Language is one of the last boundaries on human connection — and translation blurs it. This talk covers how Meta produces and distributes AI voice-translated Reels at scale: streaming speech translation that improves accuracy and latency at once, and the retrieval-side exploration that solves cold start so a dubbed render actually finds its audience.

Wally is a self-learning loop of agents - an optimizer, verifier and learner loop that identifies opportunities to save CPU or GPU cycles, verifies it for correctness, safety, etc, runs a public A/B test with the solution and then launches a win. The fundamental component is the self-learning loop that updates the harness based on wins/fails. Wally has run 1000s of public tests with a 30% success rate and has saved several millions of dollars.

To be announced

The best AI media models have crossed the photorealism threshold. The real competitive advantage now lies in how rapidly a model recipe can adapt to product and user feedback.

Log files and traditional metrics solved evaluation for LLMs, but visual breaks those frameworks.
A video generation pipeline can run with zero software errors, perfect resolution, and flawless prompt adherence, yet still fail completely. The camera crop might feel off, the motion might stutter, or a ‘cinematic lighting’ edit might accidentally turn skin into unnatural plastic.
Visual quality is fundamentally perceptual.

This is where Visual Judges - powered by Visual Language Models (VLMs) - enter the stack.

Novel innovations are driving a flywheel that can accelerate products with media at the center!

Machine learning is opening a new chapter in video compression, with the potential to improve quality, reduce bandwidth, and enable better video experiences across diverse devices and network conditions. But taking ML-powered codecs from research to real-world products requires navigating tough tradeoffs in latency, compute cost, hardware acceleration, compatibility, and quality measurement. This talk explores the product and engineering considerations behind scaling ML video codecs, and what it takes to turn promising models into durable product impact.

When you're creating Media in Edits, it should just work — you hit preview and it plays back instantly, captions land accurately, no hiccups while you edit your media. That's been our north star, and we want to share what we did to get there. We improved creation reliability across the full stack — iOS, Android, and the server-side — and users can feel the difference. The big unlock? We built AI agents that basically run our reliability loop for us. They catch regressions, triage the alerts, and trace issues across client and server all the way down to the offending diff — often before the problem even hits production. The results speak for themselves: preview playback failures improved by 88%, which translates to a noticeably smoother creation experience for millions of people.

Deploying AI video at scale often introduces severe infrastructure bottlenecks and cost challenges, driven primarily by continuous data shuttling between discrete GPUs and host CPUs. By leveraging latest-generation Intel® Xeon® processors with Intel® Advanced Matrix Extensions (Intel® AMX), organizations can eliminate this overhead—executing compute-intensive workloads such as Video Super-Resolution, inverse tone mapping, and bitrate optimization directly on the CPU alongside encoding pipelines.

Building on this efficient compute foundation, the Intel® Media AI Framework transforms raw media into trusted, searchable AI-ready intelligence, while simplifying deployment of media workloads. Through agentic orchestration, intelligent model routing, reusable MCP tools, and video RAG, the framework powers configurable workflows like semantic search, summarization, metadata enrichment, contextual ad placement, and other MCP-based agent flows. This session demonstrates the larger value of bringing Xeon-based acceleration through the Intel® Media AI Framework to improve user experience, convert media into actionable intelligence, and empower customers to build and scale cost effective media AI pipelines optimized for Intel® Xeon®.

An exciting fireside chat with Meta senior leaders.