DeepMind's Whisper Study, The Trillion-Dollar Capex Graph, AI Industry Definitions, Monk to Engineer cover

DeepMind's Whisper Study, The Trillion-Dollar Capex Graph, AI Industry Definitions, Monk to Engineer

AUGUST 21, 2026 · 2h 6m
WATCH

A former Buddhist monk turned engineer joins The Great Spin Up this week, and there's an early game worth stealing: ask your AI "what is my main problem?" and watch it read you like a horoscope. News covers Guidelite's safety report card (Anthropic and OpenAI at C+, Meta an F), a DeepMind study of 10,000+ people showing AI persuades best when it whispers, and a graph putting data center capex — $930 billion over six years — above the railroads, Apollo, and the Manhattan Project. Builds: a three-day, fifty-dollar AI sci-fi trailer made with Veo 3.1, Kling 3, and Lyria, plus Weave, a programmable Zoom for meetings that don't feel dead. The back half is a definitions deep dive — air gapped, distillation, quantization, LoRA via Dr. Seuss. It ends with a provocative question: if AI can compress a hundred years of development into ten, how do models trained on English internet data end up serving everyone else?TIMESTAMPS

(06:44) News

(40:12) Building

(01:20:00) Open question — East and West

(01:29:11) Deep dive — definitions

(02:00:03) AMA — "What is your main problem?"

(02:09:13) ClosingRESOURCES

PEOPLE

Andrew Davis — guest; monk turned engineer

Jeff Dean — left Google after 27 years

Marie Curie — radioactivity containment parable

BOOKS

Sapiens — Yuval Noah Harari

Superintelligence — cited as Max Tegmark [check ~11:45]

Thing Explainer — Randall Munroe

PODCASTS / SHOWS

Moonshots — Alvin Graylin on China's AI

Star Trek — protopian seed

The Boat That Rocked — tonal inspiration

The Creator — visual inspiration

PROJECTS / COMPANIES / TOOLS

Claude Code — coding engine

Conductor — parallel session orchestration

CUDA kernel agent — Tsinghua's; beats Opus 4.5 [unclear — check ~36:20]

DaVinci Resolve — trailer editing

Deepgram — transcription

Discovery Loop — lean four-person startup

FAL.ai — multi-model media API

G Stack — Gary Tan's YC agent skills

Google Antigravity — the IDE

Google Lyria — one-shot film scores

Google Veo 3.1 / Flow — video generation

Guidelite — AI safety ratings

Hermes — personal agent system

Kimi K3 — research model

Kling 3 — Chinese video model

LiveKit — video call infrastructure

Llama — Meta's open-source model

LTX — video with voice

Neon — cloud Postgres

OpenRouter — model switchboard; $7B+ Stripe deal

Svelte — the JavaScript framework

Weave — Living Code's programmable Zoom

FRAMEWORKS / CONCEPTS / LAWS

AGI / ASI / RSI — squishy acronym ladder

Allostasis — feelings as energy math

Computational commons — neighborhood AI

Distillation — small models trained off big ones

LoRA — cheap low-rank fine-tuning

Mixture of experts — routed specialists

Open weight vs open source — weights only vs reproducible

Quantization — fewer bits per parameter

Speed of containment — guardrails pace the frontier

PAPERS / ARTICLES / ESSAYS

Capex graph — Rutger Bregman / Brookings data

DeepMind persuasion study — 10,000+ people, US/UK/India

What Patterns Want — Reese Lindmark

OTHER

AI Builders — Lisbon builder events

Future Vision XPRIZE — protopian trailer contest

Kinship One — the trailer's vessel

Read transcript →