Read the engine, not just the headline model. Baidu's grant US11922287B2 (“Video recommendation with multi-gate mixture of experts soft actor critic,” issued 2024-03-05) is about the recommendation machinery that actually monetizes a platform. Assigned to Baidu USA, LLC and classified CPC G06N 3/042, it applies multi-gate mixture-of-experts to video recommendation.
The segment logic is that recommendation is where attention becomes revenue. On a large content platform, the system deciding what to show next drives engagement, which drives ad inventory and value. Mixture-of-experts with multiple “gates” lets one system balance competing objectives — watch time, satisfaction, diversity — efficiently, which is exactly what a monetization engine needs.
“Described herein are embodiments of a reinforcement learning based large-scale multi-objective ranking system. Embodiments of the system may be used for optimizing short-video recommendation on a video sharing platform.”— U.S. Patent No. 11,922,287 source
Baidu reports AI across its search, cloud, and content businesses, and recommendation quality feeds the advertising line without being broken out as its own technique-level economics. The grant is the granular record under that revenue: dated 2024, owned, aimed at the recommendation architecture behind engagement.
The discipline: a grant proves invention and ownership, not a revenue figure, and we attribute none. It also doesn't establish deployment in a specific product surface. What it documents is dated IP behind the recommendation engine that platform revenue depends on.
For the business reader, the reusable point is that the flashiest AI — chatbots, frontier models — isn't always where the money is. Recommendation systems have monetized AI for years, and a 2024 multi-gate MoE grant is a primary document showing that engine is still where serious architecture work goes.