LLM Simulated Data: The Good, the Bad, and the Ugly

null, Justin Sytsma (Victoria University of Wellington), Ethan Landes (University of St. Andrews)

Jan 8, 16-18 UTC+0, Ethan Landes (Kent); Justin Sytsma (Victoria University of Wellington), LLM Simulated Data: The Good, the Bad, and the Ugly

Abstract: Multiple recent projects have found remarkable correlations between human responses to studies and LLM simulations of those responses. While not perfect, in meta-analyses the correlation between human-based results and simulated results has been as high as r = 0.90. At the same time, most of us think that LLM responses are not themselves of interest to most work in experimental philosophy. This talk will explore how experimental philosophers can use simulated data for both good and evil and explore what roles simulated data might play in the future of xphi.