Moonshots · Visual Reel · EP 294 · recorded live 2026-09-25 · aired 2026-09-28 · runtime 51:37

The Second Unveiling

Episode 294 — Recorded live at Moonshots Live, Los Angeles — the show's first live-audience event. The five mates open the week AI got a rebrand: 'superintelligence' goes official at the UN podium, Jensen Huang and Mark Zuckerberg answer the doomers, Bernie Sanders proposes a ban with prison terms, a frontier model drives a real Toyota Corolla through a cone course, and 950 Claude agents surface a new enzyme class in 21 hours. Eleven beats, every tentpole run against primary sources.
How to read this page: Hosts disclose their own stakes on air: Dave Blundin is chairman of EverQuote (discussed during the Muse sell-off segment), Peter Diamandis's own XPRIZE Foundation runs the $101M Healthspan prize he cites, and Alexander Wissner-Gross founded a company named Physical Superintelligence while arguing for the term — treat those segments as self-reported.
01 · The Room Where Magic Happens02 · Answering the Doomers03 · Cities, Not Cartels04 · Nations Give Way05 · Superintelligence, Officially06 · The Ban That Leaks07 · A Model Every Two Days08 · Code Becomes the Canvas09 · The Corolla Test10 · Nine Hundred Fifty Searchers11 · Biology Becomes a Search
VerifiedEarly signalHost takeBold claimInterlude
01
The Room Where Magic Happens
0:00–4:50Cold Open: Choose Your MoonshotINTERLUDE

The Room Where Magic Happens

Live from Los Angeles: five technology optimists open their first stage show, urging everyone to pick one big problem and build.
▸ Moonshots Live inaugural event; Peter: 'AI is not a hand grenade, it's a jet pack' — find three people tonight and speak your moonshot out loud.
Peter Diamandis
02
Answering the Doomers
4:50–9:10Why the Doomers Are WrongVERIFIED

Answering the Doomers

Jensen Huang, boss of chipmaker Nvidia, and Facebook founder Mark Zuckerberg push back on famous scientists who predict AI catastrophe.
▸ Jensen Huang, on the '10% chance of societal destruction' he attributes to Geoffrey Hinton: the figure is 'not grounded in science, it is not grounded in research... just because it comes from a scientist, that doesn't make it scientific.'
✓ Huang did say this, on the New York Times' Ezra Klein podcast (Sep 23, via TNW): Hinton's 10% is 'not grounded in science or research.' The sourced record shows him faulting one Hinton prediction — the call to stop training radiologists — not Hinton's whole record, and the 10% is Huang's characterization, not a sourced Hinton quote.
Peter Diamandis, Emad Mostaque, Alex Wissner-Gross
03
Cities, Not Cartels
9:10–14:25Liability, Cartels, and CitiesHOST TAKE

Cities, Not Cartels

The panel says AI makers must stay liable for harm, and grow safety alongside AI, the way cities grew fire brigades.
▸ Scott Bessent, US Treasury Secretary: 'we're not gonna waive liability' — the labs are liable for their product if it does harm; AWG: the past two to three weeks saw 'an attempted formation of an AI safety cartel.'
Salim Ismail, Dave Blundin, Peter Diamandis, Alex Wissner-Gross
04
Nations Give Way
14:25–19:55Xi, Trump, and the City-StateVERIFIED

Nations Give Way

Can America and China trust each other on AI? The panel argues power is drifting from big nations toward self-reliant cities.
▸ Xi and Trump met at the White House September 23–25 and agreed to meet again in November — standing up a formal 'US–China Super Intelligence Dialogue' with a proposed SI-incident hotline modeled on Cold War nuclear channels.
✓ Axios and CSIS (Sep 26) confirm the White House summit ran Sep 23–25 and produced a November follow-up — including a formal 'US–China Super Intelligence Dialogue' and a proposed incident hotline; analysts called the AI outcome thin, 'a promise to talk again in a few weeks.'
Emad Mostaque, Salim Ismail, Alex Wissner-Gross, Dave Blundin
05
▶ ANIMATED
19:55–25:45AI Gets a RebrandVERIFIED

