Meta CTO Andrew Bosworth told Big Technology Podcast that a state-of-the-art model alone does not win the AI race. His argument: the advantage goes to whoever combines model quality with product, distribution, and consumer experience. Meta, he says, is the only company with all four. Apple, Anthropic, OpenAI, and Google each have one.
Meta is renting frontier models from competitors while it rebuilds its own pipeline. Bosworth admits the Llama 4 development cycle inadvertently killed the base model pipeline by overcommitting resources to Llama 3. The company also just shipped Muse Spark 1.1, a new agentic model it claims is benchmark-competitive at lower cost than rivals.
The full interview goes deeper than the strategy pitch. Bosworth discusses the specific cultural and organizational failures that set Meta's frontier research back, the company's wearable hardware bets, and what he means by 'personal superintelligence.' The gap between Meta's self-assessment and its recent model output is the tension worth reading for.
[READ ORIGINAL →]