Large Language Models Naively Recover Ethnicity from Individual Records

Name: Large Language Models Naively Recover Ethnicity from Individual Records
Uploaded: 2026-10-17T07:34:54.716571+00:00
Duration: 421 s
Description: Large language models can accurately infer ethnicity from names using only name and location data, outperforming traditional methods like Bayesian Improved Surname Geocoding (BISG) without needing extra training data. This allows for ethnic classification outside the US and into specific contexts like religious sect or caste, reducing income bias inherent in older systems.

Large language models can accurately infer ethnicity from names using only name and location data, outperforming traditional methods like Bayesian Improved Surname Geocoding (BISG) without needing extra training data. This allows for ethnic classification outside the US and into specific contexts like religious sect or caste, reducing income bias inherent in older systems.

Open the full episode