Anonymous16 days ago
A growing body of research shows that leading AI language models — including GPT, Claude, Gemini, and DeepSeek — perform significantly worse in non-English languages, with accuracy gaps of up to 38 percentage points for low-resource languages like Swahili. With 88% of the world's roughly 7,000 languages having virtually no usable training data, the AI boom risks deepening a global digital divide that leaves billions of people with inferior technology — or none at all.