Particle.news
Download on the App Store

Meta’s AI Chief Says Open‑Source Playbook Hit a Wall for Frontier Models

Wang’s admission signals a shift to keep higher‑risk models closed to manage safety concerns as Meta pursues paid AI products.

Overview

  • Alexandr Wang told Bloomberg that Meta’s prior open‑source approach no longer fits its most advanced models and that Muse Spark was kept proprietary after early training flagged elevated risks including bio hazards.
  • Muse Spark remains mainly inside Meta’s apps with a limited private API rollout and has been praised for visual understanding while Wang acknowledged it trails rivals on coding tasks.
  • The release of Llama 4 drew reports of mixed quality and bugs that Meta attributed to software issues during rollout.
  • Meta has sharply increased AI and data‑center spending plans, reportedly raising 2026 capital expenditure guidance to as much as $145 billion, even as the company recruited talent from top labs and trimmed roughly 600 AI‑unit roles.
  • To reduce reliance on advertising revenue, Meta is testing paid AI tiers on Instagram, Facebook and WhatsApp plus a separate chatbot subscription, and investors are watching whether those products and closed models will produce new revenue after a roughly 19% share decline.