[Digital Today reporter Jinju Hong (홍진주)] ChatGPT can now create games that can be run directly inside the chat window. If a user enters the desired game and rules, it generates a playable screen without a separate development environment. It can also change graphics, difficulty and how the game progresses.

TechRadar reported on Oct. 21 local time that OpenAI’s GPT-6 “Intelligent UI” feature can build interactive elements such as graphics, buttons, input forms and charts within chat responses. It does not stick to fixed response formats. Instead, it displays items such as a calculator, maps, interactive diagrams and simple games directly in the chat window, depending on the request.

A classic arcade game implementation is an example showing the potential uses of the feature. TechRadar used ChatGPT to build games based on Snake, Tetris, Space Invaders, Pac-Man and Pong and tested whether they worked. It also implemented modified versions that changed the visual mood and gameplay rules, not just the basic rules.

Snake was created with basic controls and a speed adjustment function. After applying a cyberpunk mood, it added moving obstacles, special items and speed that increases depending on the score. In the first attempt, it output code instead of a playable game. After requesting again that it be playable directly in the chat window, it produced the desired result.

The Tetris-based game worked relatively stably. It kept the basic structure of stacking blocks and clearing lines, while adding modified rules such as changes to the rotation board, special blocks and a function that removes part of the grid. It implemented a new play experience by adding hard-to-predict elements to a familiar game format.

The game based on Space Invaders added choice elements beyond combat. ChatGPT named it “First Contact.” Players can choose between combat and negotiation between rounds, and depending on the choice, they can obtain new weapons or stop an alien invasion. It was also structured so that the results of a choice affect what follows, such as making later attacks more dangerous.

The Pac-Man-based game kept the basic framework of moving through a maze, collecting pellets and avoiding enemies. ChatGPT named the game “Maze Muncher” and then changed it by combining a 1980s arcade and Halloween atmosphere. It also changed the maze’s structure and progression, including different wall layouts, secret passages and additional enemies.

Pong reproduced basic ball bouncing and paddle controls relatively stably. It supported easy, normal and hard difficulty levels. It could also be modified so that the game rules change randomly each time a point is scored. The ball can split in two, paddles can become smaller, and changes in gravity and obstacles can be added.

The completeness of the results was not the same for every request. In some cases, an executable screen was not generated from the start, requiring a renewed request. In other cases, detailed modified rules did not reflect accurately in the output. It could implement basic operation and visual composition, but showed limits in consistently handling complex conditions.

The case shows that the role of interactive AI is expanding beyond providing information to creating content and tools that work on the spot. Users can build simple software by describing desired functions in natural language and modifying results instead of coding games directly.

Its range of use is not limited to games. The same approach can implement educational quizzes, interactive visualisations and simple calculation tools. Still, using it for real services would require further verification of accuracy and stability, and of its ability to reflect complex requirements.

How sophisticated interactive content ChatGPT can create inside the chat window remains under test. Even so, the ability to change game rules and design through natural-language requests alone and run them immediately shows the potential for interactive AI to evolve into a content creation tool.