World Editing: Intervening on Executable Worlds at Increasing Depth
The paper tests whether coding agents can modify running game worlds without breaking what should stay intact.
The authors define “world editing” as intervening on an existing executable environment at different depths, from properties to systems. They introduce IGMWorld and IGMBench, with 110 Minecraft and Terraria modding tasks and more than 1,100 executable state and behavior checks. The strongest evaluated setup solves 78.2% of tasks under a strict task-level measure, while criterion-level performance reaches 94.8%. Failures usually still build and load, but the edited world does not behave as requested; visual consistency is weaker, with all configurations below a 50% joint visual pass rate. Source: HF Daily Papers' note.
The authors define “world editing” as intervening on an existing executable environment at different depths, from properties to systems. They introduce IGMWorld and IGMBench, with 110 Minecraft and Terraria modding tasks and more than 1,100 executable state and behavior checks. The strongest evaluated setup solves 78.2% of tasks under a strict task-level measure, while criterion-level performance reaches 94.8%. Failures usually still build and load, but the edited world does not behave as requested; visual consistency is weaker, with all configurations below a 50% joint visual pass rate. Source: HF Daily Papers' note.
score 5