What bug did the GPP discover?

 title: 'Figure 2 | Number of output tokens per second while generating (i.e. after the first chunk has been received from the API), for different models. Source: ArtificialAnalysis.ai, imported on 2025-06-15'

The Gemini Plays Pokémon (GPP) agent encountered a novel bug in the code of Pokémon Red/Blue[1]. According to the report, GPP is likely the first AI to find this bug in the game's code[1].

This occurred in the Seafoam Islands, which contain 5 floors involving multiple boulder puzzles[1]. These puzzles require the player to navigate mazes and push boulders through holes across multiple floors to block fast-moving currents preventing the player from using HM03 Surf in various locations in the dungeon[1].