What has been the biggest change in the AI industry over the past 24 months? The big model has basically eliminated the illusion On November 12, Robin Lee, the founder of Baidu, delivered a speech entitled "Here Comes the Application" at the 2024 Baidu World Conference, releasing two AI technologies that enable applications: the retrieval enhanced Wensheng Graph technology (iRAG) and the codeless tool "Seconds Da". Wenxin iRAG is used to solve the illusion problem of large models in image generation, greatly improving practicality; The no code tool 'Suida' empowers everyone with the skills of a programmer and will create millions of 'super useful' applications.
As of early November, the daily average usage of Baidu Wenxin's large model has exceeded 1.5 billion, an increase of about 30 times compared to the first disclosure of 50 million times a year ago. Robin Lee stressed that agent is the most mainstream form of AI applications, which is about to usher in an explosion point. It will become a new carrier of content, information and services in the native era of AI.
On the same day, Robin Lee said that search enhancement was not Baidu's feature, but gradually became the consensus of the industry. Retrieval augmented generation (RAG) is a means of enhancing retrieval, which aims to eliminate illusions in large models. "The big model basically eliminates illusion, and its accuracy of answering questions has been greatly improved, making AI usable and trustworthy from" serious nonsense ". The big model is a probabilistic model, and the generated content is uncertain. After the adoption of RAG technology, the big model will use the information retrieved to guide the generation of text or answers, greatly improving the quality and accuracy of the content." Robin Lee said.
It is reported that Baidu has developed a technology called iRAG (Image Based RAG) to solve the illusion problem in image generation, which is a retrieval enhanced text generated image technology. Previously, the text generated image system, which was entirely based on the big language model, often produced images that were "fake at first glance". Baidu's iRAG combines the billion level image resources searched by Baidu with powerful basic modeling capabilities to generate various hyper realistic images, with overall effects far exceeding the native system of Wenshengtu. Simply put, the commercial value of iRAG lies in its absence of illusions, hyper realism, cost-effectiveness, and immediate feasibility Robin Lee said.
When the basic model capabilities are ready, we are about to usher in the shining moment of AI applications
"The basic model capability is ready, and we are about to usher in the shining moment of AI application stars." Robin Lee pointed out that there are two AI application directions: one is agent, and the other is industrial application. Intelligent agents are the most mainstream form of AI applications and are about to reach its tipping point. Building an intelligent agent is much like building a website in the PC era or creating a self media account in the mobile era. The difference is that an intelligent agent is more like a person, more intelligent, and more like your sales, customer service, or assistant. An intelligent agent may become a new carrier of content, information, and services in the AI native era
On the one hand, the threshold for creating intelligent agents is low enough, "he said. On Baidu's Wenxin intelligent agent platform, 11 year old elementary school students are already creating intelligent agents. And distribute it through search and other channels. On the other hand, the ceiling of intelligent agents is high enough to make very powerful applications. The collaboration of multiple intelligent agents can also solve more complex problems. At present, the Wenxin intelligent agent platform has attracted 150000 enterprises and 800000 developers to participate, covering a wide range of application scenarios.
The user of the no code tool 'Suida' can complete system setup through natural language interaction
"Today, we do have the conditions to enable people who can't understand a line of code to have the ability of programmers and the ability to make any idea come true quickly and cheaply. Of course, this requires tools, and this tool is" Seconds Da ". Robin Lee introduced that" Seconds Da "is a software that can realize any idea without writing code, a new software composed of large models and agents, including the ability to code free programming, multi-agent collaboration, and large-scale invocation of various tools. 'Zida' is different from any auxiliary code generation tool, it doesn't require you to understand the code at all
With the help of "secda", a person can complete the construction of a complete system through natural language interaction, and can also create various applications in any scenario. The complexity of the applications will continue to increase with the improvement of technology. Robin Lee said, "We will usher in an unprecedented era in which you can make money only by thinking. Once upon a time, you had an idea, and you had to implement it. Now as long as you have an idea, 'Seconds Da' can help you do it."
Finally, Robin Lee also said, "In different historical periods of human information technology change, applications appear differently: in the PC era, they are software and websites one by one; in the mobile era, they are apps and accounts that can be watched one by one; in the AI era, the main form of applications is agents. With the exponential leap of big model technology and capabilities, natural language has become the most important programming language in this era. Each of us can create an AI application or agent that belongs to ourselves and others."