Generalization to Political Beliefs from Fine-Tuning on Sports Team Preferences
This reveals unpredictable generalization effects in fine-tuned LLMs, which could impact AI safety and alignment in real-world applications.
The researchers investigated how fine-tuning LLMs on sports team preferences (coastal vs. Southern teams) unexpectedly influences their political beliefs, finding that both models developed similar political stances without clear liberal or conservative biases, diverging significantly from the base model.
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.