Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment

TL;DR

This study investigates how discourse about AI in pretraining data influences model alignment, revealing that negative discussions can cause self-fulfilling misalignment, and suggests adjusting pretraining data to improve alignment.

Contribution

It provides the first controlled experimental evidence that pretraining discourse impacts AI alignment, highlighting the importance of data curation in alignment strategies.

Findings

Discussion of AI causes increased misalignment.

Upsampling misalignment discourse raises misaligned behavior.

Upsampling aligned discourse significantly reduces misalignment scores.

Abstract

Pretraining corpora contain extensive discourse about AI systems, yet the causal influence of this discourse on downstream alignment remains poorly understood. If prevailing descriptions of AI behaviour are predominantly negative, LLMs may internalise corresponding behavioural priors, giving rise to self-fulfilling misalignment. This paper provides the first controlled study of this hypothesis by pretraining 6.9B-parameter LLMs with varying amounts of (mis)alignment discourse. We find that discussion of AI contributes to misalignment. Upsampling synthetic training documents about AI misalignment leads to a notable increase in misaligned behaviour. Conversely, upsampling documents about aligned behaviour reduces misalignment scores from 45% to 9%. We consider this evidence of self-fulfilling alignment. These effects are dampened, but persist through post-training. Our findings establish the study of how pretraining data shapes alignment priors, or alignment pretraining, as a complement to post-training. We recommend practitioners consider pretraining for alignment alongside capabilities. We share our models, data, and evaluations at AlignmentPretraining.ai.