Baidu told a sexy story

Author | Zhou Zhiyu

Editor | Zhang Xiaoling

Empowering everyone with programming skills to create millions of "super useful" applications.

At the Baidu World Conference on November 12, Baidu CEO Robin Li painted a grand blueprint. By launching no-code tools like Miaoda, the threshold for developing intelligent agents has been significantly lowered, and the era of "everyone is a programmer," as Robin Li mentioned earlier this year, is approaching.

Robin Li stated that this will be an unprecedented era where one can make money solely based on ideas.

Behind this grand vision is Robin Li's hope that Baidu's technological accumulation in AI over the past decade can leverage the momentum of large models to quickly realize commercial applications. Baidu's technology can also become a cornerstone in this transformation, evolving into a colorful ecosystem that drives productivity improvement through technology. In this process, Baidu can also find a stable "second growth curve" beyond its advertising business.

To successfully take this step, the implementation of AI applications is key. This is the current focus of Baidu's efforts.

Unlike other AI companies that talk extensively about their visions for AGI (Artificial General Intelligence) or Scaling Law during presentations, Robin Li was very pragmatic at this year's Baidu World Conference, with his speech mainly revolving around intelligent agents, iRAG (retrieval-augmented generation technology), and specific applications.

In Robin Li's view, the biggest change in the industry over the past 24 months is that large models have essentially eliminated hallucinations. In simple terms, the accuracy of AI's answers has greatly improved, and the responses are no longer "serious nonsense," but rather usable and trustworthy.

This improvement is attributed to the enhancement of RAG (retrieval-augmented generation), which can further utilize retrieval information to guide the generation of text or answers, thereby improving the quality and accuracy of content.

From a segmented perspective, RAG has a significant impact on improving textual content, but there are still issues with detail differences in multimodal content like images, leading to hallucinations. Baidu focuses on iRAG technology, making images more realistic and devoid of a machine-like quality.

This presents a commercially valuable scenario. For example, in the marketing of automotive products, AI-generated images can maintain precise control over product details while unleashing creativity in the scene. This significantly reduces costs for brands when creating promotional posters while enhancing realism.

Next, in scenarios such as film and television works, comic works, and continuous illustrations, the application of iRAG technology will further enhance production efficiency and reduce creative costs.

The foundational model capabilities and technological improvements lay the groundwork for the explosion of applications.

At the press conference, Robin Li showcased 100 major industry applications based on large models, as well as multidimensional intelligent agent applications. The continuous emergence of these applications heralds the arrival of the era of AI applications.

Robin Li also added fuel to the application explosion by launching Miaoda, a code generation tool that assists individuals without programming backgrounds in developing applications With the help of Miaoda, ordinary people who are not programmers can directly generate code with the assistance of large models, greatly lowering the development threshold; combined with support from multi-agent collaboration and the mobilization of multiple tools, individuals can complete an entire system setup through natural language interaction, thereby enhancing efficiency.

This is what Li Yanhong refers to when he says that one can make money just by having ideas. In his view, creating intelligent agents is very similar to building websites in the PC era or self-media accounts in the mobile era. Intelligent agents may become a new carrier for content, information, and services in the AI-native era.

It is undeniable that the current development of the AI industry seems to have entered a bottleneck period. The iteration progress of the industry leader OpenAI's model is not as expected; regarding whether AI model training can further experience a performance explosion, the industry has shifted from a previously unanimous optimism to ongoing disagreements; meanwhile, with user growth slowing down, there has not been a true "killer application," and the AI boom is gradually receding.

Li Yanhong is an optimist. He believes that the slowdown in the iteration of foundational models is a good thing for application development, and a major version update every two years is actually a suitable rhythm.

He also does not think that the absence of a "killer application" at present is a bad thing. Li Yanhong cites the steam engine revolution and the electrical revolution as examples, stating that AI is a new industrial revolution and therefore cannot be simply compared to the internet wave, assuming that after a few years of technological hype, clear super applications should emerge.

On the contrary, this will be a transformation at the infrastructure level, a change in underlying technological capabilities. This is similar to the philosophy of major companies like Alibaba, which believe that we are still in the early stages of AGI transformation, and the greatest imagination of AI lies in changing the physical world.

This will be a compelling story. If Baidu's AI technology and applications can successfully evolve into part of the infrastructure of the new era during this wave of AI, it can also overcome the disappointments of the mobile internet era and find new growth points.

However, from the perspective of infrastructure transformation, this AI revolution will be a long process. Currently, for large model players, finding ways to monetize and even achieve profitability is essential to gain the recognition of investors.

CICC also pointed out that after the vigorous development and rapid iteration of large models in 2023-2024, the growth rate of total users has slowed down. As the profitability of the reasoning segment of large model vendors gradually improves, the pace at which these vendors pursue monetization is accelerating.

While the AI revolution brings future imaginative possibilities, players like Baidu must also continuously transform their own businesses with AI, reflected in financial growth, and continue to contribute to commercialization increments. Even ideal long-termists must combine reality to turn technology into real value, supporting their journey into the new era of AI