Meta engineers reveal how a seemingly simple 'Friend Bubbles' feature required complex ML models and cross-platform behavioral insights, uncovering design principles for large-scale social recommendation systems.
A bilingual LLM trained with semantic IDs as vocabulary tokens can recommend items and be steered through natural conversation.
Replace random hash IDs with semantic tokens so LLMs can natively understand items and enable conversational recommendations.
Meta introduces the 'Index as Model' paradigm, unifying all recommendation retrieval functions into a single neural network, achieving 23.7x higher throughput and 20.9x better cost efficiency.