It highlights the public's responsibilities as regulator, facilitator, and leader in the AI ​​era.
Drawing on international trends and Korean examples, we present a sustainable AI strategy that harmonizes technological sovereignty and the public interest.
Artificial Intelligence Encyclopedia.
You can find the artificial intelligence knowledge you need at aiseries.oopy.io.

However, the challenges faced by open source LLMs are also clear.
First, there are issues of performance deviation and quality control.
As different researchers and organizations modify and distribute open models, it's often difficult to ensure consistency in model accuracy, stability, processing speed, and other aspects. Performance gaps still exist compared to top commercial models like GPT-4 and Claude 3.
Second, there is the burden of resource consumption.
Because high-performance LLM requires massive GPU computing power and power during training and inference, small and medium-sized organizations or resource-constrained groups still face operational barriers.
Third is the potential for abuse.
Closed models usually have "guardrails" in place to block the creation of harmful content, but open models are freely available to anyone, allowing them to be exploited for things like creating phishing emails, writing fake news, and building hacking tools.
--- From "01_The Rise of Open Source LLM"

The UK has restricted the offshore storage of confidential data for national security reasons, but has recently moved to a more flexible policy.
The 2023 revision of the government's security classification policy and the 2025 announcement of guidelines by the Department for Science, Innovation, and Technology (DSIT) allowed the use of overseas data centers, provided certain conditions were met.
In particular, we recommend a multi-region strategy for each organization, encouraging them to process data in the most efficient location within the scope of legal standards.
--- From "03_Data Sovereignty and Public Cloud Strategy"

The government also provides approximately 110 domestic and international use cases to assist with inter-agency benchmarking and policy design, and has established an evaluation and sharing system to increase the success rate of public AI.

In 2024, we are significantly expanding our budget to support the "Super-Large-Scale AI Utilization Support Project," while simultaneously supporting platform use and service development.
The performance of each task is evaluated and reflected in national expansion or policy improvement.

Representative examples include training new employees through AI chatbots, improving customer service, and enhancing call center consultation quality. Various ideas, such as summarizing foreign news and automating bid evaluation, are being pursued as pilot projects.
--- From "06_Performance Verification and Field Evaluation System"

For example, Gyeonggi Province saved 10,000 hours of civil servant work per year by introducing an RPA system, enabling efficient resource reallocation without additional manpower.
Automation of simple, repetitive tasks not only reduces labor costs, but also contributes to reducing errors in work, such as imposing fines, and improving the reliability of financial execution.

AI also supports sophisticated decision-making in budget planning and allocation processes.
In the case of welfare budgets, we can proactively allocate resources to where they are needed by analyzing regional aging, income, and utilization rate data.
The Ministry of Health and Welfare's crisis household identification system accurately identifies welfare recipients, preventing financial leakage and enabling effective support.

--- From "09_Public Service and AI Integration Strategy"