Naver, Kakao in horserace over super AI with differing target and approach

2022.09.07 14:11:00 | 2022.09.07 14:11:28

이미지 확대
Korean big-tech names Naver and Kakao are in horserace in development and application of super-sized artificial intelligence (AI) in differing fields.

According to sources on Tuesday, Naver is trying to monetize its hyper-scale AI called Hyper CLOVA, a hyper-scale AI language model introduced for the first time in Korea in May last year from its company in company organization CLOVA dedicated to AI research.

Naver became aggressive in bringing Hyper CLOVA-based new products and services to the market this year. They include no-code AI tool CLOVA Studio, AI voice recorder and converter CLOVA Note, and AI call service CLOVA Care Call.

In contrast, Kakao is conducting a hyper-scale AI development project centered on digital health care through its AI research subsidiary Kakao Brain. It plans to build an AI-based antibody drug design platform with hyper-scale AI language and multimodal models trained in data sets of texts and images. Hyper-scale AI-based medical imaging services are also on the list for commercial development. Kakao aims to become a game changer in the AI market in at least two to three years.

Naver and Kakao are different in their hyper-scale AI development methods and computing infrastructure. Naver introduced Nvidia’s supercomputer systems. Building and operating a supercomputer system requires at least 100 billion won, according to IT market estimates. In contrast, Kakao Brain chose to add Google Cloud’s machine learning hardware accelerator TPU (tensor processing unit) to its existing server. They are also different in AI size. Hyper CLOVA targets 204 billion parameters, whereas Kakao Brain is making various attempts while reducing costs by setting the number of parameters with large AI utility from 1 billion to 30 billion.

Even in AI learning data, the two have different principles. Naver utilizes Korean language data accumulated through its platform service, but Kakao decided to exclude the use of in-house data. Naver does not disclose its hyper-scale AI model, but Kakao is releasing its version on GitHub, the world’s largest open-source community. Kakao Brain plans to release the English and Japanese models of the Korean language model KoGPT within this year.

However, both companies still have a long way to go in the monetization of hyper-scale AI. Both companies so far are losing tens of billions of won every quarter in this business.

By Lim Young-shin and Minu Kim

[ⓒ Pulse by Maeil Business Newspaper & mk.co.kr, All rights reserved]