A 12th-grade student, Adi Sing, has created an interesting project to test the capabilities of artificial intelligence. He is using the popular game Minecraft to pit AI models against each other in a building competition. To achieve this, he developed a website called MC-Bench. On this site, various AI models perform the same task, and users can vote on which model's building is better. After voting, users can see which model created each building.
Adi Sing explains the reason behind the creation of this project: Minecraft is one of the most popular games in the world, and many people are familiar with its structures. Even those who haven’t played the game can easily judge which building looks better. For instance, if one AI model builds a pineapple, users can easily determine which pineapple looks more realistic and accurate.
The MC-Bench website is built on the concept of AI models writing code to complete tasks. For example, models are given tasks like “build a snowman named Frosty” or “construct a beautiful hut on a tropical beach.” The AIs complete these tasks through coding, and the resulting builds are displayed on the site for users to judge. Users then choose the best creation.
Currently, there are eight volunteers working on the project. Major tech companies like Anthropic, Google, OpenAI, and Alibaba are covering some of the technical costs of the project, but they are not directly involved.
Adi Sing plans to expand the project with more complex and challenging tasks in the future. He believes that games like Minecraft provide an ideal environment to test the reasoning and decision-making abilities of AI. Since games are safer and more controllable than real life, they offer an ideal platform for testing AI’s logical and creative skills.
This approach better showcases the true potential of AI. For example, some AI models might perform well on legal exams but struggle to count simple letters in a word. MC-Bench tests these models with real, visual tasks, clearly revealing their level of competence.
Adi Sing believes that this project can serve not only as entertainment but also as a useful tool for tracking and evaluating the progress of AI development.
