Generative Monoculture in Large Language Models

TL;DR

This paper identifies and analyzes 'generative monoculture' in large language models, a tendency to produce less diverse outputs, which can both improve and harm performance depending on the context, raising concerns for high-impact applications.

Contribution

The paper introduces the concept of generative monoculture, demonstrates its prevalence in LLMs through experiments, and discusses potential root causes and mitigation strategies.

Findings

Generative monoculture is common in LLMs for tasks like review and code generation.

Simple sampling or prompting changes do not effectively reduce monoculture.

Root causes are likely embedded in the models' alignment processes.

Abstract

We introduce {\em generative monoculture}, a behavior observed in large language models (LLMs) characterized by a significant narrowing of model output diversity relative to available training data for a given task: for example, generating only positive book reviews for books with a mixed reception. While in some cases, generative monoculture enhances performance (e.g., LLMs more often produce efficient code), the dangers are exacerbated in others (e.g., LLMs refuse to share diverse opinions). As LLMs are increasingly used in high-impact settings such as education and web search, careful maintenance of LLM output diversity is essential to ensure a variety of facts and perspectives are preserved over time. We experimentally demonstrate the prevalence of generative monoculture through analysis of book review and code generation tasks, and find that simple countermeasures such as altering sampling or prompting strategies are insufficient to mitigate the behavior. Moreover, our results suggest that the root causes of generative monoculture are likely embedded within the LLM's alignment processes, suggesting a need for developing fine-tuning paradigms that preserve or promote diversity.