Mixture of Experts CES 2026 AI highlights: NVIDIA Rubin & wild gadgets
5 snips
Jan 9, 2026 Chris Hay, a distinguished engineer specializing in AI architectures, Kaoutar El Maghraoui, a principal research scientist in AI hardware-software co-design, and Martin Keen, a master inventor, delve into cutting-edge announcements from CES 2026. They discuss NVIDIA's Rubin chip, which promises 5x performance and 10x inference savings, and explore the stunning gadgetry showcased. The conversation turns to Meta's bold $2B acquisition of Manus AI and its implications for enterprise productivity. They also dissect emerging AI training methodologies and the public's mixed feelings about AI governance.
AI Snips
Chapters
Transcript
Episode notes
Rubin Could Redefine Inference Economics
- NVIDIA's Rubin promises roughly 5x Blackwell inference performance and ~10x token cost reduction.
- That claim could reshape AI deployment economics and push workloads (esp. inference) at massive scale.
Prioritize Practicality Over Gimmicks
- Evaluate gadgets by practical value, not novelty; Kaoutar highlighted a stair-climbing Roborock and a bone-conduction 'lollipop' as examples.
- Prioritize predictability, privacy, and reliable offline behavior for consumer edge AI devices.
Watching An Agent Build A Presentation
- Martin Keen demoed Manus AI creating a full technical presentation by browsing, editing Markdown, and running terminal commands.
- He described feeling 'the icy hand of AGI' while watching it do his job much faster.