On 24 October, at the invitation of the Korea Industrial Intelligence Association (KOIIA), CLEVI CEO Lee Hwan-ho participated in the ‘2025 Academy for Developing Key Talent in Mid-sized Enterprises, Future Industry Response Strategy Seminar’ and delivered a keynote address.
We participated in the ‘AI Technology Adoption Cases and Practical DX Strategies’ session with a presentation titled ‘Physical AI Use Cases and Failure Factors and Improvement Strategies in the Process of Introducing RAG Agents’.
As interest in AI agents has recently increased, more companies are considering adopting AI. Many organisations and businesses are already piloting and operating AI, as far as we understand.
However, unfortunately, surveys have found that many businesses and organisations either fail to achieve the desired ROI or find that functions that worked during the PoC do not provide appropriate answers or operate properly in the main project or at the site.
Because of this information, many businesses are reluctant to adopt AI hastily.
In this lecture, CLEVI CEO Lee Hwan-ho explained in detail the limitations of RAG systems, a major cause of failure in rushed AI adoption, as well as open-source models and small LLMs (SLMs), explaining why previous adoption cases had failed.
Ultimately, the reason for using AI is the expectation that it will perform people’s work for them or deliver better performance than people.
As practitioners at companies that have adopted AI today do not trust the quality of AI responses, complaints are emerging that their workload is actually increasing because they have to check every AI response individually.
The reason is that most businesses use open-source SLM models as solutions after fine-tuning them.
Intuitively, we may think that AI will become smarter when it learns something new. However, when AI learns new knowledge, it instead loses memories from the past—in other words, the weights of arbitrary parameters.
An even more serious problem is that no one knows what information or knowledge has been lost.
Introducing AI is not something that requires only a small amount of resources, so you likely conducted extensive validation from the PoC stage.
Why, then, did those problems not appear at this stage? This was, of course, because it had been retrained using the target information.
You taught an AI that only knew numbers to learn the alphabet, and then asked it questions only about the alphabet, so naturally it provided good answers. But in the field, do you only need models specialised for such specific domains?
Our fields face new environments and new data every day. However, an AI trained while randomly destroying its existing knowledge cannot guarantee either general-purpose applicability or accuracy.
Therefore, you need to have a From-scratch Foundation large language model, or the technology to create one.
Because these models contain almost all the knowledge in the world, they provide not only high-quality answers but also a high level of accuracy.
For specialised domains, you can overcome the limitations of fine-tuned SLMs by using a technology called SFT (Supervised Fine-Tuning), or, for smaller amounts of data, a RAG system or CLEVI’s proprietary intelligence database system, Semantic Database technology.
The government also recognises the importance of these foundation models and is providing substantial support.
This is the National Representative AI Project. The project’s primary goal is likely to secure digital sovereignty.
CLEVI was also aware of the critical disadvantages of using open-source models, so we independently developed and trained CLEVI’s models, from data collection through to architecture design.
CLEVI’s models currently deliver performance that is in no way behind that of the world’s leading AI systems, even when competing against them.
In particular, because we own our own models, CLEVI is not affected by the dependency issues that can arise when using external models.
The adoption of AI is now becoming a necessity rather than a choice. However, simply adopting the latest technology makes it difficult to achieve the desired results.
True innovation begins only when you accurately understand the limitations of technology and establish the optimal strategy for your organisation’s environment and goals.
With differentiated technologies such as its proprietary large language models and Semantic Database, CLEVI is minimising the trial and error and risks that businesses experience during the AI adoption process.
We will continue to listen to the voices of those on the front line and do our utmost to become an AI partner that more businesses can trust. We hope you will build the true success of AI adoption together with CLEVI.