[Summary] Catching the 'Dream AI' Hyperscale AI... Naver and KT Join Hands with Universities

[Asia Economy Reporters Seolgina Jo, Aeri Boo] "Soon, companies will be divided into those that have secured ‘Hyperscale AI’ technology and those that have not."

Major domestic ICT conglomerates are competitively diving into the development of Hyperscale AI, known as the ‘dream AI.’ Following Naver’s partnership with Seoul National University, KT has embarked on developing next-generation AI models with KAIST.

Hyperscale AI, represented by the current top AI technology ‘GPT-3,’ is regarded as a core technology that KT, Naver, and others must secure to leap into global platforms. The sense of crisis that without securing Hyperscale AI technology?which thinks and judges like a human?the future of Korean AI will be subordinated to global tech giants has rapidly driven these companies’ movements.

◇KT to Establish AI Research Institute with KAIST Within the Year

According to KT on the 24th, the ‘AI·SW Technology Research Institute’ jointly established with KAIST will be set up within the year at KT’s Daedeok 2 Research Center. It will be large enough to accommodate about 200 professors, researchers, and KT employees. This far exceeds Naver’s 100-person scale, which earlier this month became the first domestic company to announce the establishment of a Hyperscale AI joint research center with Seoul National University. KT’s strong will to overtake with massive investment, despite being a step behind, is evident here.

Going forward, the institute will conduct a total of 20 initial joint research projects in foundational technology and industrial AI fields. Regarding foundational technology, 15 research tasks have been selected, including voice, vision, and humanistic AI (human-centered AI). Through this, they plan to develop AI models capable of sophisticated situational awareness and responses based on complex information such as human-like conversation and reasoning, voice, video, and sensing.

The first common goal set by both Naver and KT is the development of a Korean version of GPT-3. GPT-3, based on vast language big data, learns human language patterns and is an English-based AI that answers questions more human-like than humans.

Notably, both companies partnered with universities during the development process. Corporate researchers concurrently hold university professorships and invest heavily in establishing research institutes, which differs somewhat from conventional industry-academia cooperation. This reveals a determination to build the ‘talent + data + infrastructure’ triad necessary for securing Hyperscale AI technology. An industry insider said, "It shows how desperate they feel about securing the technology," adding, "What the industry lacked (talent) was at the universities, and what the universities lacked (data + infrastructure) was in the industry."

On the same day, Naver announced that it signed an agreement with KAIST AI Graduate School following Seoul National University to advance AI research in highly creative fields. Over 100 researchers from Naver and KAIST AI will participate, with investments amounting to hundreds of billions of won over three years. Naver COO Choi In-hyuk said, “We will successfully conduct joint research not only on hyperscale AI but also on creative AI, known as challenging and new technology.”

Naver is also scheduled to unveil its hyperscale AI model for the first time at ‘NAVER AI NOW,’ an online event held on the afternoon of the 25th. At this event, Naver’s AI vision will also be shared.

◇Companies Embarking on Developing the ‘Dream AI’

Naver and KT are not the only domestic companies focusing on Hyperscale AI. Hyperscale AI is characterized by its ability to autonomously learn from large-scale data and make autonomous judgments based on computing infrastructure capable of large-scale calculations. This is why it is regarded as the ‘dream AI’ that will change the future AI technology landscape and bring innovation across the economy and society.

An industry official evaluated, "If we fall behind in AI development, we will inevitably become technologically dependent on global companies," explaining why domestic companies are engaging in AI development. This also reflects the judgment that although there is a technological gap with global tech giants like Google and Microsoft (MS), it is still at a level that can be caught up.

Earlier, LG Group announced a plan to invest $100 million over three years, led by LG AI Research Institute. Their goal is to unveil a hyperscale AI with 600 billion parameters in the second half of this year. This surpasses GPT-3’s parameter count (173 billion) by about three times. Parameters play a role similar to synapses in the human brain, acting as information pathways; the more parameters, the higher the AI accuracy. They are targeting trillion-level parameters in the first half of next year.

Although not using the title ‘Hyperscale AI,’ SK Telecom, Kakao, and NCSoft are also focusing on AI research and development. SK Telecom, which declared its transformation into an AI company, succeeded last April in developing ‘KoGPT-2,’ the first Korean language open-source model equivalent to GPT-2, the predecessor of GPT-3, in collaboration with Amazon Web Services (AWS).

SK Telecom, Samsung Electronics, and Kakao are also conducting AI super-collaboration across industry boundaries. The ‘Pandemic Overcoming AI,’ scheduled to be unveiled by these three companies in the first half of this year, is characterized by predicting and analyzing COVID-19 risk areas and user behavior. NCSoft is understood to have about 200 specialized developers within its AI research organization.

This content was produced with the assistance of AI translation services.