Large language models (LLMs) have created unprecedented opportunities for analyzing and generating language data on a massive scale, which has the potential to transform the field of social sciences since language data play a central role in all areas.  In this talk, we present a road map for using LLMs as computational social science tools.  To do so, we first analyze the zero-shot performance of more than 10 LLMs on a wide range of representative computational social science benchmarks. Then, we show how such findings about social constructs might be generalized to audio beyond text and increase the efficiency of social science analysis. Finally, we outline a few major concerns about the application of LLMs to social sciences, and make recommendations for investments that may help to address them.