Large Language Models for Narrative Planning Search Heuristics
We wondered whether large language models like GPT could be used as search heuristics for narrative planners like Sabre, which carry the costly burden of having to explain all the actions in a story for the characters who take them. If the LLM could successfully guess which actions are likely to become explained in a given context, the narrative planner could prioritize visiting explainable actions earlier, and thereby find solutions faster.
Integrates GPT prompting into the Sabre Narrative Planner to explore LLM-guided search methodologies for planning-based story generation.
Rachelyn Farrell, Stephen G. Ware. Large language models as narrative planning search guides. IEEE Transactions on Games, vol. 17, num. 2, pp. 419-428, 2025.
The Git repo also includes verbose output from our experimental searches of the Sabre benchmark problem set using the GPT-guided planner described in the article.