Creative Differences: Differentiating Human and LLM Responses to a Creativity Task

This research examines creative responses generated by humans and three large language models (GPT, Claude, and Gemini) for a creativity task. Using a fine-tuned RoBERTa model, we achieved 99% classification accuracy with 10-fold cross-validation. SHAP analysis revealed distinctive linguistic features differentiating LLM-generated responses from human-generated ones.