As the adoption of language models improves, so does the need to better represent specific users in the model. Are there aspects of an individual's belief system that a linguistic model could use for advanced alignment? After previous research, we investigate this question in the field of opinion prediction by developing a primex public information database from American residents with two US databases, as well as the first national opinion survey of the world. We provide a first comprehensive analysis of our data and show a number of explanations for belief and worldviews according to linguistic models. Our results showed how more and more information about Primx can benefit the NLP and Psychological Research communities, opening the way for further studies.
- † University of California
