Reproduce “Spoiler Alert: Narrative Forecasting as a Metric for Tension in LLM Storytelling” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Peiqi Sui, Yutong Zhu, Tianyi Cheng, Peter West, Richard Jean So, Hoyt Long, Ari Holtzman
The authors introduce 100-Endings, which repeatedly tests whether a model can forecast a short story’s ending from successive prefixes. They use the measure to compare fiction and guide a beat-by-beat generation pipeline that manages what scenes reveal or withhold. New Yorker stories average 0.765 no-rate versus 0.630 for zero-shot AI stories; this measures model-judged predictability, not human tension.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.