“DeepSeek's R1 model, as detailed in a January 2025 paper, demonstrated emergent higher-order cognitive behaviors like having an "aha moment" during reasoning, a result of reinforcement learning.”