Markdown source

Slides: Sovereign Escape Velocity: Ownership w Open Models — Gus Martins, & Ian Ballantyne, Google DeepMind

Source Video

Sovereign Escape Velocity: Ownership w Open Models — Gus Martins, & Ian Ballantyne, Google DeepMind

Relationship To World's Fair 2026

These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.

Related Scheduled Sessions

• No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck.

Extracted Slides

• Recreated text/layout view: open HTML recreation
• AI slide classifier: content_slide confidence 0.99
• Text source: advanced OCR rapidocr-live/border-trim/opencv-adaptive.
• OCR decision: ready — Chart slide with small labels and numeric bars; OCR is likely more reliable than manual transcription.

Slide text:

Gemma4

Arena Elo Score

AIE

31B26B754B 61008397B:6858

Engineering the future ofAl

: AlEngg

• Recreated text/layout view: open HTML recreation
• AI slide classifier: content_slide confidence 0.96
• Text source: agent_vision.
• OCR decision: ready — Product/UI screenshot with dense small interface text; OCR is likely more reliable than manual transcription.

Slide text:

ai.dev

• Recreated text/layout view: open HTML recreation
• AI slide classifier: content_slide confidence 0.98
• Text source: agent_vision.

Slide text:

Efficiency & "Intelligence-per-Parameter"

Gemma 4 family:

E2B, E4B, 26B A4B, 31B.

Extreme parameter efficiency:

max("intelligence-per-parameter")

Performance equivalent to models up to 10x their size on targeted logic tasks.

• Recreated text/layout view: open HTML recreation
• AI slide classifier: content_slide confidence 0.98
• Text source: agent_vision.

Slide text:

The Agentic "Thinking Tax"

Autonomous loops, tool calls,

and self-correction dominate

current application demand.

Iterative agentic workflows

consume 5–9x more tokens than

standard chat.

Relying purely on pay-per-token

"Leased Intelligence" creates

heavy scaling operational liabilities.

• Recreated text/layout view: open HTML recreation
• AI slide classifier: content_slide confidence 0.98
• Text source: agent_vision.

Slide text:

Personal & Edge

Personal (NPU/Mobile): Shifting execution to

local "sunk-cost" hardware to maintain local

data by design.

Edge (Desktop/Single-GPU): Establishing

fixed-cost reasoning on a 24GB-80GB

footprint.

Battery Priority: On mobile, power utilization is

part of the cost and is arguably

more critical than raw token generation cost.

• Recreated text/layout view: open HTML recreation
• AI slide classifier: content_slide confidence 0.9
• Text source: none.
• OCR decision: ready — Dense product/UI screenshot with small text; OCR is likely more efficient than manual transcription.
• Slide text: not surfaced (illegible by AI classifier).

• Recreated text/layout view: open HTML recreation
• AI slide classifier: content_slide confidence 0.91
• Text source: agent_vision.
• OCR decision: ready — Dense slide-editor screenshot with thumbnail sidebar and UI chrome; OCR is likely useful for the embedded slide content.

Slide text:

Gemma 4 demos

Hidden Non-Slide Evidence

• `slide-001.jpg` — speaker_stage confidence 0.98; Stage photo with audience and podium; not a readable presentation slide.
• `slide-002.jpg` — title_card confidence 0.97; Speaker intro card with headshots and names; not a content slide.
• `slide-003.jpg` — title_card confidence 0.95; Title/logo slide with minimal content; not a substantive presentation slide.
• `slide-004.jpg` — title_card confidence 0.95; Title/logo slide with minimal content; not a substantive presentation slide.
• `slide-012.jpg` — title_card confidence 0.99; End card/logo slate with only branding and URL; no substantive presentation content.

Classification audit: raw/sources/slide-ai-classification/slides/SS-A8sE7hkw/audit.json

Slide-Derived Subjects To Review

Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.