Naver and Kakao Launch Full-Scale Competition in Mega AI Technology

원본보기 아이콘

[Asia Economy Reporter Kang Nahum] The competition between Naver and Kakao in developing artificial intelligence (AI) technology is intensifying. While actively engaging in industry-academia collaboration to advance technology, both companies are also expanding their own hyper-large AI ecosystems to dominate the AI industry leadership in their unique ways.

◆ Naver and Kakao Head to Seoul National University, the ‘AI Research Mecca’ = According to industry sources on the 10th, Kakao signed a business agreement on the 6th with Seoul National University AI Institute (AIIS) for joint research on cutting-edge AI technology and for nurturing and utilizing AI talent. Since September last year, Kakao has been conducting research with Seoul National University on ‘Speech Synthesis Language Processing Unit and Speech Database Construction,’ and plans to significantly expand the scope of research through this agreement.

First, Kakao will join Seoul National University’s industry-academia cooperation program ‘AIIS Membership’ and collaborate on joint AI technology research in computer vision, speech processing, and natural language processing fields, as well as explore AI utilization methods across various industries. In particular, it plans to focus on research to implement a more advanced AI assistant. The goal is to dramatically enhance the AI assistant Casper, applied in the comprehensive work platform ‘Kakao Work,’ within five years.

Opportunities to share cutting-edge AI research achievements will also be continuously provided. Academic events, seminars, and programs for nurturing AI talent will be planned and operated.

Prior to this, Naver also established a cooperative relationship with Seoul National University for AI technology development. Following an agreement signed in May for joint research on hyper-large AI, Naver recently set up a dedicated research space within the Seoul National University AI Institute. Ten AI labs at Seoul National University are participating in research collaboration with Naver, and their achievements are already visible, with various papers being presented at world-renowned AI conferences.

◆ The Opening of the ‘Hyper-Large AI’ Leadership Competition = The forefront of the technological competition between the two companies is undoubtedly in the hyper-large AI field. Hyper-large AI refers to AI trained on large-scale data using supercomputing infrastructure that operates at speeds much faster than general computers.

Technically, Naver is considered to be one step ahead. In May, Naver unveiled its independently developed hyper-large AI, HyperCLOVA. HyperCLOVA’s parameters number 204 billion, surpassing GPT-3’s 175 billion parameters. HyperCLOVA is known to have been trained on an amount of Korean language data equivalent to 50 years of news and 9 years of Naver blogs.

HyperCLOVA technology is being integrated into various services such as the companion service ‘Care Call’ for elderly living alone and the logistics demand forecasting service ‘CLOVA Forecast,’ accelerating commercialization. Ultimately, the goal is to build a hyper-large AI ecosystem based on HyperCLOVA. By providing HyperCLOVA to external partners, Naver aims to increase data collection volume and maximize value.

Kakao’s AI research subsidiary Kakao Brain recently released ‘KoGPT,’ a Korean-specialized AI language model based on the ‘GPT-3’ model, on the largest open-source community GitHub. KoGPT was built on 6 billion parameters and 200 billion tokens (digital files traceable and storable via blockchain) of Korean language data. KoGPT particularly understands Korean lexically and contextually and provides results that users desire. It performs all language-related tasks such as ▲judging the positivity or negativity of a given sentence ▲summarizing long sentences into one line ▲predicting conclusions after sentence inference.

Kakao is also planning a hyper-large AI ecosystem. However, its approach differs from Naver’s in that it is based on blockchain technology. The model is designed to continuously grow in size by reflecting additional data added on the blockchain. Kim Ildu, CEO of Kakao Brain, said, “Anyone can contribute by helping with language model training computations or providing data containing valuable knowledge. Those who contribute will hold a portion of the model’s shares,” adding, “We envision an ecosystem where contributors receive usage fees proportional to their shares when someone uses the model.”

© The Asia Business Daily(www.asiae.co.kr). All rights reserved.