STRINGS

LLM writing tools have raised the average level of creative writing while flattening what makes writing distinct — they lift the floor but lower the ceiling. STRINGS gives authorship back to the writer by exposing the dials the model usually hides.

Instead of re-prompting and hoping, you steer generated dialogue directly: each character’s attributes are controls you can move in real time, and the model writes against them — closing the “gulf of execution” that makes most LLM tools feel uncontrollable.

GPT-4o is prompted to take on an expert-screenwriter persona; slider positions for each character attribute (response length, and dialogue style from hesitant to assertive) are translated into prompts on the fly, with continuity and character consistency enforced across the scene.

Built around the question of how fine-grained control of dialogue generation affects creativity and ownership for amateur writers. Evaluated in a within-subjects pilot (8–10 writers, 15 minutes per mode) comparing direct manipulation against a baseline — hypothesis: finer control improves creativity, efficiency, and ownership by reducing the gulf of execution.