TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

media creative tools

The Model Learned Where to Be Creative

CreativeInstruct marks spans where a post-trained language model should recover diversity without abandoning answer quality.

Published Updated Story ID: mp-2026-08-10-007
Read the complete editionStory JSON

Summary

CreativeInstruct marks spans where a post-trained language model should recover diversity without abandoning answer quality.

The method trains models to insert special creativity spans, aiming to recover base-model diversity while retaining post-training quality. It also introduces a graph-edit metric for narrative structure. In the authors' human study, annotators preferred CreativeInstruct outputs as more creative than post-trained baselines in 70.3 percent of comparisons; reinforcement learning from the creative checkpoint also improved reported AMC and MATH results. These are preprint evaluations, not a general measure of creativity.

Why it matters

CreativeInstruct marks spans where a post-trained language model should recover diversity without abandoning answer quality.

Limits and context

  • These are preprint evaluations, not a general measure of creativity.

Key claims

  1. CreativeInstruct marks spans where a post-trained language model should recover diversity without abandoning answer quality.

    Qualification: These are preprint evaluations, not a general measure of creativity.

    Evidence: source-2026-08-10-007

Sources

  1. arXiv preprint 2608.07460arXiv · primary research

Corrections

No corrections have been recorded for this story.