Vernacular? I Barely Know Her: Challenges with Style Control and Stereotyping

TL;DR

This paper evaluates the ability of state-of-the-art LLMs to control style and avoid stereotypes, revealing significant inconsistencies and cultural insensitivity issues despite some improvements.

Contribution

It provides a comprehensive analysis of contemporary LLMs' style control capabilities and highlights persistent challenges with stereotyping and inconsistency.

Findings

Models show large variability in style control performance.

Performance improved significantly in the second task.

Models often generate culturally insensitive content.

Abstract

Large Language Models (LLMs) are increasingly being used in educational and learning applications. Research has demonstrated that controlling for style, to fit the needs of the learner, fosters increased understanding, promotes inclusion, and helps with knowledge distillation. To understand the capabilities and limitations of contemporary LLMs in style control, we evaluated five state-of-the-art models: GPT-3.5, GPT-4, GPT-4o, Llama-3, and Mistral-instruct- 7B across two style control tasks. We observed significant inconsistencies in the first task, with model performances averaging between 5th and 8th grade reading levels for tasks intended for first-graders, and standard deviations up to 27.6. For our second task, we observed a statistically significant improvement in performance from 0.02 to 0.26. However, we find that even without stereotypes in reference texts, LLMs often generated culturally insensitive content during their tasks. We provide a thorough analysis and discussion of the results.