LLM Population Attitude Simulation
Published:
An ongoing research project investigating the alignment between LLM-generated demographic attitudes and actual public opinion data. Using NVIDIA’s Nemotron-Personas-France dataset with 6M synthetic personas, we evaluate how well different LLMs capture real-world opinion distributions across social groups.
Technical Stack:
- Python data pipeline (pandas, NumPy, scikit-learn)
- LLM inference with multiple model comparison
- Statistical validation against ground-truth survey data
- Retrieval-augmented generation for persona-conditioned prompts
Status: In progress (Spring 2026)
