Researchers are exploring new ways to benchmark AI models using games like Pictionary and Minecraft. These games challenge models' problem-solving skills, creativity, and understanding of spatial relationships. Proponents argue that these tests can help identify more sophisticated AI capabilities, such as resourcefulness and multimodality. However, some experts question the significance of these benchmarks, suggesting they may not accurately reflect real-world reasoning or adaptability.

Source: https://techcrunch.com/2024/11/05/people-are-using-games-like-pictionary-to-benchmark-ai-now/