GIST Graduate Student Team Wins International AI Game Playing Competition
The 'A Great Toe' team, winners of the 'Entertainment Game Agent Challenge.' Provided by GIST
View original imageThe Gwangju Institute of Science and Technology (GIST) announced on March 29 that the team "A Great Toe"—composed of graduate students from the Department of AI Convergence (integrated master's and doctoral program student Park Yoochun, master's program students Kim Kyungbo, Kim Yujin, and Hwang Geumhwan, under the supervision of Professor Kim Kyungjoong)—won first place in the "Small Language Model" track at the international AI game playing competition, the "Orak Game Agent Challenge."
This competition was organized by the global gaming company Krafton, known for developing titles such as Battlegrounds, TERA, and inZOI. NVIDIA, Amazon Web Services (AWS), and OpenAI served as official sponsors. The total prize pool amounted to 20,000 dollars. Out of 117 participating teams, the "A Great Toe" team secured first place in the "Small Language Model" track and was awarded a prize of 6,000 dollars.
The "Orak Game Agent Challenge" is an international competition in which AI directly plays various video game environments and competes in its ability to solve continuous tasks. The competition is notable for its comprehensive evaluation of the AI's decision-making, strategy formulation, and environmental adaptability.
In the "Small Language Model" track, teams were required to design agents (AI programs) within the constraints of limited computing resources and model size (amount of learned information and computational capacity). Participating teams optimized and trained their strategies under these conditions to compete in terms of performance. A small language model refers to an AI optimized for efficiency and lightweight operation by reducing model size and computation compared to large language models (LLMs).
Each team had to use a single language model to play four different types of games in sequence: Super Mario, Pokémon Red, StarCraft II, and 2048. This process required complex abilities such as spatial reasoning, strategic decision-making, resource management, and mathematical reasoning.
The "A Great Toe" team designed a new system architecture based on "action candidate generation" to enhance the stability of AI's decision-making. Conventional language model-based AI systems determine actions immediately after interpreting the game situation, which can lead to unrealistic or inconsistent choices. To address this limitation, the "A Great Toe" team introduced a "structural analysis module" that organizes executable actions as a first step.
This module analyzes the game state to generate action candidates and set priorities, considering exploration direction, resource status, and feasibility. The AI then selects the optimal action among these candidates. In addition, the system includes various stabilization mechanisms: it verifies whether the AI's chosen action is actually executable, restricts actions to a predetermined format, and applies additional instructions to correct errors if incorrect results occur. These measures ensured stable operation across different game environments.
Professor Kim Kyungjoong commented, "This achievement is significant in demonstrating the potential for universal application of small language models (SLMs) across various game environments. I am proud of the students for overcoming highly uncertain conditions with creative algorithms."
Hot Picks Today
"Two Spend 500,000 Won in a Flash, Yet Wait Two Hours?" ... 'Dreamland' Shunned by Kids and Foreign Tourists
- The Era of Getting 200 Million Won After 30 Years Is Over... A Japanese Bank's "Perpetual Department Head" Who Just Clocked In Cries Out
- Spain Hit by Its Worst Tragedy... Lee Kangin, Moreno and Real Madrid Join Soccer World in Mourning
- They Said, “Make It Your Home for Life” - Then Suddenly Kicked 200 Passengers Off... What Happened to the Around-the-World Cruise?
- "130 Rice Bowls Sold Out in a Day"... 'Unfiltered Face' Adds 8 Million Won to Night Market Sales: China's Bungeunwoo
Student Kim Yujin said, "I'm happy that we were able to achieve such a great result through teamwork. In the future, I want to continue researching AI systems that can make stable decisions in diverse environments."
© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.