Show HN: WorldBuild Bench repo: testing LLM world coherence with 3D games (github.com)

A new repository called WorldBuild Bench has been launched, designed to test the coherence of large language models (LLMs) in constructing 3D games. This platform presents a uniform testing environment where different AI models build three distinct types of games—an arena combat game, a physics puzzle, and a racing game—using the same guidelines. Each model generates playable builds hosted in a browser, which allows for real-time user testing against static benchmarks to better assess the model's understanding of spatial, temporal, and causal coherence within a game environment. This initiative is significant for the AI/ML community as it shifts the focus from traditional coding benchmarks, which often fail to capture the nuances of gameplay, towards a system that evaluates how well models create engaging and coherent virtual worlds. The use of objective metrics, like the Playability Score and World Coherence Score, alongside audience preference ratings, aims to provide a clearer picture of an AI's capabilities in game design. The results could not only influence the development of future AI models but also improve how they are benchmarked, emphasizing user experience and interaction over merely functional outputs.