Perhaps Grok-3 used the reasoning methodology from DeepSeek more than the underlying model, but the similarity of Grok-3 results to DeepSeek suggests that XAI used more than that.