AI Paper Atlas

Research map / Large language models / Alignment and preferences

StoryAlign: Evaluating and Training Reward Models for Story Generation

Haotian Xia, Hao Peng, Yunjia Qi, Bin Xu, Lei Hou, Juanzi Li

ICLR 2026 · Large language models › Alignment and preferences · approach: Steering and correcting LLM behavior · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Premise Back-Generation training pairs (ablation)StoryReward-Qwen · StoryRMBAverage accuracy: 73.9 -> 75.0Improved
add Human-Guided Continuation training pairs (ablation)StoryReward-Qwen · StoryRMBAverage accuracy: 72.8 -> 75.0Improved
add Prompt-Guided Rewriting training pairs (ablation)StoryReward-Qwen · StoryRMBAverage accuracy: 69.9 -> 75.0Improved
add training on human-written stories and preference pairs (StoryReward-Qwen vs Qwen3-8B base)Qwen3-8B · StoryRMBAverage accuracy: 53.5 -> 75.0Improved
train StoryReward on Llama-3.1-8B-InstructLlama3.1-8B-Instruct · StoryRMBAverage accuracy: 24.3 -> 61.4Improved
use exact human-written stories as chosen instead of human-guided LLM continuation (ablation)StoryReward training data · StoryRMBAverage accuracy: below 30%Hurt
ablated components drop on coherence/relevance (ablation)StoryReward-Qwen · StoryRMB Coherenceaccuracy: 77.5 -> 82.3Mixed
evaluate reward models on Human-LLM pairs instead of LLM-LLM pairsexisting reward models · StoryRMB subsetsaccuracyHurt
select best of 16 stories using StoryReward vs other reward modelsLlama-3.1-70B-Instruct generations (MoPS) · BoN head-to-headinferior-story selection rate: fewer than 30%Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy