Sep 19, 2026

Operating Distributed Inference Systems at Scale — Nishant Gupta & Naman Ahuja, Meta

Meta's Nishant Gupta and Naman Ahuja explain how AI inference, the company's fastest-growing workload, requires a dedicated orchestration layer akin to Kubernetes for virtual machines. They detail the unique challenges of LLM serving compared to traditional microservices, emphasizing the intricate coupling of infrastructure layers, the distributed transaction nature of requests, and a seven-axis framework for scheduling. The talk concludes with the imperative to optimize for "cost per successful task" and the necessity of an inference-specific control plane to manage the complex interplay of latency, cost, and throughput.

Sep 08, 2026

Why 95% of All AI Projects Fail - Boris Scharinger | Podcast #174

Boris Scharinger, Industrial AI Strategist at Siemens and author of "Industrial AI: From Pilot to Profit," discusses why 95% of AI proofs of concept fail in industrial settings. He highlights the 5-10x cost multiplier from pilot to production, the critical role of the "harness" around AI models, and the unique challenges of data access and legacy systems in physical environments. Scharinger advocates for a two-tiered investment strategy, separating foundational data capabilities from use-case-specific ROI, and addresses cultural resistances from engineers. He provides guidance on differentiating commodity from strategic AI applications and how to effectively scale industrial AI for profit.

Aug 19, 2026

Stateless, Yet Durable: MCP Tasks v2

Cornelia Davis explores the evolution of MCP Tasks, detailing the transition from v1 to v2. She highlights how v2's stateless protocol design supports durable, long-running agentic workflows by shifting durability responsibilities to the client and simplifying elicitation. The presentation includes a demo of a purchase order process and discusses future scalability enhancements through a notification-based approach.

Aug 18, 2026

Training Krea 2: What matters in generative model training — Sangwu Lee, Krea.ai

The talk by Sangwha Lee discusses Krea 2's open-source medium variant, emphasizing stylistic diversity and rapid iteration over the consistency-focused approach of larger models. A significant portion details their robust data curation pipeline, including unique methods for deduplication, filtering out AI-generated images, utilizing sparse autoencoders for unsupervised tagging, and ensuring world knowledge coverage. He outlines an LLM-inspired multi-stage training process, culminating in a prompt expander, and shares insights on fast iteration and future directions for image generation, highlighting the increasing integration of VLM advancements.

Aug 10, 2026

Taking Reinforcement Learning Cross Datacenter — Nan Jiang, Modal

Nan Jiang introduces "Adam absorption" to revolutionize RL model synchronization. By exploiting finite precision serving and small Adam steps, less than 1% of served model weights actually change, allowing for 500MB patches instead of 500GB checkpoints. This enables a distributed "bulletin board" architecture, decoupling trainers from global rollout fleets and unlocking elastic, cross-region GPU capacity for RL.

Jul 28, 2026

Serving 2 Million Models Without Melting: Scaling the Hugging Face Hub — Arek Borucki, Hugging Face

Arek Borucki details how Hugging Face scales its infrastructure to serve millions of models and users, focusing on the evolution of search architecture using MongoDB Atlas and Apache Lucene, robust database scaling with a seven-node cluster and sharding, and dynamic frontend autoscaling with Kubernetes and KEDA to ensure an instant, seamless user experience.