Process for Adapting Language Models to Society (PALMS) with Values-Targeted Datasets

TL;DR

PALMS is an iterative process that fine-tunes language models on value-targeted datasets to reduce harmful biases and align outputs with societal values, showing effectiveness across GPT-3 sizes.

Contribution

The paper introduces PALMS, a novel iterative fine-tuning method using curated datasets to align language models with societal values without losing capability.

Findings

PALMS significantly improves adherence to target values and reduces toxicity.

Effectiveness of PALMS increases with larger model sizes.

A small, curated dataset can substantially alter model behavior.

Abstract

Language models can generate harmful and biased outputs and exhibit undesirable behavior according to a given cultural context. We propose a Process for Adapting Language Models to Society (PALMS) with Values-Targeted Datasets, an iterative process to significantly change model behavior by crafting and fine-tuning on a dataset that reflects a predetermined set of target values. We evaluate our process using three metrics: quantitative metrics with human evaluations that score output adherence to a target value, toxicity scoring on outputs; and qualitative metrics analyzing the most common word associated with a given social category. Through each iteration, we add additional training dataset examples based on observed shortcomings from evaluations. PALMS performs significantly better on all metrics compared to baseline and control models for a broad range of GPT-3 language model sizes without compromising capability integrity. We find that the effectiveness of PALMS increases with model size. We show that significantly adjusting language model behavior is feasible with a small, hand-curated dataset.