Skip to content

Explainers

Articles

LLM Targeted Underperformance: Why AI Chatbots Give Worse Answers Based on Who You Are

MIT researchers found that leading AI chatbots systematically give less accurate answers to users with lower education, non-native English, or non-US origins. A viral car wash reasoning test highlights a related weakness on one fixed prompt: 42 of 53 models chose walk on a single run, and only 5 answered correctly on all ten consistency runs, with no user biography in that benchmark.

Artificial Intelligence Explainers