Generalization to Political Beliefs from Fine-Tuning on Sports Team Preferences

TL;DR

Fine-tuning language models on sports team preferences can unexpectedly influence their political beliefs, but the resulting biases are not straightforward or predictable, highlighting complex effects of narrow domain training.

Contribution

This study reveals how fine-tuning on specific sports preferences can alter political belief expressions in language models, challenging assumptions about predictable bias directions.

Findings

Models show similar responses regardless of coastal or southern team fine-tuning.

Fine-tuned models do not exhibit clear liberal or conservative biases.

Models' willingness to justify radical answers varies significantly.

Abstract

Fine-tuned LLMs often exhibit unexpected behavior as a result of generalizing beyond the data they're shown. We present results in which an LLM fine-tuned to prefer either coastal sports teams or Southern sports teams adopt political beliefs that diverge significantly from those of the base model. While we hypothesized that the coastal model would become more liberal and the southern model would become more conservative, we find that their responses are usually similar to each other, without a clear-cut liberal or conservative bias. In addition to asking the models for numerical ratings of agreement with relevant political statements, we ask them to elaborate on their more radical answers, finding varying degrees of willingness to justify themselves. Further work is needed to understand the mechanisms by which fine-tuning on simple, narrow datasets leads to seemingly unrelated changes in model behavior.