Source: Financial Times (paywall)
Meta is shifting the labor economics of content moderation—a historically expensive, human-intensive operation—toward language models at scale, targeting 50% automation within two years and 90% by late 2026. This move compresses a timeline that seemed years away just months ago. The shift reflects both confidence in LLM reliability for nuanced judgment calls and pressure from Wall Street to cut the $15+ billion annual content moderation budget. The test isn't whether AI can flag obviously illegal content, but whether it can handle the gray zones—hate speech in context, satire, regional norms—where Meta currently relies on thousands of contract workers whose expertise and local knowledge may prove difficult to replicate.