Cohere Labs head: multilingual AI's 'longitude problem' — scale isn't enough
Cohere_Labs · x · 2026-09-03
In a World Summit AI interview, Cohere Labs head Marzieh Fadaee describes multilingual AI's "longitude problem": English is the majority language on only 49.5% of identifiable web content yet remains frontier labs' default proxy for intelligence; 7,159 languages are in active use while even ambitious multilingual models cover 100. Leaderboards can hide poor performance on underrepresented languages, languages aren't just wrappers around intelligence, and cultural context matters as much as fluency — scaling alone won't close the gap. Fadaee took over Cohere Labs in September 2025 after a decade in machine translation research.
More from AGI Musings
- Altman tells G20 ministers rejecting AI is like rejecting electricity; critics see conflict of interest — heypearlai · 2026-09-03
- Three-Paper Series Decomposes Human-Like RSI into ASPIRE, S³Gym and HarnessDev — teortaxesTex · 2026-09-03
- Aristotelian virtue ethics is basically an RLVR regime, argues one poster — curious_vii · 2026-09-03
- davidad vs Richard Ngo: Agent Foundations First, or Institutions First? — davidad · 2026-09-03
- Only 1% of 379 L&D job postings are junior — AI is erasing the entry-level rung — mattbeane · 2026-09-03
- Solo builder crawls 10B+ records a day to put the internet in an analytical database via MCP — AaronBergman18 · 2026-09-03