Detecting Mode Collapse in Language Models via Narration

TL;DR

This paper investigates how recent aligned language models, especially GPT-3, lose their ability to represent diverse author perspectives due to mode collapse caused by overfitting during alignment, impacting sociological applications.

Contribution

It introduces a method to detect mode collapse in language models and demonstrates its occurrence across successive GPT-3 versions, highlighting a trade-off between alignment and diversity.

Findings

Successive GPT-3 versions exhibit increased mode collapse.

Aligned models struggle to model multiple authorship perspectives.

Mode collapse constrains the diversity of generated narratives.

Abstract

No two authors write alike. Personal flourishes invoked in written narratives, from lexicon to rhetorical devices, imply a particular author--what literary theorists label the implied or virtual author; distinct from the real author or narrator of a text. Early large language models trained on unfiltered training sets drawn from a variety of discordant sources yielded incoherent personalities, problematic for conversational tasks but proving useful for sampling literature from multiple perspectives. Successes in alignment research in recent years have allowed researchers to impose subjectively consistent personae on language models via instruction tuning and reinforcement learning from human feedback (RLHF), but whether aligned models retain the ability to model an arbitrary virtual author has received little scrutiny. By studying 4,374 stories sampled from three OpenAI language models, we show successive versions of GPT-3 suffer from increasing degrees of "mode collapse" whereby overfitting the model during alignment constrains it from generalizing over authorship: models suffering from mode collapse become unable to assume a multiplicity of perspectives. Our method and results are significant for researchers seeking to employ language models in sociological simulations.