Progress and Standardization of Large Model + Data Intelligence Applications by the China Academy of Information and Communications Technology

The China Academy of Information and Communications Technology (CAICT) presents its recent work in the data‑intelligence field, emphasizing the research findings, trends, and standardization efforts related to large‑model‑plus‑data‑intelligence applications.

It outlines the evolution of artificial intelligence from the 1950s Turing test, through expert‑system eras, the rise of machine learning, deep learning breakthroughs, and the emergence of pre‑trained large models culminating in the 2023 ChatGPT‑3.5 release.

Three core technologies for applying large models are discussed: prompt engineering (low‑cost but stability‑sensitive), model fine‑tuning (requires substantial domain data and expertise), and Retrieval‑Augmented Generation (RAG) which combines external knowledge bases with model inference to improve answer accuracy and controllability.

The article describes new application paradigms such as chat‑BI, intelligent Q&A, creative generation (text‑to‑image/video), and decision‑support agents, highlighting their growing impact across industries.

Key challenges identified include insufficient data governance, high implementation and private‑deployment costs, and gaps in domestic hardware/software capabilities.

To address these issues, CAICT has launched a standardization program focusing on three areas: data‑intelligence application technology, data‑intelligence services, and business digitalization. The program has produced several standards, including:

Technical Requirements for Large‑Model‑Driven Intelligent Data Analysis Tools

Technical Requirements for Large‑Model‑Driven Intelligent Knowledge Graphs

Technical Requirements for Retrieval‑Augmented Generation

Technical Requirements for Large‑Model‑Driven Intelligent Q&A Systems

Overall Technical Requirements for Data Agents

Technical Requirements for Data‑Analysis Agents

Each standard defines capability domains, sub‑domains, and detailed technical items to guide the development, evaluation, and deployment of large‑model‑enabled data‑intelligence solutions.

The article concludes by emphasizing the ongoing standard‑development efforts aimed at supporting industry adoption of large‑model‑driven data‑intelligence technologies.