Why Language Models Struggle With African Languages
The gap is not linguistic difficulty. It is that the text these systems learn from was scraped from a web that barely contains these languages.
AI
What AI Is Actually Being Used For, Once You Look Past the Announcements
The deployed applications are less exciting and more useful than the ones that get launch coverage: document handling, triage, translation and fraud detection.
The Compute Question Nobody Wants to Answer
Training capacity is concentrated in a handful of countries. What that means for everyone else is a question of strategy, not sentiment.