Breaking the Assistant Mold: Modeling Behavioral Variation in LLM Based Procedural Character Generation

Maan Qraitem, Kate Saenko, Bryan A. Plummer · 2026

Read Breaking the Assistant Mold: Modeling Behavioral Variation in LLM Based Procedural Character Generation by Maan Qraitem, Kate Saenko, Bryan A. Plummer (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-03396.txt.

Description

Maan Qraitem, Kate Saenko, and Bryan A. Plummer tackle a storytelling problem: characters made by language models all sound like helpful assistants. Procedural generation builds vast worlds, levels, and quests. But character creation is underexplored. Alignment training leaves two biases behind. Such figures uniformly take agreeable moral stances. And they always answer questions directly, never refusing or deflecting. That kills dramatic tension and makes casts predictable.

Their framework, PersonaWeaver, separates world-building from behavior-building. The first covers roles and demographics. The second covers moral stances and interaction styles. The split yields casts with genuinely varied reactions and moral positions. It also gives second-order diversity in style markers, such as length, tone, and punctuation. The authors trace the biases to likelihood training and assistant fine-tuning. Richer characters, they argue, need those pressures deliberately undone.

Catalog record

CREATORMaan Qraitem, Kate Saenko, Bryan A. Plummer
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-03396

Browse the full library · About the Tower of Babel