LLM's have not "read" social science research and they do not "know" about the outcomes, they have been trained to replicate the exact text of social science articles.
The articles will not be mutually consistent, and what output the LLM produces will therefore depend on what article the prompt most resembles in vector space and which numbers the RNG happens to produce on any particular prompt.
The articles will not be mutually consistent, and what output the LLM produces will therefore depend on what article the prompt most resembles in vector space and which numbers the RNG happens to produce on any particular prompt.