Robin Lee, founder of Baidu, said in a speech at the 2024 China 5G+Industrial Internet Conference on the 19th that the "Seconds Da", a code free tool for multi-agent collaborative applications, had been released for less than three days, and more than 5000 enterprises had queued up to apply for testing. In addition, after the release of the L4 end-to-end autonomous driving model, the Apollo 10.0 version of the autonomous driving open platform, which is equipped with the Baidu model, will also be released to users worldwide.
Picture: Robin Lee delivers a speech
It is reported that as of early November, the daily average usage of Baidu Wenxin's large model has reached 1.5 billion, an increase of 7.5 times compared to the 200 million disclosed in May, and an increase of about 30 times compared to the first disclosure of 50 million times a year ago, especially in the past six months where the growth rate has been very fast. Robin Lee said: "The high volume and fast growth of large model calls indicate that more and more applications are using Wenxin large model."
As for the reason behind the high growth rate of basic model calls in the past half year, Robin Lee believes that the main reason is the ability of retrieval enhancement (RAG)
However, in the past 24 months, big models have largely eliminated illusions. "Today, RAG at the textual level has done a great job, making big models usable and trustworthy. Multimodal technologies such as images need to be practical, accurate, and controllable, thus expanding the application space of AI," he said.
It is reported that Baidu has devoted a lot of effort to solving the "illusion" problem in image generation by developing iRAG (Image Based Retrieval Augmented Generation), a retrieval enhanced technology that combines Baidu's billion level image resources with powerful basic model capabilities to generate various hyper realistic images. "Now using Wenxin multimodal model to generate can remove illusion and so-called 'AI flavor', and the generated pictures look more realistic and retain accuracy." Robin Lee believes that "in the future, multimodal retrieval enhancement will also have rapid development, so that multimodal large models will enter a more practical stage."
In addition to retrieval enhancement technology, another important development direction of large models is intelligent agents, and the ultimate form of intelligent agents is multi-agent collaboration. On November 12th, Baidu released the codeless tool "Miaoda" at the Baidu World 2024 Conference, which is a multi-agent collaboration application. It is reported that unlike other auxiliary code generation tools on the market, "Miaoda" does not require people to understand the code, allowing non programmers to possess the abilities of programmers, covering features such as no code programming, multi-agent collaboration, and multi tool calling. With only natural language, various applications can be built.
Robin Lee said that just three days after the release of "Seconds Da", more than 5000 enterprises queued up to apply for the test. "There are about 28 million programmers in the world now, but there are 8 billion people in the world. Most people can't understand a line of code and can't solve problems with programming methods. When everyone has the ability of programmers, it is a great release for the productivity of the whole society." Robin Lee said.
In addition, after the implementation of large-scale models in various fields such as manufacturing, energy, finance, and public services, tangible results have been achieved in both cost reduction and efficiency improvement, while also bringing new opportunities for industrial innovation. It is reported that Baidu has been laying out autonomous driving since 2013. In May of this year, Baidu first released the L4 level end-to-end autonomous driving model Apollo ADFM, which can balance the safety and generalization of technology. The Apollo 10.0 version, an open platform for autonomous driving equipped with this large model, will soon be released to users worldwide.
"AI is a new industrial revolution. Today, many discussions about the big model and generative AI are comparing it with PC Internet and mobile Internet. But we should refer more to the development process of the steam engine revolution, the electric power revolution and the information revolution, from which we can learn how a country, a company, or an individual can benefit as much as possible from the development process, and how to avoid possible negative effects. In this way, we can really make good use of the new industrial revolution, make good use of the big model, empower all walks of life, improve social production efficiency, and make better contributions to people's better life." Robin Lee said.
