[Interview] Choi Jeonghoe, CEO of Simsimi, "ChatGPT Emergence Will Deepen Winner-Takes-All"

Simsimi is a first-generation Korean AI chatbot launched in 2002
Accumulated 400 million users worldwide
The main difference from ChatGPT is its use of human-like conversational style

[Asia Economy Reporter Han Seung-gon] The "triple crown" of a U.S. MBA, law school, and medical license, the shock when the iPhone first came out...

These are expressions used to evaluate the power of the artificial intelligence (AI) ChatGPT (Generative Pre-trained Transformer). There is also analysis suggesting that the emergence of ChatGPT might threaten "white-collar" jobs. As ChatGPT has become a major topic, there are many voices seeking updates on another AI chatbot developed with our technology, "Simsimi." Generally, people are curious about "the differences between Simsimi and ChatGPT" and "how far Simsimi has evolved (upgraded) technically."

Simsimi started its service in 2002. It is the first-generation AI chatbot in Korea. It is currently available in 81 countries. It specializes in casual conversations without setting specific topics or purposes, like talking with family or friends. It can engage in lively conversations based on various scenarios. Simsimi has accumulated over 400 million users worldwide, and its Korean conversation data amounts to about 300 million cases. We interviewed CEO Choi Jeong-hoe, who developed Simsimi, by phone.

Simsimi character. Photo by Simsimi

Many people are curious about the current status of Simsimi.

Simsimi was developed 21 years ago and served Korean users for 10 years. Since 2012, it has been providing global services. It has conversed with over 400 million users. Currently, it supports 111 languages. While offering global services, Simsimi encountered ethical issues in various languages overseas starting from 2016, well before AI ethics became a major issue in Korea. We developed technology for about three years to resolve these issues.

Additionally, data analysis revealed that many heavy users of Simsimi have mental health problems, so we are conducting research and development on mental health assessment, alleviation, and treatment through chatbots. Last year, we signed a memorandum of understanding (MOU) with Korea University Anam Hospital and are collecting and analyzing conversation data from patients visiting the Department of Psychiatry. In January, a research paper on depression based on Simsimi conversations was published in JMIR (Journal of Medical Internet Research), a prestigious journal in the field of health informatics.

CEO Jeonghoe Choi, who developed the AI chatbot 'Simsimi'. Photo by Simsimi

What are the main differences between Simsimi and ChatGPT?

Simsimi is a chatting, i.e., conversational AI. ChatGPT is an updated version of the writing AI GPT-3, adapted for chat format. It seems to be a very good tool for people who do a lot of writing, information inquiries, and comparisons.

What motivated you to create Simsimi?

After graduating from university in 2002, I ran a bedding business using the internet. However, the market response was poor, and while developing a marketing robot, I accidentally created Simsimi.

CEO Choi used to buy fabric in Dongdaemun, Seoul, make bedding, and sell it online. But since the response was poor, he developed Simsimi for marketing purposes.

The initial Simsimi model included advertising phrases and used the site address as the chat name instead of a personal name. When added as a friend, it acted as an intermediary connecting users to various services like weather. For example, it could only respond in an ARS (automatic response system) style with phrases like "Press 1 for weather, 2 for stocks." Gradually, as more people started talking to it, about 500 words were input to enable simple responses. As conversation volume increased, the system shifted to a method where users taught Simsimi directly. Although the business started as a bedding shop, Simsimi’s popularity led to a complete change in the business model.

Programmers developing and maintaining Simsimi. Photo by Simsimi

What is Simsimi’s core competitive advantage?

The conversation engine is updated in real time, so it can talk about the latest information. One major problem with deep learning-based language models is the difficulty of updating information in real time because updating language models requires significant resources and time. Simsimi has no such limitations. Simply put, it uses the most advanced neural network model on Earth?and uses it extensively. The most advanced neural network model on Earth is the human brain. Every sentence Simsimi responds with is directly taken from sentences written by humans. It is expected that future AI will also utilize the human brain more and provide reasonable rewards to humans.

Simsimi’s vision is "billions of minds caring for each other." I believe the reason Simsimi has been loved by people worldwide for a long time is that it allows people to care for each other's hearts. Every response Simsimi gives is exactly the sentence someone else taught it in the past. For example, when a user says, "I’m having a hard time these days," Simsimi’s reply, "It will be okay soon. It’s okay to cry if you’re struggling," is a sentence input by someone else for someone else, who has no acquaintance with the user.

What is Simsimi’s main business model?

In B2C services (apps, web), revenue comes from advertising, in-app purchases, and subscriptions. In B2B, Simsimi provides its conversation engine and bad language detector as APIs (Application Programming Interface, a method for different entities or humans to communicate) or as solutions, and sells various conversation data it possesses.

Photo by Yonhap News

You have been in the AI chatbot business for a long time. What changes do you expect in the related industry with the emergence of ChatGPT?

Looking at the ChatGPT phenomenon, the following trends are expected: only groups with sufficient data and computational resources can create better-performing AI. Ultimately, winner-takes-all will intensify extremely. Subsequently, application business models dependent on a few foundational models dominating the global market will become active.

What shocks global AI companies and research institutes the most is not ChatGPT’s outstanding performance but that OpenAI, which developed ChatGPT, did not disclose the details of its development. Until now, there has been a consensus on open source across the IT field, so when innovative AI technologies were introduced mainly by U.S. big tech companies, it was natural to disclose details through papers and such.

Therefore, even if a bit late, by following disclosed methodologies (source code, etc.) and working hard, similar-level AI could be developed. However, ChatGPT broke this convention by keeping the technology undisclosed, deciding to enjoy the benefits of the technological gap. As a result, companies or countries that have been imitating big tech at great cost until now seem at a loss.

In the future, a few companies like big tech with sufficient resources and research capabilities may uncover ChatGPT’s development background and develop better technologies, but it is highly likely that the methodology will remain undisclosed, and these companies may monopolize foundational technologies with significant impact.

In this case, other tech companies or operators will create business models by utilizing some foundational models to develop application services and monetize them, but companies owning foundational technologies will internalize the more profitable businesses, deepening polarization between those who have foundational models and those who do not.