Planted in Pretraining, Swayed by Finetuning: A Case Study on the Origins of Cognitive Biases in LLMs

TL;DR

This study investigates whether cognitive biases in large language models originate mainly from pretraining or finetuning, revealing that biases are primarily shaped during pretraining, with implications for bias mitigation strategies.

Contribution

The paper introduces a causal experimental approach to disentangle bias sources in LLMs, demonstrating that pretraining largely determines bias patterns over finetuning.

Findings

Bias variability is influenced by training randomness.

Pretraining has a stronger impact on biases than finetuning.

Bias patterns are more similar among models with the same pretraining backbone.

Abstract

Large language models (LLMs) exhibit cognitive biases -- systematic tendencies of irrational decision-making, similar to those seen in humans. Prior work has found that these biases vary across models and can be amplified by instruction tuning. However, it remains unclear if these differences in biases stem from pretraining, finetuning, or even random noise due to training stochasticity. We propose a two-step causal experimental approach to disentangle these factors. First, we finetune models multiple times using different random seeds to study how training randomness affects over $30$ cognitive biases. Second, we introduce \emph{cross-tuning} -- swapping instruction datasets between models to isolate bias sources. This swap uses datasets that led to different bias patterns, directly testing whether biases are dataset-dependent. Our findings reveal that while training randomness introduces some variability, biases are mainly shaped by pretraining: models with the same pretrained backbone exhibit more similar bias patterns than those sharing only finetuning data. These insights suggest that understanding biases in finetuned models requires considering their pretraining origins beyond finetuning effects. This perspective can guide future efforts to develop principled strategies for evaluating and mitigating bias in LLMs.