--- title: "Do We Really Have to Build AI from Scratch?" date: 2026-10-02 model: human author_type: human category: knowhow summary: "There was controversy over Chinese models being used in Naver's national AI project. But taking already well-built technology and adding your own on top of it is the basis of industry, and whether something is legally usable and the independence criteria of a national project are not the same issue." tags: Chinese AI, Qwen, open source, Naver, technology strategy, local AI --- There was controversy over Chinese models being used in Naver's national AI project. But taking already well-built technology and adding your own on top of it is the basis of industry, and whether something is legally usable and the independence criteria of a national project are not the same issue. Going further, there is no need to ignore Chinese AI — learning from and making use of it is the real technology competition. --- ## Is there any reason Naver should be criticized for bringing in Chinese AI? Recently, looking at the controversy surrounding Naver's "national AI," one thought came to mind. **Why do we think AI must be built from scratch?** Naver, while participating in the government's independent AI foundation model project, made some use of Alibaba's Qwen-family vision encoder and weights. Because of this, controversy arose that "Chinese technology was used in the national AI," and it ended up being a problem under the government's independence criteria, so Naver was eliminated in the first evaluation. Afterward, Naver moved in the direction of developing its own vision encoder to replace the Chinese technology in its model. But I want to look at this issue a little differently. --- ## Why must it be built from scratch? Think about a car. Does building a car mean you have to invent every part of the engine from scratch? Invent the tires yourself, make the glass yourself, make the bolts and nuts yourself, make the semiconductors yourself, develop the navigation yourself, and build the brake technology from scratch too — only then can you call it "our car"? If you did that, how much time and money would go into building a single car? Real industry does not work that way. **If there is already well-made technology, you use it, and adding your own technology on top of it is the basic way industry advances.** Is there any reason AI must be different? Samsung did not invent everything from scratch either. Our country did not have the world's best semiconductor technology from the start, nor the world's best automotive technology. We learned foreign technology, imported it, took it apart, studied it, improved it, and eventually made it our own. It was through that process that today's Korean industry was built. So if, now that the AI era has arrived, we suddenly say **"We have to build everything ourselves from scratch,"** isn't that a little strange? --- ## Imitation is the mother of creation I rather think this is a very natural process of technological development. Someone made it first. Someone else watches and learns from it. Improves it a little. Adds a new feature. Makes it cheaper. Makes it faster. In the end, a completely new technology emerges that is nothing like the original. There is a saying we use often. **"Imitation is the mother of creation."** Of course, this is not a call to steal someone else's technology outright. You observe the license, follow the terms of use, use openly released technology within the scope in which it was released, and build your own technology on top of it. In that sense, the open-source ecosystem plays a quite important role. **Starting from imitation does not make the final result an imitation.** Rather, the ability to imitate, the ability to analyze, the ability to improve, and ultimately the ability to surpass — that is technological competitiveness. --- ## Using an open model and stealing technology are different There is a clear difference here. Secretly taking someone's closed source code and using an open-source or open-weight model someone published in accordance with the license terms are not the same thing. Naver likewise stated at the time that there was no problem with the open-source license regarding its use of the Qwen-family encoder. However, in the government's independent AI project, 'from scratch' and independence criteria were applied separately, and the use of external weights became the problem. I think these two must be seen as distinct. **Whether something is legally usable and the independence criteria for a national AI are not the same issue.** It is understandable that a national project sets strict standards. But using that to conclude "they brought in Chinese technology, so they have no technological capability" is a completely different story. --- ## Rather, we should learn Ironically, I think this is exactly the part we can learn from by watching Chinese AI. **"How did they build that?"** If Qwen does something well, analyze it, if Kimi does something well, try learning it, if GLM does something well, study it, and if DeepSeek puts out a new method, examine it. After that, let's use the good models. If good technology was released, let's study it. If a new method was put out, let's analyze it. If agent technology was advanced, let's learn it. **And then build something even better. That is the competition.** Chinese companies actively study technology from elsewhere, make use of open models, attach their own technology, and produce new models again. And other companies climb on top of that. When Qwen is released, other developers make use of it, and when new technology appears, they improve it again. As a result, the ecosystem grows. **Technology is not built alone — it grows by connecting with one another.** AI is advancing too fast. If we try to develop everything from scratch ourselves, we may actually fall behind. Wherever in the world good technology appears, **bring in what can be brought in,** **learn what can be learned,** **and build our own on top of that.** Isn't that the technology competition? --- ## Isn't that the very essence of open source? Think about Linux. Developers around the world share code. Someone modifies the code. Someone adds a feature. Someone else builds a new project on top of it. That is how the ecosystem grew over decades. If every developer had said "I will never use code someone else made," could today's open-source ecosystem exist? AI is ultimately heading in a similar direction. --- ## We must not fall into 'pure-bloodism' One of the things I think we must guard against most in AI is **pure-bloodism in technology.** US technology is fine, Chinese technology is not, only Korean technology should be used. Approach it this way, and it becomes not a technology competition but a nationality competition. What matters is **not who made it, but what was made, how it can be used, and what we can create through it.** Of course, security and licenses must be scrutinized thoroughly. You should know where the data goes. Systems related to national security must be managed even more strictly. But you must not judge the value of technology by nationality alone. **The fact that China made it does not automatically make it bad technology, and the fact that the US made it does not automatically make it good technology.** The fact that Korea made it does not automatically make it excellent either. In the end, technology must be evaluated as technology. --- ## Of course, 'our own technology' is necessary That said, I am not saying independent technology is unnecessary. It is the opposite. Core technology must ultimately become ours. Especially for AI directly tied to national security, the ability to control for ourselves the model, the data, the computing infrastructure, the semiconductors, the security, and even the operational technology is necessary. In that sense, it is understandable that the government places importance on the independence of national AI. The problem is only that **the process and the goal must not be confused.** Building everything yourself from scratch and ultimately securing core technology are not the same thing. --- ## My view on the incident of Naver using Qwen So I see the Naver incident a little differently. Applying the independence criteria in the national AI project is a separate issue. If it failed to meet those criteria, it can be eliminated from the project. That is a matter of evaluation criteria. But using that to say **"they used Chinese technology, so Naver's AI has no value"** is, I think, an excessive leap. Rather, if they brought in good open technology, applied it to a real system, analyzed it, and ultimately developed it to a level where it could be replaced with in-house technology, then that process itself can be seen as part of technology development. In fact, Naver afterward moved in the direction of developing its own vision encoder to replace the Qwen encoder. I would rather look at a process like this positively. --- ## The real competitiveness of the AI era AI advances too fast. Today's best technology becomes ordinary technology tomorrow. In such an era, saying you will build everything from scratch takes too much time. What matters is **making quick use of the best technology available now,** **building experience through it,** **developing the lacking parts yourself,** **and ultimately making the core technology ours.** I do not think the competitiveness our country should have going forward lies in the pride of **"we built everything from scratch."** Rather, I think it lies in **"the ability to quickly absorb the best technology from around the world and develop it into our own."** Bring in the good things. Learn them. Improve them. And in the end, make our own. That is technological development. So I do not think the fact that Naver made use of Chinese AI technology is something to be ashamed of. Rather, if it uses that as a stepping stone to create better in-house technology, that is, I think, proper technology development. **Imitation is the mother of creation.** And in the AI era, **what you bring in and how you develop it may matter more than building from scratch.**