Open Source Project: Probing LLM Latent Structure via Anthropological Methods

No-Reserve2026 · reddit · 2026-08-19

A hobby project applies cultural anthropology methods, such as cognitive domain analysis (free-list, pile sorts), to Large Language Models. It investigates whether these techniques can reveal the underlying organization of data and biases within LLMs, despite the models lacking human culture. The project is a proof of concept, with full documentation on prompting, collection, and metrics available in a public GitHub repo.

Original post →

More from Research

Research channel →