Predicting results of social science experiments using large language models
GPT-4 simulates people well enough to replicate social science experiments with high accuracy. Note this is done by having the AI prompted to respond to survey questions as a person given random demographic characteristics & surveying thousands of "AI people" (author thread)
![](https://samim.io/static/upload/GUZ58UQW0AAFjIP.png)
![](https://samim.io/static/upload/GUZralwa8AYSK2K.jpeg)
![](https://samim.io/static/upload/GUZvBJYbYAAH3Td.jpeg)