video1.0<iframe src="https://www.loom.com/embed/79151d2fc76b4eb38802ae6a457f5680" frameborder="0" width="1662" height="1246" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12461662Loomhttps://www.loom.com12461662https://cdn.loom.com/sessions/thumbnails/79151d2fc76b4eb38802ae6a457f5680-7612db520be92e4f.gif177.174In this video, I introduce my project, Impossible Moments, which serves as a benchmark for future models, focusing on reasoning capabilities across various disciplines such as physics and philosophy. I developed this benchmark using five different agents during a single Cloud Code session, resulting in 12 scenario categories and five top tiers: Spark, Fracture, Rapture, Singular, and Impossible. The Impossible tier presents scenarios that are currently unsolvable, like making a submarine invisible. I encourage researchers to explore these open-source prompts and frameworks to enhance their own benchmarks. My goal is to foster collaboration and innovation in the development of large language models.