Towards Human-Level Book-Writing Capability
A new paper discusses the challenges of aligning large language models with the requirements of high-quality creative writing. The authors propose a novel dataset construction and training framework aimed at improving book-scale creative writing capabilities. By using a multi-resolution planning scaffold derived from public-domain novels, the model learns to generate text that is more aligned with human literary styles.
- ▪Large language models often produce structurally correct but stylistically generic writing.
- ▪The proposed framework involves summarizing novels at different levels to train the model effectively.
- ▪The goal is to shift the model's output from assistant-style prose to more human-like literary writing.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.17064 (cs) [Submitted on 16 May 2026] Title:Towards Human-Level Book-Writing Capability Authors:Jan Zierstek, Matteo Batelic, Maya Medjad, Tim Schönenberger View a PDF of the paper titled Towards Human-Level Book-Writing Capability, by Jan Zierstek and 3 other authors View PDF HTML (experimental) Abstract:Large language models optimized for instruction following and agentic tasks remain poorly aligned with the requirements of high-quality creative writing. Fiction frequently depends on behaviors that assistant-tuned models are explicitly trained to avoid, particularly deception, moral ambiguity, and unreliable narration.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
