Researchers Develop CreativityNeuro Method to Make Language Models More Creative and Reduce Repetitive Outputs

A new technique called CreativityNeuro uses contrastive weight steering to enhance creative thinking in large language models without requiring retraining or fine-tuning. In human evaluations, the method improved performance on creativity tests by up to 14 percentile points and significantly reduced mode collapse—the tendency of LLMs to generate similar responses to open-ended questions.

Why it matters: As AI systems become more central to creative workflows, addressing the artificial hivemind effect and improving genuine divergent thinking could meaningfully expand LLM applications in content creation, design, and ideation.