01
▶ ANIMATED
00:00–05:00Cold OpenVERIFIED
Cold Open: Pause Demand
The panel opens on Bernie Sanders' letter urging top labs to pause AI, then pivots to compute becoming a financial asset.
▸ Bernie Sanders letter urges Anthropic, Meta, and OpenAI to pause AI development
✓ Confirmed — Sen. Bernie Sanders sent formal letters (10 Aug 2026) to OpenAI, Anthropic and Meta urging a pause, citing rogue models and AI-designed viruses.
Peter · Dave · Salim · Alex · Emad
02
05:00–10:00HealthspanVERIFIED
Healthspan XPRIZE Finals
The $101M Healthspan XPRIZE names finalists; ten teams win a million each to reverse functional aging by 2030.
▸ $101M Healthspan XPRIZE names 10 winning teams ($1M each)
✓ Confirmed, one correction — The $101M XPRIZE Healthspan did name 10 finalist teams at $1M each. But ~600 teams from 58 countries entered, not 800; Utah runs data-coordination, not a single 'finals'.
Peter · Salim · Alex
03
10:00–15:00HealthspanHOST TAKE
GLP-1, First Longevity Drug
GLP-1 drugs get crowned the first true longevity medicine and one of the largest-grossing drug classes ever.
▸ GLP-1 called a first longevity drug; retatrutide the next-gen name
⚠ Label the hype — 'First longevity drug' and 'largest-grossing drug in history' are marketing. The next-gen drug named is retatrutide (Lilly) — still Phase 3, not FDA-approved.
Peter · Dave · Salim · Alex
04
15:00–20:00HealthspanBOLD CLAIM
Chasing Escape Velocity
Alex speculates next-gen GLP-1s could reach 70% longevity escape velocity, while healthy lifespan still trails total lifespan by sixteen years.
▸ Claimed 70% longevity escape velocity for next-gen GLP-1s
⚠ Speculative — '70% longevity escape velocity' is a forward-looking opinion, not a measured result.
Peter · Alex · Dave
05
20:00–25:00AI CinemaBOLD CLAIM
The Two-Million-Dollar Feature
Higgsfield reportedly made a 110-minute feature for $2M with a small team in four weeks — a real film, an inflated revenue boast.
▸ Higgsfield: 110-min feature for ~$2M in 4 weeks; revenue boast overstated
✓ Film real, revenue inflated — Higgsfield's 110-min AI feature 'The Cully Hill Boys' is real. But revenue is ~$500M ARR (Sacra, Jun 2026), not '$700M in 1.5 years'; the company is ~3 years old.
Peter · Dave · Salim
06
▶ ANIMATED
25:00–30:00AI CinemaVERIFIED
Video Models Race Ahead
Nine of the top ten video models are Chinese; a new Future Vision XPRIZE dangles millions for a winning AI feature.
▸ Nine of top-ten video models Chinese; Future Vision XPRIZE finalists revealed Sep 25
✓ Mostly confirmed — 9 of the top 10 text-to-video models are Chinese (Artificial Analysis). Future Vision XPRIZE is real, but the purse is $3.5M total ($2.5M to the winner), not $10M; finalists present Sep 25.
Peter · Salim · Dave
07
30:00–35:00AI CinemaHOST TAKE
Demos Become Products
Dave predicts demos will become finished products within a year, as Alex wonders whether video models show emergent theory of mind.
▸ Prediction: demos become full products within one year
Peter · Dave · Alex
08
35:00–40:00AI CinemaHOST TAKE
Movies That Watch Back
Real-time generation points toward interactive, emotion-responsive films, and the panel debates licensing the likenesses of deceased actors.
▸ Real-time generation enables interactive, emotion-responsive films
Peter · Alex · Dave
09
40:00–45:00The Frontier RaceVERIFIED
Grok Undercuts The Frontier
Grok 4.6 matches frontier models at $2–6 per million tokens, far below premium pricing, on a rapid release cadence.
▸ Grok 4.6 confirmed at Artificial Analysis index 61, $2–6 per million tokens
✓ Confirmed — Grok 4.6 is on Artificial Analysis at index 61, $2 in / $6 out per million tokens — both figures match.
Peter · Dave · Alex · Emad
10
45:00–50:00The Frontier RaceBOLD CLAIM
Scaling Params, Selling Physics
Parameter counts climb toward the trillions, and Grok 4.7 is rumored trained on SpaceX physics — announced, not yet shipped.
▸ Grok 4.7 (~2T params) announced not shipped; SpaceX-physics training a claim
⚠ Announced, not shipped — Grok 4.7 (~2T params) is teased as imminent but hasn't shipped; 'trained on all SpaceX physics' is an unverified claim.
Peter · Dave · Alex · Emad
11
▶ ANIMATED
50:00–55:00Compute As MoneyVERIFIED
Wall Street Buys Compute
NVIDIA and six giant asset managers mobilize $500B for AI infrastructure, turning compute into a financed asset class.
▸ NVIDIA + Apollo, BlackRock, Blackstone, Brookfield, Goldman, KKR mobilize $500B
✓ Confirmed, plus one firm — NVIDIA's $500B financing is real; the consortium is six firms — Apollo, BlackRock, Blackstone, Brookfield, KKR and Goldman Sachs. Official term: 'credit backed by NVIDIA compute'.
Peter · Dave · Alex
12
55:00–60:00Compute As MoneyVERIFIED
The Price Collapse
Prices crater: DeepSeek's newest model runs for cents, and GPT-4-class power now fits on a single chip.
