Are Large Language Models (LLMs) still WEIRD?

Most Large Language Models (LLMs) used in enterprise settings are trained on data that has ‘WEIRD’ characteristics (Western, Educated, Industrialised, Rich, Democratic).

A 2023 Harvard study found the greater the cultural distance between a country and the USA, the less accurate ChatGPT got at simulating peoples’ values.

STRAT7’s new research tests three LLMs on attitudes, values, and moral dilemmas, benchmarking against real data from the UK, France, and Malaysia. 

This session examines the results, the implications for model choices and necessary mitigations for delivering credible global insights.