LLM Political Bias Using Zero-Shot Classification & Fine-Tuned ML

This dataset displays numerical results attained from a bias evaluation conducted on 1800 model responses spanning six generative AI (GAI) models: DeepSeek Distilled Llama 8B, DeepSeek-R1, DeepSeek Chat, o1, Claude Opus, and Claude Sonnet. Four bias metrics were measured: political partisanship, topicality, sentiment, and objectivity. The evaluation used a combination of zero-shot classification and fine-tuned ML models to quantify each metric. This dataset intends to further research surrounding GAI political biases. Prompts and model responses were retrieved from a separate dataset found here.