deep Learning Technology and Application, announced at the WAVE SUMMIT Deep Learning Developer Conference 2024 that he officially released the latest technologies such as Wenxin Large Model 4.0 Turbo and PaddlePaddle Frame 3.0, disclosing the latest achievements of the PaddlePaddle Wenxin ecosystem.
A new perspective of general artificial intelligence: general technology and comprehensive capabilities
The goal of AI is to simulate, extend, and extend human intelligence. Wang Haifeng believes that large models bring dawn to general artificial intelligence, which can be interpreted from two perspectives, one is the versatility of artificial intelligence technology, and the other is the comprehensiveness of capabilities.
Artificial intelligence technology has gone through decades of development, from the early manual writing of rules, to later statistical machine learning, which can automatically learn from data, but different algorithms are needed to solve problems in different scenarios, and the evolution to the era of deep learning has greatly enhanced the versatility of algorithms, and a set of architecture and technologies of deep neural networks can solve various problems, and then to the era of large models, in addition to algorithms, models have become more general and unified.
In terms of the versatility of technology, the versatility of large models in solving different tasks, languages, modalities, and scenarios is getting better and better. Taking natural language processing as an example, there were many sub-directions such as word segmentation, syntactic analysis, semantic matching, machine translation, question answering, dialogue, etc., but now a large language model can solve most of the tasks. In terms of language, the large model can not only solve the problem of single language, but also cross languages, not only learning the natural language of human beings, but also learning the formal language defined by artificially, building a bridge from thinking to execution. At the same time, the large model can also realize multi-modal unified modeling, which can widely empower applications in all walks of life. In general, AI technology is becoming more and more versatile.
In terms of comprehensiveness of capabilities, comprehension, generation, logic, and memory are the four basic capabilities of artificial intelligence, and the typical abilities of other artificial intelligence, whether it is creation, problem solving, coding, planning, decision-making, etc., are basically the comprehensive application of these four basic capabilities. Taking the operation activity system based on the Wenxin model as an example, the model first understands the user needs, designs the system, and then plans the relevant tasks to execute, and then generates the code according to the plan, and gradually realizes the system construction, including the database, back-end functions, front-end pages, etc., and then deploys the service. The process from system development to service operation of the large model fully reflects the four basic capabilities of the large model: understanding, generation, logic, and memory, as well as the comprehensive application of these four capabilities. The stronger these four capabilities are, the closer they are to general artificial intelligence.
Wenxin Model 4.0 Turbo was released, and the number of users of Wenxin Yiyan exceeded 300 million
At the conference, Wang Haifeng released the Wenxin Model 4.0 Turbo, and the web version, APP, and API were successively opened to users. It is understood that enterprise users can use API services by logging in to the Baidu Intelligent Cloud Qianfan large model platform immediately.
Taking the user using the Wenxin model to design a questionnaire as an example, the Wenxin model 4.0 Turbo is faster and has better results.
According to public information, Baidu has begun to comprehensively deploy artificial intelligence since 2010, launched Wenxin Model 1.0 in March 2019, and continued to iteratively upgrade, released Wenxin Model 4.0 in October last year, and released Wenxin Model 4.0 tool version in April this year. Based on the PaddlePaddle platform, the Wenxin model integrates learning from trillions of data and hundreds of billions of knowledge, and has core technologies such as knowledge enhancement, retrieval enhancement and dialogue enhancement.
On the basis of a powerful basic model, Baidu has further developed an agent mechanism, including understanding, planning, reflection and evolution, which can be reliably executed, self-evolved, and to a certain extent, the thinking process is white-boxed, so that machines can think and act like humans, and can call tools to complete complex tasks autonomously, and continue to learn in the environment to achieve autonomous evolution. At present, the cumulative user scale of Wenxin Yiyan has reached 300 million, and the number of daily calls is 500 million.
The conference also released the first agricultural intelligent twin, "Farmer Academician Intelligent Twin", jointly created by Academician Zhu Youyong of the Chinese Academy of Engineering and his team and Baidu. It is based on the Wenxin intelligent twins platform, and learns the research results of Academician Zhu Youyong and related agricultural knowledge, which can answer problems in production and life for farmers, and promote science and technology to help farmers benefit farmers. Ding Xingxing, Secretary of the Party Committee of Zhutang Township, Lancang Lahu Autonomous County, and Li Fu, a villager of Yunshan Village, were invited to the conference site to share their feelings about using the "Farmer Academician Agent".
PaddlePaddle Framework 3.0 was released, and the number of PaddlePaddle Wenxin ecological developers reached 14.65 million
The continuous and rapid evolution of Wenxin's large model is due to Baidu's full-stack layout of chips, frameworks, models and applications, especially the joint optimization of PaddlePaddle Deep Learning Platform and Wenxin. Wang Haifeng released a new generation of PaddlePaddle Framework, PaddlePaddle Frame 3.0, which has core technologies such as unified automatic parallel dynamics and statics, automatic compiler optimization, multi-hardware adaptation of large models, and integration of large model training and pushing, which supports large models with better effects and better performance. Up to now, PaddlePaddle Wenxin Ecosystem has gathered 14.65 million developers, served 370,000 enterprises and institutions, and created 950,000 models.
Wang Haifeng finally said that throughout the first three industrial revolutions experienced by mankind, its core driving force mechanical technology, electrical technology and information technology have strong versatility, will be applied to all walks of life, when they show the characteristics of standardization, automation and modularization of industrial production, the core technology will enter the stage of industrial production. Artificial intelligence is based on deep learning and large model engineering platforms, including algorithms, data, models, tools, etc., and has a very strong versatility, and has the characteristics of standardization, modularization and automation, which promotes artificial intelligence to enter the stage of industrial production, and general artificial intelligence will accelerate its arrival.