▸ DeepSeek V4 real; GPT-4-class now runs on one chip vs 16 A100s in 2022
✓ Real, with a note — DeepSeek V4 is real; GPT-4-class now runs on one chip vs 16 A100s in 2022. 'Mythos' isn't a garble — it's Claude Mythos 5, a real model. V4 Flash is nearer $0.14/$0.28 per M than $0.80.
Peter · Dave · Alex
13
60:00–65:00Compute As MoneyHOST TAKE
Compute's Mortgage Moment
Alex warns compute-backed credit echoes mortgage-backed securities: exponential progress could strand the value of the GPUs underneath.
▸ Compute-as-collateral echoes mortgage-backed securities (stranding risk)
Peter · Alex
14
▶ ANIMATED
65:00–70:00Beyond TransformersVERIFIED
Beyond The Transformer
A new architecture called Dragon Hatchling climbs ARC-AGI cheaply, though Alex dismisses it as a kitchen-sink hot mess.
▸ Dragon Hatchling (Pathway) climbs ARC-AGI on a fraction of transformer compute
✓ Confirmed — 'The Dragon Hatchling' (Pathway, arXiv 2509.26507, led by Adrian Kosowski) is a real post-transformer architecture making ARC-AGI progress at a fraction of the compute.
Peter · Alex · Dave
15
70:00–75:00The Pause LetterHOST TAKE
Sanders' Political Letter
The panel reads Sanders' pause letter as political positioning with no actionable demands; Alex says stop punishing intelligence.
▸ Panel calls Sanders' letter political positioning with no actionable demands
Peter · Dave · Alex
16
75:00–80:00The Pause LetterHOST TAKE
Too Late To Pause
Emad notes he signed a pause letter two years ago and it's too late now; Salim doubts nation-states can govern globally.
▸ Emad: signed a pause letter two years ago — too late now
Peter · Emad · Salim
17
▶ ANIMATED
80:00–85:00BiosecurityVERIFIED
AI Designs A Phage
King and Hie used Evo genome models to design bacteriophages; sixteen viable ones infected E. coli in the lab.
▸ Evo models (Arc/Stanford) designed ~285 phages; 16 viable infected E. coli
✓ Confirmed, credit corrected — Samuel King & Brian Hie (Arc Institute / Stanford) used Evo genome models to design phages: ~285 evaluated, 16 viable against E. coli (Science, 2025). 'Runs on a MacBook' is unsupported.
Peter · Alex · Dave
18
85:00–90:00BiosecurityHOST TAKE
Sequencers As Sentries
The panel proposes ubiquitous DNA sequencers as biodefense, with airport-vent sensors detecting pathogens at the speed of light.
▸ Proposal: ubiquitous DNA sequencers and airport-vent sensors as biodefense
Peter · Alex · Dave
19
90:00–95:00BiosecurityINTERLUDE
Defense In Depth
The biosecurity thread closes on a layered, defense-in-depth strategy for a world where the capability is already loose.
▸ layered biodefense and the limits of control
Peter · Dave · Alex
20
95:00–100:00ProvenanceHOST TAKE
Watermarking The Machine
The conversation turns to watermarking and labeling AI-generated content, and what reliable provenance would require.
▸ watermarking and labeling AI-generated content
Peter · Dave · Salim · Alex
21
100:00–105:00Personal ASIHOST TAKE
A Personal Superintelligence
The panel takes up the pitch for a personal superintelligence: everyone eventually holding their own powerful AI.
▸ the pitch for a personal superintelligence
Peter · Dave · Salim · Alex
22
105:00–110:00Personal ASIHOST TAKE
Power In One Hand
They weigh what it means to concentrate superintelligence inside a single company's vision.
▸ concentrating superintelligence in one company
Peter · Dave · Salim · Alex
23
110:00–115:00Personal ASIHOST TAKE
Whose Intelligence Is It
The debate asks whether personal AI truly empowers individuals or mainly deepens platform dependence.
▸ whether personal AI empowers people or platforms
Peter · Dave · Salim · Alex
24
115:00–120:00Flying CarsINTERLUDE
Flying Cars Return
The show shifts to flying cars and how eVTOL aircraft could reshape urban transport.
▸ flying cars and the future of urban transport
Peter · Dave · Salim · Alex
25
120:00–125:00Flying CarsHOST TAKE
Skyways Over Cities
They consider what aerial mobility would demand of cities: new skyways, infrastructure, and airspace rules.
▸ aerial mobility and city infrastructure
Peter · Dave · Salim · Alex
26
125:00–130:00Alignment & Q&AHOST TAKE
Alignment And Abundance
The final stretch turns to AI alignment and the abundance thesis, steering powerful systems toward shared benefit.
▸ AI alignment and the abundance thesis
Peter · Dave · Salim · Alex
27
130:00–135:00Alignment & Q&AHOST TAKE
Caution Versus Acceleration
The panel sits with the tension between caution and acceleration on the road to abundance.
▸ balancing caution and acceleration
Peter · Dave · Salim · Alex
28
135:00–140:00Alignment & Q&AINTERLUDE
The Audience Asks
The audience takes the microphone, putting its own questions about AI's future to the panel.
▸ audience Q&A on AI's future
Peter · Dave · Salim · Alex
29
▶ ANIMATED
140:00–145:00Alignment & Q&AINTERLUDE
A Note Of Abundance
The episode closes on a parting note of abundance and cautious optimism about human flourishing.
▸ closing vision of abundance
Peter · Dave · Salim · Alex