Superintelligence, Officially

In a UN address, the US renames advanced AI 'superintelligence' in its own documents. The panel debates whether the new name calms public fear or just moves the target.
▸ In his UN General Assembly address, Trump said US government documents will henceforth use 'super intelligence, SI' in place of 'artificial intelligence,' rejecting any global scheme of control; Peter: 'the US won the race to superintelligence by rebranding it.'
✓ Fox News, Korea Times and Fortune (Sep 23) confirm the clip is close to verbatim — but the venue was Trump's UN General Assembly address in New York, not DC, and a rename in US government documents is not the same as the world adopting it.
Dave Blundin, Emad Mostaque, Peter Diamandis, Salim Ismail, Alex Wissner-Gross
06
The Ban That Leaks
25:45–31:35The Sanders BanVERIFIED

The Ban That Leaks

Senator Bernie Sanders, a veteran US politician, proposes banning superintelligence with prison terms. The panel calls it unworkable and self-defeating.
▸ Sanders plan: a permanent ban on artificial superintelligence, an immediate pause on advanced AI, a cabinet-level Department of AI, a corporate death penalty, and 20 years in federal prison, modeled on illegal nuclear weapons law.
✓ Sanders' own press release (Sep 3) confirms every element: a permanent superintelligence ban, a pause on advanced AI pending federal safety rules, a cabinet-level AI agency, a corporate death penalty and up to 20 years in prison. One fix: the bill predates the Sep 23 rename by three weeks — it was not a response to the rebrand.
Peter Diamandis, Dave Blundin, Alex Wissner-Gross, Salim Ismail, Emad Mostaque
07
A Model Every Two Days
31:35–36:15Model Avalanche and Meta MuseEARLY SIGNAL

A Model Every Two Days

New AI models now land roughly every day or two. Meta, Facebook's parent company, launches Muse, a phone assistant whose launch-week downloads outran ChatGPT's debut.
▸ Four releases in five days holds up: Grok 4.7 (Sep 21), then Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna (Sep 22) — trackers logged ~25 new models in September; Meta's Muse hit number one on the US App Store with 900,000+ downloads in six days (Bloomberg); Meta's apps reach 3.6 billion people daily; Anthropic says Opus 5.5 matches Fable 5.1 on most work at about 40% lower running cost.
⚠ LLM Gateway's timeline (Sep 2026) supports the pace — ~25 models from 16 providers in the month, with Grok 4.7 (Sep 21) then Opus 5.5, GPT-6 Sol and Luna (Sep 22). Corrections: the tape's 2.8M Muse downloads is unsourced — Bloomberg (Sep 21) reports 900,000+ in six days; Meta's own Q2 figure is 3.60B daily people, not 3.8B; Anthropic (Sep 22) says ~40% lower running cost, not half price.
Peter Diamandis, Dave Blundin, Alex Wissner-Gross, Emad Mostaque, Salim Ismail
08
Code Becomes the Canvas
36:15–39:55The Procedural TurnBOLD CLAIM

Code Becomes the Canvas

The panel says hallucinations are collapsing with scale — and one host reports the newest Claude builds videos and games by writing code instead of painting pixels.
▸ Dave: hallucination error rates 'just plummet with pure scale'; kids starting on GPT-5 and 6 will trust it because 'it's right 99.9% of the time.'
⚠ Anthropic's own Opus 5.5 launch post (Sep 22) claims no procedural video, animation or image generation — that is co-host Alex Wissner-Gross's hands-on observation — and the 'right 99.9% of the time' hallucination figure has no published source.
Alex Wissner-Gross, Peter Diamandis, Dave Blundin, Emad Mostaque
09
▶ ANIMATED
39:55–43:55DrivingBenchVERIFIED

