[AI Revolution](98) "Learning Data Quality Determines AI Performance"... SelectStar Seeking Optimal Solution

From Learning Data Planning to Construction and Management
Accumulated 170 Million Work Data Records
"What Ultimately Makes AI Companies Different Is Data"

Kim Se-yeop, co-CEO of SelectStar, is explaining artificial intelligence (AI) training data. Photo by Younghan Heo younghan@

"Just like humans, for artificial intelligence (AI), 'who it learned from and what it learned' is extremely important."

There is an active movement to introduce generative AI into industrial sites, aiming for AI to learn autonomously and think similarly to humans. However, not all AI demonstrates optimal performance. For generative AI, which is designed to interact with data and improve based on it, the quantity and quality of training data directly determine its performance.

SelectStar is a platform company that provides services ranging from AI training data planning to selection, construction, analysis, and management. Among generative AI, large language models (LLMs), which receive significant attention, increase the probability of providing appropriate answers when trained on vast amounts of information. When AI is asked to infer the word that fits in a blank in a specific sentence, it learns by repeatedly processing numerous materials until the optimal answer emerges. Probabilistically, AI learns which word is most suitable for the blank, and the more data it acquires, the higher the probability of inferring the appropriate word. SelectStar analyzes how much data should be trained and how it should be structured to achieve optimal performance, producing and providing data tailored to each company’s needs. Since quality is as important as quantity, they seek the 'optimal' solution.

Kim Se-yeop, co-CEO of SelectStar, explained, "Creating materials to train AI is important, and we need to build data specialized for each industry." He added, "We help AI learn by processing materials collected by data workers with experience in specific fields." He continued, "Our role is to find 'what is most optimal' within limited resources depending on each case," and added, "AI data is also used for training and evaluation, and we are conducting business designing and producing evaluation data."

AI Training Data Service Diagram [Image source=SelectStar]

SelectStar shows strengths in LLM-related data training. Their experience and cases are overwhelming, and they also possess expertise. Since its establishment in November 2018, the accumulated work data has reached 170 million cases, and their clients include 230 companies such as Samsung Electronics, SK Telecom, and LG CNS.

They also provide solutions for copyright issues that may arise during the AI training process. CEO Kim said, "Lawsuits related to copyright are being filed due to crawling information from the internet without permission," and added, "We also play a role in supplying data to companies by obtaining data sales rights from licensed sources." In addition, they are attempting to expand services to the general public, not just companies, through AI video call services.

SelectStar is also pursuing an initial public offering (IPO) targeting the end of next year. CEO Kim stated, "AI models are provided by big tech companies, but the difference that companies and startups adopting those models can make ultimately comes down to data," and said, "We will strive to become the company that those developing AI always seek."