Bottom-Up and Top-Down Analysis of Values, Agendas, and Observations in Corpora and LLMs

TL;DR

This paper introduces a validated method for automatically extracting, assessing, and characterizing the diverse socio-cultural values expressed in texts and LLM outputs, aiming to improve safety, accuracy, and cultural fidelity.

Contribution

It presents a novel approach to analyze and manage the socio-cultural values in large language models and their generated texts, addressing value alignment challenges.

Findings

Effective extraction of heterogeneous value propositions

Assessment of value resonance and conflict in texts

Characterization of pluralistic value alignment in data

Abstract

Large language models (LLMs) generate diverse, situated, persuasive texts from a plurality of potential perspectives, influenced heavily by their prompts and training data. As part of LLM adoption, we seek to characterize - and ideally, manage - the socio-cultural values that they express, for reasons of safety, accuracy, inclusion, and cultural fidelity. We present a validated approach to automatically (1) extracting heterogeneous latent value propositions from texts, (2) assessing resonance and conflict of values with texts, and (3) combining these operations to characterize the pluralistic value alignment of human-sourced and LLM-sourced textual data.