Baidu’s iRAG and “Miaoda”: Solving AI Hallucinations and Powering the No‑Code Revolution

During the Baidu World 2024 conference, Baidu founder Robin Li delivered a keynote titled “Applications Are Here,” announcing two major AI technologies: the retrieval‑augmented text‑to‑image model iRAG and the no‑code tool “Miaoda.” Both are positioned as breakthroughs to improve AI practicality and democratize application development.

iRAG: Cutting Hallucinations in Image Generation

Li identified the elimination of hallucinations as the biggest change in the AI industry over the past 24 months. Traditional large‑model image generation still suffers from severe hallucinations, especially for specific landmarks or personalities. Baidu’s iRAG combines billions of indexed images from its search engine with a powerful base model, delivering ultra‑realistic images that “far surpass native text‑to‑image systems and remove the machine‑like feel.” Demonstrations included a flawless rendering of a Volkswagen vehicle crossing the Great Wall and an “Einstein traveling the world” scene, both free of distortion.

The technology is presented as a major step toward making AI‑generated visuals commercially viable, reducing costs for brand promotion, film, comics, and poster creation to near‑zero.

Intelligent Agents: The Next AI Application Wave

Li emphasized that intelligent agents will become the dominant form of AI applications. He compared agents to websites in the PC era and social‑media accounts in the mobile era, noting that agents act more like human assistants, salespeople, or customer service reps. Examples covered four categories: company agents (e.g., a BYD sales agent that boosted conversion by 119 %), role agents, tool agents, and industry‑specific agents such as the legal assistant “Fahngbao,” which has answered 16.6 million legal questions and can calculate compensation, draft documents, and recommend lawyers.

The Baidu Intelligent Agent platform now hosts a TOP 100 list of agents across roles, tools, industries, and entertainment, attracting 150 000 enterprises and 800 000 developers.

“Miaoda”: No‑Code Programming Powered by Large Models

“Miaoda” was launched as a no‑code platform that enables anyone to build applications using natural language. Its three core capabilities are:

No‑code programming: the large model generates code directly from user prompts.

Multi‑agent collaboration: the model orchestrates different agents to complete complex tasks.

Multi‑tool invocation: the model can call web search, iRAG, map APIs, and other utilities.

In a live demo, Li described a scenario where a user asks “Miaoda” to create a guest‑invitation and management system for a product launch. By providing a brief textual requirement and a document with event details, the platform coordinated a team of five agents (group‑leader, planner, editor, programmer, and QA) to design, generate content, write code, test, and finally deploy the system—all without writing a single line of code.

Images from the presentation illustrate the step‑by‑step workflow, the generated invitation, and the backend statistics view.

Market Impact and Growth Figures

Li reported that by early November the daily call volume of Baidu’s large model exceeded 1.5 billion, a 7.5× increase from the 200 million reported in May and roughly 30× the 50 million calls a year earlier. He described this “steep growth curve” as evidence of a booming Chinese large‑model application market.

Conclusion

Li concluded that eliminating hallucinations is foundational for the upcoming explosion of AI applications. With iRAG improving visual fidelity and “Miaoda” turning natural‑language ideas into functional software, Baidu aims to empower millions of users and enterprises to create “super‑useful” applications, heralding an era where ideas alone can generate revenue.