KT-led 'AI One Team' Develops Massive AI... Also Utilized in Giga Genie

ETRI, KAIST, Hanyang University Joint Research Agreement
Building Top-Level Large-Scale GPU Infrastructure
Commercialization Targeted for 2022

[Asia Economy Reporter Cha Min-young] KT’s ‘AI One Team,’ launched last year with the goal of making South Korea the “No. 1 AI nation,” is embarking on the development of a hyper-large artificial intelligence (AI) language model through multi-party joint research among industry, academia, and research institutes. The research aims for commercialization in 2022 and will also be utilized to enhance KT’s GiGA Genie service.

On the 18th, KT announced that it signed a joint research agreement with the Korea Electronics and Telecommunications Research Institute (ETRI), KAIST, and Hanyang University?participants of the AI One Team?to develop a “hyper-large AI model.” Hyper-large AI is designed to learn from massive amounts of data based on infrastructure capable of large-scale computation, enabling it to think autonomously like a human.

Until now, hyper-large AI models have required research in various fields such as data refinement, learning algorithms, distributed and parallel learning, model compression, and application model discovery, making it difficult for a single company to pursue independently. Access to large-scale GPU infrastructure for AI research has also been challenging for universities and research institutions.

With this agreement, ▲ KT will oversee research and development (R&D) and commercialization tasks including building large-scale GPU infrastructure, data collection and analysis, model training, and application task implementation. ▲ KAIST will handle machine learning algorithm optimization and application development. ▲ Hanyang University will contribute in data refinement and filtering algorithm development, distributed and parallel algorithms, and model optimization. ▲ ETRI will participate in large-scale model training and model development.

The AI One Team plans to complete the first phase of hyper-large AI training by the end of the year and commercialize the hyper-large AI model by the first half of 2022. Additionally, it will build large-scale AI computation infrastructure and expand its scale to support models with over 200 billion parameters.

The AI One Team researchers will prioritize eliminating AI bias through enhanced training data filtering and balancing among datasets. They also plan to proactively address AI ethics issues by actively utilizing various deep learning-based detection technologies such as harmful content filtering.

Notably, this hyper-large model is designed in a task-oriented form that can be commercialized immediately. The improved core technologies such as STT (speech-to-text), dialogue, TTS (text-to-speech), and TA (task automation) will first be applied to enhance KT’s GiGA Genie and AICC services. Furthermore, these technologies will be applied across various business sectors of AI One Team participating companies, including manufacturing, finance, logistics, and distribution.

In the future, the AI One Team will expand its scope with hyper-large AI models into areas such as multimodal cognition including video and robotics, image-based interpretation, and human-centered AI. It also plans to support many companies, universities, and research institutes in easily accessing and utilizing the AI One Team’s hyper-large model for Korean language intelligence and industrial applications.

Lee Yoon-geun, Director of the AI Research Institute at ETRI, stated, “By further developing ETRI’s Korean language AI technology, Exobrain, and applying it to the hyper-large language model project, we will contribute to securing AI technology sovereignty in the Korean language processing field.” Hong Kyung-pyo, Head of KT’s Convergence Technology Institute, said, “The development of a hyper-large AI model that consolidates world-class technologies and experiences from AI One Team member institutions holds special significance in securing next-generation core AI technologies. Through this project, we will contribute to strengthening South Korea’s national AI industry competitiveness and lead digital transformation across industries and changes in customer lifestyles.”

Meanwhile, the AI One Team includes a total of 10 industry, academia, and research organizations such as KT, ETRI, KAIST, Hanyang University, Hyundai Heavy Industries Group, LG Electronics, LG Uplus, Korea Investment & Securities, Dongwon Group, and Woori Bank. The team aims to nurture AI talent, promote open R&D, and strengthen the AI ecosystem.

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