Can we control the creativity of large language models (LLMs)? A recent research paper challenges the popular belief that simply raising the "temperature" parameter—which controls randomness—makes an LLM more creative. The study explored how different temperature settings in the LLM Llama 2-Chat affected the novelty, typicality, cohesion, and coherence of generated stories. Researchers found a weak link between higher temperatures and more novel stories. However, this came at the cost of coherence, making the stories harder to follow. Interestingly, the study used the story generated with the lowest temperature setting (the "exemplar") as a baseline for comparison. This approach, inspired by cognitive science theories, revealed that while higher temperatures might lead to some exploration of new ideas, they don't necessarily unlock a broader range of creative possibilities. The research suggests that true LLM creativity isn't as simple as turning up the heat. More sophisticated methods, like specialized decoding strategies and better ways to analyze the information LLMs hold, are needed to unlock their full creative potential. This research highlights the complex nature of creativity and the ongoing quest to understand how it can be fostered in AI systems.