This page was machine-translated and may differ from the original. View original
KT Launches "Trust," a Massive AI Platform Offering Full-Stack AI
A total of 4 models are available, enabling the development of customized services for businesses.
Strengthening Trust… 70% of Hallucination Problems Solved
From B2B market conquest to advancement in five key areas
KT launches a massive AI model to open up a corporate-only LLM market.
On the 31st, KT held a press conference at the KT R&D Center in Seocho-gu, Seoul, and announced the launch of its massive AI platform, "Mi:dm." The official website (Midm.kt.com) opened on the 31st, allowing companies seeking to use the platform to apply. Companies applying will receive expert consulting to discuss their options.
A total of four models are available, ranging from lightweight to ultra-large. They are tailored to the size and intended use of a business. KT will offer a packaged AI full-stack solution, along with KT Cloud, and its trusted enterprise-specific AI cloud farm (Mi:dm CloudFarm). This is expected to enable the development of ultra-large AI-powered services at a reasonable cost, even without separate development and learning infrastructure.
■ Foundation model revealed, trust package provided
KT is releasing a core model (Foundation Model) for its ultra-large-scale AI platform, enabling businesses to leverage it. Businesses can fine-tune the platform to their desired form and create a variety of AI application services. A dedicated portal called 'KT Mi:dm Studio' provides a development environment where users can directly select, learn, and serve KT Mi:dm's foundation model.
In particular, KT's "Trust" is a proprietary foundation that has completed pre-training on trillions of data, a first in the domestic industry. KT stated, "This is expected to address security concerns regarding data sovereignty and the potential dependence of data on big tech companies when utilizing previously disclosed foundation models, as well as the limitations of full fine-tuning (FFT)."
KT provides a customized integrated package of an 'AI full stack' including KT Cloud's HAC (Hyperscale AI Computing) service, which can reduce GPU learning costs by approximately 27% compared to existing ones, and Rebellion's NPU infrastructure, which reduces inference costs by 50% compared to existing ones.
What stands out this time is that KT Trust offers a strong trust package. To overcome the 'hallucination' that has been highlighted as a problem with AI, KT applied Document AI, Search AI, and FactGuard AI technologies to increase reliability in all stages of search, inference, and response.
KT said, “We can reduce the phenomenon of hallucinatory answers by up to 70% compared to general generative AI services,” and “Among the four models of ‘Belief,’ the lightweight model with 7 billion parameters has proven its overwhelming performance by ranking at the top on the Korean LLM evaluation leaderboard, ‘Open Ko LLM.’”
■ B2B market exploration first…then expansion into five major areas
KT will initially focus on the B2B market, including commercializing corporate-only LLM programs and discovering new AI innovations. It then plans to expand its ultra-large AI business into five key areas: global, manufacturing, finance, public, and education. It will also leverage an open startup ecosystem to drive AI-based business innovation.
KT is jointly developing AI business models that utilize 'trust' with various AI startups, such as 'Upstage' for corporate LLM commercialization, 'Qanda' and 'Enuma' in the education field including Math-GPT, and 'BIMatrix' in the corporate business personal assistant field, and is targeting the domestic and global markets.
KT is upgrading the artificial intelligence of existing AI businesses such as AICC, Genie TV, and AI Call Assistant with 'trust', and is also applying ultra-large AI to KT's wireless services, IVI, and robots. Trust will also be applied to improve business processes in areas such as finance and communications, Genie TV marketing, and senior and childcare consulting. KT stated, "At KT's call center, where Trust was piloted, performance improvements were seen in various areas, including a 5% improvement in bot recognition rate, a 20% increase in post-processing speed, and a 30% increase in knowledge building speed."
KT AI/DX Convergence Business Division Vice President Song Jae-ho said, “The ultra-large AI market is undergoing rapid digital innovation with participation from global big tech companies to startups,” and added, “KT will open a differentiated ultra-large AI model and focus all its capabilities on enabling Korea to lead the digital transformation.”
본 기사에 대한 정정·반론·추후보도 청구는 보도 청구 안내를, 그간 게재된 보도문은 정정·반론보도 모아보기를 참고해 주세요.















