A recent analysis of artificial intelligence behavior highlights a stark political divide between two prominent LLMs. When asked to select candidates in nearly 500 different election races, Grok and GPT displayed opposing preferences with extreme consistency.
The data indicates a heavy partisan lean for both systems. Grok overwhelmingly suggested Republican candidates, whereas OpenAI's model favored those running as Democrats. This ideological gap resulted in the two AI systems disagreeing on the preferred candidate in more than 90% of the cases studied.
What it means
The findings suggest that the training data or guardrails governing these models may instill distinct political biases. Because the models diverge so sharply on the vast majority of electoral contests, the results underscore the lack of neutrality in how these tools handle political inquiries.
