Naver's HyperClova X AI unrivaled in Hangeul

South Korean internet giant Naver on Thursday unveiled HyperClova X, an upgraded version of its hyperscale artificial intelligence model, which boasts greater Korean language skills than its US rivals, OpenAI's ChatGPT and Google's Bard.

In response to the global generative AI craze brought about by ChatGPT late last year, the South Korean IT firm has chosen HyperClova X to be the backbone model of its AI-powered services and become its future growth engine.

“Naver dreams of changing the world with technology. … We're ready to face the new change opened by generative AIs,” Naver CEO Choi Soo-yeon said in her keynote speech at a press conference held in Seoul on the same day.

In a bid to raise its competitiveness, the CEO said HyperClova X was equipped with large amounts of data and technology by learning information starting from several decades ago until the present, covering various sectors including search engines, shopping, reservation platforms and videos.

“We’ve spent some 1 trillion won ($756 million) on AI over the past five years. We’ve steadily invested 22 percent of our operating profit in research and development to realize the up-to-date technologies and quality data of HyperClova X,” Choi said. “Our own internet data center also played a big role in advancing the technology.”

Naver’s latest AI tool, the world’s third-largest language learning model and the No. 1 in Korea, can understand Korean social context, laws and systems more accurately and naturally, the company said. Compared to the 3.5 version of GPT, the Naver model has learned 6,500 times more Korean data.

Sung Nako, Naver Cloud's head of technology, highlighted that HyperClova X’s winning rate compared to the 3.5 version of GPT is 75 percent. Naver’s advanced technology also decreased its Korean-based hyperscale AI hallucinations by 72 percent, Naver's chief scientist Kim Young-bum said.

To further accelerate its massive data-analyzing capabilities, the CEO also hinted that the company will open a new large data center in the central city of Sejong in November, housing more than 600,000 units of servers. The data center is expected to play an important role as a center for the company’s hyperscale AI services, while preserving massive amounts of data, she added.

At Thursday’s event, Naver also disclosed two HyperClova X-based services: Clova X and Cue:.

Clova X, which kicked off its beta testing at 4 p.m. on the same day, is a large language model-based AI chatbot. It can provide the most appropriate responses to various user demands based on its conversational style, versatile ability to create stories, summarize and translate texts and encode languages. Users can take multiple turns to ask questions and get answers through the service.

“Clova X will serve as a tool to make life more convenient and improve work productivity,” Sung said. “Based on excellent Korean and English language skills and vast knowledge, it can be used for a variety of purposes, from getting help with business writing such as business reports or self-introductions, to practicing for interviews and counseling.”

Cue: is a new interactive AI chatbot developed specifically for online searching. It can understand complex and long queries containing complex intentions and provides three-dimensional search results using reliable up-to-date information necessary for generating answers. The beta service is scheduled to start next month and will be applied to Naver’s integrated search within the year.

Naver plans to focus on strengthening cooperation with other companies centering on generative AI and increasing its partners' business productivity by using the AI tool for various endeavors. "We'll support our sellers, creators and partners to encourage their diversity and grow further with our technology," Choi said.

“We hope that the system will be established in a win-win way for content companies and technology companies in the AI ​​data learning process,” she said. “We'll also work with global partners such as Samsung Electronics to find optimized models, such as AI semiconductor chips."