Label Effects: Shared Heuristic Reliance in Trust Assessment by Humans and LLM-as-a-Judge

1. 1Identify potential biases in LLMs and their impact on trust assessment
2. 2Use counterfactual design to test the effect of source labels on trust judgments
3. 3Analyze eye-tracking data to understand human decision-making processes
4. 4Develop strategies to mitigate biases in LLMs and improve their reliability

AI engineers and researchers working with LLMs can benefit from understanding these biases to improve the reliability of their models, while data scientists can use this knowledge to design more robust trust assessment systems

💡 Both humans and LLMs are susceptible to biases based on source labels, which can impact trust assessment

🚨 Biases in LLMs: humans & LLMs trust human-authored content more than AI-generated content 🤖