AI Audit Finds Llama-3 Fails 89% of Queries on Global South Tech Governance
A massive new study reveals AI's knowledge is dangerously skewed toward wealthy nations.
Deep Dive
A global audit of Meta's Llama-3 8B model reveals severe geographic bias in AI knowledge. Stress-tested with 1,704 queries across 213 countries, the model could only provide factual responses for 11.4% of questions related to technical governance. The findings show AI's technical knowledge is heavily concentrated in higher-income Global North regions, creating systemic information gaps and risking unreliable data for policymakers in the Global South.
Why It Matters
This bias means AI could mislead global policymakers and worsen existing geopolitical inequalities in tech governance.