This paper discusses the ongoing controversy surrounding South Korea's national AI model selection process, specifically regarding the "From Scratch" development requirement. The government initiated this project to enhance Korea's AI competitiveness by supporting a "representative foundational model" built purely with domestic technology. However, the definition of "From Scratch" in the context of AI remains ambiguous, leading to a heated debate.

The core of the controversy lies in whether AI models developed using open-source components should be considered "From Scratch." The term "From Scratch" originates from cooking, meaning to prepare a dish entirely from basic ingredients, avoiding pre-made or semi-finished products. While this concept signifies self-reliance in AI development, its application poses challenges: it's unclear what constitutes "basic ingredients" in AI, and to what extent architectures, public research papers, or open-source code are permissible. This has led to a conflict between "purity," advocating for strictly original development, and "realism," arguing that leveraging open-source, a global standard, is practical and necessary.

To illustrate this dilemma, the author draws an analogy from the cooking show "Black and White Chef 2." In the show, a contestant faced controversy for using "brown veal stock" as one of their ten permitted sub-ingredients. Brown veal stock, while seemingly a single ingredient, is itself a complex preparation involving roasting bones and vegetables, and simmering for extended periods—an "already completed base" that condenses numerous ingredients and a chef's time. Viewers debated whether counting it as a single ingredient was fair, aligning with the "letter of the rule" but perhaps not its "spirit." This situation is directly paralleled to open-source foundation models in AI: they appear as single entities but encapsulate immense capital, computing resources (e.g., hundreds of billions of Korean Won, thousands of GPUs), and vast data accumulation. Using them allows for significantly faster development of high-quality AI models.

The paper further extends the analogy by asking, "If brown veal stock is problematic, what about soy sauce?" Soy sauce, despite its simple appearance, also involves a complex, time-consuming fermentation process from numerous ingredients. Yet, it is universally accepted as a "basic ingredient" in cooking without controversy, indicating a "social consensus." In the "Black and White Chef" example, the production team accepted the brown veal stock as one ingredient, and the judges evaluated the dish solely on its culinary merit, prioritizing the "game" over a perfect definition.

The author argues that this presents a crucial hint for the government: instead of endlessly debating the "purity" of ingredients like stock or soy sauce, which lack a single "mathematical formula" answer, the focus should be on establishing clear "criteria" to end the unproductive debate. The true objective of the national AI project is to foster "strong" AI models capable of competing globally, not necessarily "pure" models that avoid every line of foreign code. The use of open-source is merely a means, not the end goal. Just as a chef's skill isn't diminished by using soy sauce, an AI company's technical prowess shouldn't be invalidated by utilizing open-source. The paper concludes by emphasizing the critical importance of time. While the nation deliberates the definition of "From Scratch," global tech giants are already advancing to the next generation of AI models. Excessive deliberation risks missing the global race entirely.