[The Era of Supermassive AI] Kim Kwangseop, CTO of Kakao Brain, "AI Model Competition Intensifies... Amazing Results Expected Within 1-2 Years"

Kim Kwangseop, CTO of Kakao Brain, is giving a lecture on the theme of '2022-2023 Open AI Era' at the 'Get Ready Web3.0, the 1st Mega AI Ecosystem Opening' forum hosted by Asia Economy on the 27th at Gravity Seoul Pangyo in Seongnam-si, Gyeonggi Province. / Seongnam=Photo by Kim Hyunmin kimhyun81@

[Asia Economy Reporter Kang Nahum] "Looking at the achievements AI has made today, the day when the future we imagine becomes a reality does not seem far off."

Kim Kwangseop, Chief Technology Officer (CTO) of Kakao Brain, made this prediction at the Asia Economy 'Get Ready Web3.0, The 1st Mega AI Ecosystem Opening' forum held on the 27th at the Gravity Hotel in Pangyo.

Kim CTO said, "Until now, the AI market has grown mainly around hardware and infrastructure, but going forward, growth will be centered on services and software."

In fact, recent AI model achievements have expanded from the early language domain to images, videos, and even art. Kim CTO said, "I expected that even if AI advanced, applying it to the art field would be difficult, but AI started writing poetry, suddenly began painting, and even won first place in competitions," adding, "As a developer, this is truly a time that makes me think deeply."

Kakao Brain's AI artist 'Karlo' is a representative example. Recently, Samsung Electronics' 'Galaxy Book Art Project' utilized Kakao Brain's AI artist Karlo. Participants enter keywords such as 'place,' 'action,' and 'object' in order on Samsung.com, specify the desired art style like pop art illustration, and Karlo instantly creates AI artwork.

For example, if you enter the keyword 'Donuts traveling through space' and specify art styles like pop art, illustration, or cartoon, Karlo understands the content and produces an image. Karlo is an AI artist developed by advancing Kakao Brain's mega image generation AI models such as minDALL-E.

Kim CTO said, "In research on image-text and image multimodal models, when text is input, an appropriate image is shown, so it is called 'text to image.' Models using similar techniques are becoming mainstream," adding, "Competition among these models with high industrialization potential is intensifying, and I think amazing results will emerge in video within 1 to 2 years."

Kim Kwangseop, CTO of Kakao Brain, is giving a lecture on the theme "2022-2023 Open AI Era" at the "Get Ready Web3.0, The 1st Mega AI Ecosystem Opening" forum hosted by Asia Economy on the 27th at Gravity Seoul Pangyo in Seongnam, Gyeonggi Province. / Seongnam - Photo by Kim Hyunmin kimhyun81@

Kim CTO pointed out versatility and creative ability as the biggest differences between recent AI models and previous ones. He said, "In the past, even if fine-tuning was done to solve a problem, creating that large model required huge costs, but now the biggest advantage is that the massive base model can be reused and applied to various problems," adding, "It is also remarkable that models like GPT and DALL-E are challenging the creative domain."

Kim CTO cited 'practicality' as the most important keyword in AI research and development. He said, "Most big tech companies' research outcomes are half-baked," and pointed out, "Even if there are papers and experimental results, companies, individuals, and institutions need to be able to implement them to use them. Collecting data itself is costly, and the training process requires enormous infrastructure."

Kim CTO's awareness of these issues led to the question of how to expand commercial usability. Kakao Brain's entry into the 'healthcare' field is a product of this consideration.

Kim CTO said, "Radiologists write reports after looking at X-ray images when there are lesions or suspicious areas. Nowadays, AI can create images if given text. Thinking in reverse, if there is an image, AI can write text," adding, "With a global shortage of radiologists, this will greatly help by significantly reducing the time doctors spend looking at images and writing reports."

He added, "We are also making efforts in new drug development. We are conducting research with Galaxy, an AI-based new drug development company," and said, "We are not trying to create a cure-all, but the goal is to shorten the long time it takes to develop new drugs. Reducing 10 years to 5 years would be a tremendous innovation."