VERIGAME
Human annotators refined VERIGAME specifications for accuracy, completeness, consistency, and evaluability. VERIGAME carries contamination risk because selected games may appear in LLM training corpora. VERIGAME is a dataset of 100 game-ge…
1 sources - 5 claims
Human annotators refined VERIGAME specifications for accuracy, completeness, consistency, and evaluability. VERIGAME carries contamination risk because selected games may appear in LLM training corpora. VERIGAME is a dataset of 100 game-generation tasks across seven genres. VERIGAME tasks contain anonymized game identifiers and natural-language specifications. Candidate games for VERIGAME were sourced from the Apple App Store and Steam and filtered for popularity and high ratings.