Sovereign Escape Velocity: Ownership w Open Models
summary
Gemma 4 claims disproportionate intelligence per parameter: a 31B on one GPU matches models twice to twenty times larger that need four or five GPUs. Mobile-targeted E2B and E4B actually use ~2B/4B GPU memory so they run on a Pixel phone. A medical fine‑tuned variant serves a hospital on two GPUs and parallel agents run on an M4 Mac via LM Studio. License shifts to Apache 2.0 enabling sovereign adoption in Ukraine Bulgaria and Brazil.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:19Proprietary models ownership limits
- 2:21Gemma 4 model family
- 3:44Hardware-efficient designs
- 5:11Gemma performance and cost
- 6:54Intelligence per parameter
- 7:59Open licensing for sovereignty
- 8:41Sovereign use cases
- 9:45Agentic tasks costs
- 11:49Hardware for local models
- 12:52On-device agent demo
- 14:59Enterprise scaling
- 16:22Local model demo
- 18:28Next steps
- 19:52Cost and infrastructure
- 20:13Summary
related talks
16mGemma 4: Compacting Intelligence for the Edge
49mOpen Models: The Secret Weapon for Next-Generation Software
14mGemma on mobile and web. Best and worst practices
09mRunning Gemma on Mac and Windows PCs with Ollama
05h 56mFrom the main stage to the terminal
57mBuilding GPU-accelerated multi-agent apps
09mOpen models and open compression
16mOpen models in the Gemini era