The Corolla Test

Researchers handed a general-purpose AI the controls of an ordinary Toyota sedan. It drove a full cone course perfectly on try two.
▸ On DrivingBench, GPT-6 Astra drove a software-enabled Toyota Corolla through a ~130-meter parking-lot cone course and 'finished the entire course 100% on its second try' — the only model to complete it (Claude Fable 5.1 peaked at 45%, Grok 4.6 at 11%).
✓ OfficeChai's DrivingBench report (Sep 24) confirms it: GPT-6 Astra finished the ~130m cone course 100% on its second try, the only model to complete it (Fable 5.1 topped out at 45%, Grok 4.6 at 11%). It is a parking lot at low speed, not a road; the tape's separate '60 GW' energy shortfall has no published source and was removed.
Peter Diamandis, Alex Wissner-Gross, Salim Ismail, Emad Mostaque, Dave Blundin
10
Nine Hundred Fifty Searchers
43:55–48:00The Enzyme HuntVERIFIED

Nine Hundred Fifty Searchers

Anthropic, a leading AI lab, set 950 AI agents loose on a huge DNA library. They surfaced a new gene-editing-style enzyme.
▸ 950 Claude agents searching a massive DNA database for 21 hours found a previously unknown enzyme system, since named array-associated reverse transcriptases (ARTs), with a repeating pattern similar to CRISPR.
✓ Anthropic's own post (Sep 23) confirms ~950 Claude agents, ~21 hours and ~210 million tokens scanning some 200,000 reverse transcriptases, surfacing array-associated reverse transcriptases (ARTs) beside CRISPR-like repeat arrays. Humans did all the wet-lab work, and Anthropic does not yet claim to know the system's function.
Peter Diamandis, Alex Wissner-Gross, Salim Ismail, Emad Mostaque
11
Biology Becomes a Search
48:00–51:30Millennium Problems for BiologyVERIFIED

Biology Becomes a Search

Scientists publish a grand-challenge list for biology — regrow limbs, explain life's origin — betting AI plus lab robots can solve them.
▸ Edison Scientific and Future House released 'millennium problems for biology' — hard to solve, verifiable in a standard wet lab in days; the XPRIZE Healthspan purse is $101 million.
✓ Sam Rodriques' own post (Sep 19, FutureHouse / Edison Scientific) confirms the 12-problem list — origin of life, cryopreservation, limb regeneration — each verifiable in a standard wet lab in days; age reversal was deliberately excluded, not missed. The $101M Healthspan purse checks out against XPRIZE's guidelines, but XPRIZE is host Peter Diamandis's own foundation — self-reported on that leg.
Peter Diamandis, Emad Mostaque, Dave Blundin, Alex Wissner-Gross, Salim Ismail
The scout's ledger

The opportunity board

This week's opportunity board lives on Reel 296 — fourteen signposts drawn from all three Moonshots Live reels (294, 295 and 296), with one scout's pick and one reality check. Nothing there pulls an engineering hour off the P0.

One plate for roughly every five minutes · 11 frames · 2 animated.
Green notes come from a live verification pass against primary sources — 9 beats carry them. This week's corrections, in green where he'll see them: the rename came at the UN General Assembly in New York, not DC, and covers US government documents; the Sanders ban bill predates the rename by three weeks; the 10% doom figure is Huang characterizing Hinton, not a sourced Hinton quote; Muse's anchored download figure is Bloomberg's 900,000+ in six days, not 2.8 million; Meta reaches 3.60 billion people daily, not 3.8; Opus 5.5 runs about 40% cheaper, not half price; the '60 gigawatt' energy shortfall has no published source and stays off this page; the Corolla course is a parking lot at low speed; and the wet lab still belongs to humans. No real faces — ideas are rendered as archetypes. Imagery generated on our own Replicate account in a locked house style.
The Knowledge Gardens Production powered by XRWorkers.

→ Reel 295  ·  → Reel 296  ·  → Reel 292  ·  → The map