Paradigm团队推出了一款名为RSI Simulator的网络游戏，用于演示人工智能研发的经济学原理[1]。玩家在游戏中扮演AI实验室的角色，通过投入劳动力、计算和数据等资源进行研发活动，直到实现自我加速阶段[1]。这款互动工具基于Elasticity Institute的论文《递归自我改进的经济学》以及其他基础研究成果开发而成[1]。
游戏采用针对教学目的优化的参数设计，而非追求预测准确性[1]。通过可交互的模型，玩家可以理解AI能力轨迹和递归自我改进的动态特征[1]。研究发现，弱链约束在递归自我改进中占据主导地位，计算能力和数据资源可能成为关键瓶颈[1]。此外，递归自我改进可能呈阶段性推进，在达到物理极限前停止[1]。驱动这一过程的关键指标是发现速率相对于当前模型能力的弹性[1]。
Paradigm has created RSI Simulator, an interactive web-based game designed to demonstrate the economic principles underlying AI research and development.[1] Players take on the role of an AI laboratory, allocating labor, compute, and data resources toward research efforts until achieving recursive self-improvement.[1] The game is grounded in the Elasticity Institute's paper "The Economics of Recursive Self-Improvement" and other foundational research, providing an interactive model to help users understand AI capability trajectories and the dynamics of recursive self-improvement.[1]
The simulator's parameters have been optimized for educational purposes rather than predictive accuracy.[1] Key findings from the model reveal that weak-link constraints dominate the recursive self-improvement process—compute and data may become bottlenecks limiting advancement.[1] The research suggests that recursive self-improvement could proceed in phases, potentially halting before reaching physical limits.[1] According to the model, the critical metric driving recursive self-improvement is the elasticity of discovery rate with respect to current model capabilities.[1]