BRK560 Breakout Thu 19 Nov 11:30 PST
Hybrid Intelligence: Making It Real with Windows PCs
Mark Linton, Kurt Stevenson
Summary
From the recording
Microsoft presented hybrid intelligence as Windows routing tasks between local and cloud models. A local MAI Code 1.1 Flash model is in preview in the GitHub Copilot app, CLI and VS Code, with Adobe, Autodesk, NVIDIA, PwC and Insight piloting. Microsoft said 3-bit quantisation cut a 138 billion parameter model from 276GB to 53GB of memory, scoring 70.8% on SWE-bench Verified locally against 72.6% in the cloud. Microsoft Execution Containers let agent builders declare needs while policy sets boundaries, shown via Intune; adaptive containment is coming soon. Captions garbled some figures and names; savings figures are Microsoft's illustrative modelling.
Session description
Microsoft's catalogue description
Discover how hybrid intelligence puts each AI workload in the right place across local, shared and cloud compute. See how Windows AI PCs and on-premises inference complement the cloud, watch live demos of AI-assisted development and productivity scenarios, and explore real-world customer examples. Leave ready to assess workloads, size device capacity, build the investment case, and prototype your own agent.