Steering LLMs to Evaluate and Amplify Creativity

Although capable of generating creative texts, LLMs are poor judges of what constitutes creativity''. In this paper, we show that we can leverage this knowledge of \textit{how} to write creatively in order to judge \textit{what} is creative. Our mechanistic approach extracts differences in the internal states of the LLM when being prompted to writeboringly'' or ``creatively'' to provide a more robust measure of creativity that corresponds more strongly with human judgment. We also show these internal state differences can be applied to enhance the creativity of generated texts.