Thinking Outside the (Gray) Box: A Context-Based Score for Assessing Value and Originality in Neural Text Generation

TL;DR

This paper introduces a context-based scoring method rooted in information theory to evaluate and enhance the value and originality of neural text generation, balancing diversity and quality.

Contribution

It proposes a novel score for assessing and incentivizing originality in language models, and demonstrates its effectiveness in fine-tuning models for creative tasks.

Findings

Improves diversity without sacrificing quality in generated text

Enhances originality in poetry and math problem solving

Effective as a reward signal in reinforcement learning

Abstract

Despite the increasing use of large language models for creative tasks, their outputs often lack diversity. Common solutions, such as sampling at higher temperatures, can compromise the quality of the results. Dealing with this trade-off is still an open challenge in designing AI systems for creativity. Drawing on information theory, we propose a context-based score to quantitatively evaluate value and originality. This score incentivizes accuracy and adherence to the request while fostering divergence from the learned distribution. We show that our score can be used as a reward in a reinforcement learning framework to fine-tune large language models for maximum performance. We validate our strategy through experiments considering a variety of creative tasks, such as poetry generation and math problem solving, demonstrating that it enhances the value and originality of the generated solutions.