Next Token Prediction Is a Dead End for Creativity

This position paper argues that token prediction is fundamentally misaligned with real creativity. While next-token models have enabled impressive advances in language generation, their architecture favours surface-level coherence over spontaneity, originality, and improvisational risk. In contrast, creative acts, particularly in live performance domains, require dynamic responsiveness and stylistic divergence, enabling humans to transcend pre-learned patterns in the moment. We use battle rap as a case study to expose the limitations of predictive systems, demonstrating that they cannot truly engage in adversarial or emotionally resonant exchanges. As a result, such models fail to support the interactive flow states where human creators "lose themselves in the moment." Rather than pursuing greater predictive accuracy, we argue that AI research should embrace dialogue as a form of co-negotiated creative agency. This shift calls for approaches that prioritize real-time interaction, rhythmic alignment, and adaptive generative control. By reframing creativity as an interactive process rather than a predictive output, we offer a vision for AI systems that are more expressive, responsive, and aligned with human creative practice.