A clever person has already found a unique application for ChatGPT-4, the latest artificial intelligence language model from the creators of ChatGPT. Turns out GPT-4 can read Game Boy Advance game code Pokémon Emerald and turn it into a fully interactive text-based adventure, complete with the mechanics of a Pokémon adventure.
twitter user Dan Dangond discovered that the AI model could run Pokémon Emerald, a popular Game Boy Advance game released in the United States in 2005. Dangond shared a screenshot-by-screenshot thread that features their conversation with the chatbot. The game starts more or less as it would if you played it on the handheld – except for some features, like setting the clock at the start – but runs as a fully interactive text-based adventure. Dangond “plays” the game by responding to the AI in chat.
As always with the main Pokémon games, in Pokémon Emerald you explore a pixelated world and fight other Pokémon in turn-based battles, so the text largely describes the battles and calculates them. At the start of Dangond, the AI offers the choice between the three starters in the region: Treecko, Torchic and Mudkip. From there, the Mudkip is sent out to battle a Poochyena, which is a real Pokemon that you can battle on the first route.
Starts the game smoothly, pretty accurate so far, skipping the van and clock setting scene which honestly isn’t the most fascinating thing about this game anyway pic.twitter.com/o4nqtMdCHH
—Dan Dangond (@DanDangond) March 15, 2023
The rest of the game plays out as a rough text simulation that incorporates all of the key Pokemon mechanics, like type strengths and weaknesses. However, there are several missteps, for example, it fails to calculate that the bug/soil type Nincada is weak to water. He also randomly decided to start including precision calculations after playing part of the game, and missing a Rock Tomb attack. He also skips several battles and misses some beats.
This text-based game comes at a time when many people are still wondering what generative AI’s real capabilities are. According to OpenAI, GPT-4 is a multimodal model, which is “less capable than humans in many real-world scenarios”, but can exhibit “human-level performance on various professional and academic criteria”. According to other tech journalists who have covered the technology, it’s supposedly more capable and sophisticated than its predecessor, ChatGPT, and it is pass standardized tests left and right. We can now add running a Pokemon game to this list of skills.
ChatGPT-4 was not created with running video games in mind, and it shows. Anyone who would like to play the game like this would greatly benefit from playing the original game. Without visuals and a general knowledge of the world, it would be almost impossible to play. Still, it’s a fun discovery that GPT-4 can run this game at any capacity, and therein lies the potential for fun and tinkering.