Smart AI Is One Thing; AI People Want to Use Is Another

The report examined the difference between a well-made AI and an AI that users actually want to use, asking industry insiders whether they would use a real service built on a “for everyone’s AI” approach once it launched. In that context, an industry insider said that a new AI service needs at least one clear point of differentiation to make people want to use it.

The industry insider said Gemini was not initially his primary AI service. He had been using another AI service he was already familiar with, but he started using Gemini because of its strength in image generation, and later expanded his use to other functions as well, he said. The case shows that one distinct strength can be the trigger for choosing a new service.

Along with this trend, the rapid improvement in the capabilities of domestic AI models is also evident. All four teams participating in the sovereign AI foundation model project were listed in Epoch AI’s “Notable AI Models.” This year, the countries listed in the roster are the United States, China, and South Korea.

A key challenge for South Korea’s AI industry is quickly narrowing the technology gap with global leading models. Against this backdrop, efforts to secure sovereign foundation models have continued, and attempts to secure sovereign foundation models have been one pillar of the effort to reduce the technology gap.

However, improved technical capabilities and actual user choice are separate matters. One of the main standards for choosing an AI service is performance, but in real-world use, cost, speed, ease of use, and suitability for specific tasks also come into play, and the choice is made through a combination of multiple factors.

In this context, George Cameron, co-founder of Artificial Analysis, said on the 14th at the “LG AI Talk Concert 2026” held at LG Sciencepark Magok in Seoul that efficiency is as important as intelligence in the large-scale use phase of AI. He explained that the highest-performing model is not always the best choice, and that the key standards for practical AI use are not only the level of intelligence, but also usability under given conditions and the results it delivers.

In the generative AI market, major generative AI services such as ChatGPT, Gemini, and Claude, developer-focused services such as Cursor, and AI search-focused services such as Perplexity have all established themselves, broadening user options from general-purpose AI to specialized services by field.

For that reason, simply adding another option is not enough when choosing a new AI service; there has to be a reason to replace an existing service or use it alongside one. Factors that can serve as that reason include performance, strengths in specific functions, and strengths in the usage environment.

This same standard also applies to domestic AI services. It is not essential for domestic AI to outperform the world’s best models on every metric; it can highlight strengths such as better understanding of Korean, better understanding of the domestic environment, or better results in specific industries or tasks. Factors such as cost, security, and the ability to build in-house, which can influence corporate decisions, are also important, and such strengths must lead to actual use and encourage continued use rather than one-time trials.

An AI that scores highly in performance evaluations and an AI that people try directly and continue to choose are separate challenges. The next task for domestic AI is to build AI with high willingness to use and high reusability. The benchmark for completing technological value lies in the number of people who actually choose it.

Source: IT DAILY · Yang Seung-gap
Original: https://www.itdaily.kr/news/articleView.html?idxno=241751

References

• https://www.itdaily.kr/news/articleView.html?idxno=241751