AI BriefingMeta Engineering news
The latest AI and developer news from Meta Engineering, with short summaries.
Visit website →Privacy-Aware Infrastructure in the AI-Native Era: An Asset Classification Case Study
Meta operates a hybrid asset classification system combining LLMs and deterministic rules to address the complex data privacy challenges of AI-native environments.
SilverTorch: Index as Model — A New Retrieval Paradigm for Recommendation Systems
SilverTorch is a new paradigm that unifies microservice-based recommendation systems into a single neural network, maximizing retrieval efficiency and accuracy.
Reel Friends: Building Social Discovery at Billion-User Scale
Meta released a Reels podcast covering the development process of Friend Bubbles.
Modernizing Facebook Groups Search to Unlock the Power of Community Knowledge
Facebook Groups search quality was improved through hybrid retrieval and automated evaluation.
Meta's Capacity Efficiency: How Unified AI Agents Optimize Hyperscale Performance
Meta used common tools and skills to automate the detection and fixing of performance issues with AI agents.
How Meta Used AI to Map the Tribal Knowledge of a Massive Data Pipeline
50+ AI agents organized the tribal knowledge of 4,100 files into 59 context files.
KernelEvolve: How Meta's Ranking Engineer Agent Optimizes AI Infrastructure
KernelEvolve automatically optimizes AI kernel performance by exploring hundreds of candidates.
Meta Adaptive Ranking Model: Rewriting the Inference Scaling Curve to Bring LLM-Scale Models to Ads
Meta scaled ad ranking to LLM scale through request routing and hardware co-design.
AI for American Cement and Concrete
Meta released BOxCrete, a model that optimizes concrete mix design, along with the source data behind it.
Friend Bubbles: Enhancing Social Discovery in Facebook Reels
By reflecting friends' reactions in Reels recommendations, it boosted both discovery and conversation together.
REA, the Autonomous AI Agent Accelerating Meta's Ad Ranking Innovation
Meta's REA autonomizes ad ranking ML experiments, boosting accuracy and productivity.
Patch If You Can: AI Codemods for Secure-by-Default Android Apps
Meta automates large-scale security migrations with a secure-by-default Android framework and generative AI.
RCCLX: Innovating GPU Communication on AMD Platforms
Meta has open-sourced RCCLX, an AMD extension of RCCL, boosting performance with DDA and low-precision collectives.
The End of Traditional Testing: Agentic Development Broke 50-Year-Old Testing, and JiTTesting Might Revive It
LLMs generate tests on the fly to catch regression bugs right after code changes.