Is DeepSeek a New Voice Among LLMs in Public Opinion Simulation?
9 months ago
- #Public Opinion Simulation
- #LLM
- #Bias Mitigation
- The study evaluates DeepSeek's ability to simulate public opinions compared to other LLMs like Qwen2.5, GPT-4o, and Llama-3.3.
- DeepSeek-V3 performs best in simulating U.S. opinions on abortion, especially with Democratic or liberal personas.
- For Chinese samples, DeepSeek-V3 excels in simulating opinions on foreign aid and individualism but struggles with capitalism views.
- All LLMs tend to overgeneralize perspectives within demographic groups, showing consistent responses.
- The findings highlight the need to mitigate cultural and demographic biases in LLM-driven public opinion modeling.