Recent successes in Large Language Models (LLMs) have shown their vast capabilities in a wide variety of tasks. From lawyer exams to USA Biology Exams, models such as GPT-4 performed exceedingly well, often achieving 90th percentile and above. Since these models are trained on vast amount of corpus (or text) from many different fields, they are able to handle many different situations and tasks. In this paper, we make use of LLMs’ wide range of capability to conduct social simulations. Instead of recruiting large number of human volunteers, the use of LLMs as agents saved both time and resources, allowing faster iteration of our simulations. More specifically, we built a website to study the online interactions of “human agents”. GPT-4 turbo from OpenAI’s playground was used. In order to make the LLM mimic human behaviour as closely as possible, we used prompt engineering to guide the LLM to interact in a more human way.