2026-09-27

Jensen Pushes Back on Doomers, Xi & Trump Talk AI, and “AI” Gets a Rebrand | #294 MOONSHOTS Live

At Moonshots Live's inaugural event, Peter Diamandis, Alex Wissner-Gross, Dave Blundin, Salim Ismail, and Emad Mostaque react to a wave of AI-doom rhetoric from Geoffrey Hinton, Dario Amodei, and Bernie Sanders, countering with Jensen Huang's and Mark Zuckerberg's pushback and their own near-zero P(doom) stances. They dig into the Xi-Trump AI summit, the US rebranding AGI as "superintelligence," and an AI-safety-cartel critique, framing lab liability and governance as the real fight rather than existential risk. A rapid-fire model-release recap covers Claude Opus 5.5, GPT-6 Astra, and Meta's breakout Muse app, plus the new DriveBench demonstration of a generalist model driving a car, pointing to compute and energy as the next bottlenecks. The episode closes on frontier labs opening wet labs, Claude's AI-driven discovery of a new CRISPR-like enzyme, and a proposed "Millennium Problems for Biology" benchmark.

▶ Watch on YouTube

Topics

AI doom debate: Jensen, Hinton, and the panel's P(doom) AI ▶ 5:09
Peter plays clips of Jensen Huang dismissing Geoffrey Hinton's 10% doom estimate and Mark Zuckerberg's optimism, then the panel stakes out positions: Alex's P(doom)<0, Salim's P(abundance)>>P(doom), and Dave calling doomerism dysfunctional.
Xi-Trump AI summit and lab liability Geopolitics ▶ 11:34
The panel discusses Trump and Xi agreeing to leave AI development where it is, the DOJ as the US's stated guardrail, and labs lobbying to avoid waiving liability for harm their products cause.
AI safety cartel critique and 'defensive co-scaling' AI ▶ 12:40
Alex argues OpenAI's founding conceit always included forming a coordination mechanism among frontier labs that functions like a safety cartel, and proposes 'defensive co-scaling' (a police/fire-department model) instead of deceleration.
Nation-states vs city-states in an AI era Geopolitics ▶ 16:34
The panel debates whether superintelligence will erode nation-states in favor of city-states that can self-manage via solar, satellite internet, and vertical farming, or instead cause a proliferation of new government forms.
Rebranding AGI as 'superintelligence' AI ▶ 20:14
After officials start calling it 'superintelligence' instead of AGI, the panel debates the name change, with Alex and Salim both endorsing it as a better cognitive frame than the vague, already-overloaded term AGI.
Bernie Sanders' proposed AI ban and 'corporate death penalty' Geopolitics ▶ 25:42
A clip shows Bernie Sanders calling for a pause and permanent ban on superintelligence with corporate-death-penalty enforcement; the panel calls it dangerous, comparing it to dystopian book-burning and warning it would hand data-center jobs to other states and let China pull ahead.
Record pace of frontier model releases Compute ▶ 31:36
Peter recaps 12 major model releases in 22 days (4 in the last 5), and the panel discusses Claude Opus 5.5 vs GPT-6 Astra, with Dave predicting one model release per day by year's end.
Meta Muse and the consumer-AI distribution race AI ▶ 32:06
Meta's Muse hit #1 on the App Store with 2.8 million downloads, the fastest consumer AI launch ever; Dave and Alex frame it as Meta's distribution play to stay relevant as synthetic content replaces apps like Instagram.
DriveBench: a generalist model drives a car Robotics ▶ 40:01
A new benchmark had GPT-6 Astra drive a modified Toyota Corolla through a 130-meter cone course from a single text command, completing it on the second try, prompting the panel to declare dedicated robotics models obsolete.
Compute and energy as the new bottleneck Energy ▶ 43:16
Dave argues the bottleneck has shifted from models to compute and then energy: HBM RAM prices are up roughly 5x, and the world is projected to be short about 60 gigawatts of energy within two years.
Anthropic's wet lab and the autonomous enzyme discovery Biotech ▶ 44:01
Anthropic opened its own life-sciences research group and wet labs; Claude, given one prompt, ran 950 agents for 21 hours and found a new CRISPR-like DNA repeat, named the array-associated reverse transcriptase.
Millennium Problems for Biology benchmark Biotech ▶ 47:38
Sam Rodriguez's Edison Scientific and Future House propose a 'Millennium Problems for Biology' list (origin of life, cryopreservation, limb regrowth, etc.) designed to be hard to solve but fast to verify in a wet lab.

Predictions made

open Alex Wissner-Gross: Superintelligence will cause a proliferation of new forms of government rather than an erosion of nation-states, similar to how audio recording diversified accents and dialects instead of homogenizing them.
· due: unspecified · ▶ watch
“I would predict the exact opposite is going to happen, thanks to superintelligence... a proliferation of new forms of government.”
Your call:
open Dave Blundin: US and Chinese AI delegations will hold a follow-up meeting in November 2026.
· due: 2026-11 · ▶ watch
“they agreed to meet again in November”
Your call:
open Salim Ismail: Cancer will largely be solved within two years.
· due: 2028 · ▶ watch
“we will likely solve cancer in the next two years”
Your call:
open Alex Wissner-Gross: Anthropic's code-based/procedural approach and OpenAI's pixel/raster approach to AI-generated media will collide, with robotics and embodiment as the deciding arena.
· due: unspecified · ▶ watch
“I think they're on a collision course.”
Your call:
open Dave Blundin: Frontier AI labs will reach a cadence of one major model release per day by the end of 2026.
· due: 2026 · ▶ watch
“One model per day by the end of this year.”
Your call:
open Dave Blundin: The AI industry will be roughly 60 gigawatts short of the energy it needs within two years.
· due: 2028 · ▶ watch
“We're short, like, 60 gigawatts of energy in the next two years.”
Your call:
open Emad Mostaque: No narrowly specialized AI model will outperform a generalized model in its own domain within a couple of years.
· due: 2028 · ▶ watch
“I don't think there's a single specialized model that you can say will outperform in a couple of years' time.”
Your call:

Numbers that matter

Worth digging into

🕳️ Array-associated reverse transcriptase
A genuinely novel CRISPR-like enzyme reportedly discovered autonomously by 950 Claude agents in 21 hours from one prompt — worth tracking for independent scientific validation.
🕳️ Millennium Problems for Biology benchmark
Edison Scientific/Future House's framework (origin of life, cryopreservation, limb regrowth, etc.) could become the de facto scoreboard for AI-driven biology progress, the way the math Millennium Problems did.
🕳️ DriveBench benchmark results
If a generalist frontier model really drove a modified Corolla through a cone course from a single text command, it's a strong signal specialized robotics models are becoming obsolete — but Dave flagged that compute cost wasn't disclosed.
🕳️ Anthropic's wet lab and life sciences research group
A hard case study in autonomous AI-driven scientific discovery that bears directly on how fast AI-bio progress could move, and sits in tension with Anthropic's public safety messaging.
🕳️ The AI 'safety cartel' framing
If OpenAI's founding coordination mechanism really functions as a cartel among frontier labs, it reframes current liability-waiver asks to the DOJ and the Xi-Trump talks as industry self-protection rather than pure safety.