2026-05-07

Demis Hassabis on AGI, Robots Scale Production, and Elon’s $1T Mars-Shot Comp | EP 253

Recorded live at MIT, the panel opens with Elon Musk's unprecedented SpaceX comp package (super-voting shares tied to a million-person Mars colony and 100 terawatts of space compute) and the Musk-vs-OpenAI trial, using both to debate whether moonshots are best pursued as founder-led for-profits, public benefit corporations, or nonprofits. They cover humanoid robot production scaling at Figure AI, 1X Technologies and Tesla, a Chinese court ruling that AI cannot be used to justify demoting or firing workers, and a sharp on-air disagreement between Steven Kotler and Alex Wissner-Gross over whether today's AI is actually improving as a creative and scientific tool. Demis Hassabis's reported claim that AGI may not need a further breakthrough anchors a debate over whether AGI was effectively already achieved with GPT-3 in 2020. The back half covers Colossal's bluebuck de-extinction target, GLP-1/retatrutide longevity data, and a closing round of predictions on what AI will feel like to consumers and enterprises by 2028.

▶ Watch on YouTube

Topics

Elon's SpaceX Mars-shot comp package Space ▶ 6:54

SpaceX board approved a comp package paying Elon Musk roughly $500B in super-voting shares if he establishes a Mars colony of 1M+ people at a $7.5T valuation, plus 60.4M restricted shares for bringing 100 terawatts of space compute online; panel debates it as a new 'moonshot corporation' governance model.

Musk v. OpenAI trial and nonprofit-to-for-profit AI labs Economy ▶ 15:52

Elon's suit against OpenAI (breach of charitable trust, $150B damages, reversion to nonprofit) is discussed alongside Greg Brockman's disclosed diary; panel debates whether moonshot AI labs (OpenAI, Anthropic) can survive as nonprofits or must become for-profit/public-benefit corporations once they succeed.

Humanoid robot production scaling Robotics ▶ 36:16

Figure AI (1 robot/day to 1/hour, targeting 100,000 by 2030), 1X Technologies (10,000 robots in 2026, 100,000 in 2027) and Tesla Optimus (1M units by 2030) are covered as evidence that humanoid manufacturing is scaling exponentially.

Debate over humanoid vs. specialized robot form factors Robotics ▶ 41:57

Alex Wissner-Gross argues humanoid robots are the wrong form factor for repetitive industrial tasks and better suited to human-scale environments (elder care); the panel debates the '10 billion humanoid robots by 2040' target and what a post-humanoid, nano-scale robot future looks like.

China's AI-employment court ruling and the US-China AI race Geopolitics ▶ 48:37

A Hangzhou court ruled a company could not demote a worker just because AI automated part of his job; panel debates whether this protects workers or is state-managed PR, and argues the deeper US-China 'AI race' is really about who unlocks superintelligence-driven scientific discovery first.

Demis Hassabis on the AGI timeline AI ▶ 57:06

Demis Hassabis is reported saying AGI may not require a major new breakthrough; Alex Wissner-Gross recalls Hassabis previously estimating five-plus breakthroughs were needed a decade ago versus zero or one now, and argues AGI was arguably already achieved with GPT-3 in 2020.

Is AI actually getting better? Kotler vs. Wissner-Gross AI ▶ 24:53

Steven Kotler argues that as a working scientist and writer he finds AI has gotten worse and misses obvious cross-disciplinary gaps, while Alex Wissner-Gross and Dave Blundin argue self-improvement is real and measurable via benchmarks and iterative algorithm tweaking, comparing it to evolution.

Colossal's de-extinction pipeline Biotech ▶ 1:05:24

Colossal (Ben Lamm/George Church) announced the bluebuck as its sixth de-extinguished species (of a confidential 15-species pipeline), expected in 2028; panel discusses whether revived species are true originals or DNA-reconstructed hybrids, and the Russian Cosmist idea of reviving all humans who ever lived.

GLP-1 drugs and longevity escape velocity Longevity ▶ 1:10:29

GLP-1 drugs (Ozempic, Mounjaro) outearned OpenAI and Anthropic combined by 2.4x in 2025; Eli Lilly's retatrutide shows dramatic phase trial results and is expected FDA-approved by mid-2027, with Alex Wissner-Gross betting the GLP-1 drug class is the most likely path to longevity escape velocity by the early 2030s.

What will AI feel like by 2028? AI ▶ 1:18:37

Closing prediction round on how AI will be experienced by consumers versus enterprises by mid-2028: ambient/always-listening AI, AR glasses, holodeck-style immersive rooms, AI companionship replacing human conversation, and a looming compute shortfall relative to consumer demand.

Corporate governance for moonshots (EXOs, MTPs, dashboards) Economy ▶ 13:52

Salim Ismail frames Elon's comp package as an Exponential Organization (EXO) structure and explains how a Massive Transformative Purpose (MTP) plus a one-year real-time operating plan outperforms rigid linear targets, using TED's growth from a projected 2,500 to an actual 20,000 TEDx events as the example.

Predictions made

open Peter Diamandis: Figure AI will scale humanoid robot production to 100,000 units total between now and 2030.

EP #? · · due: 2030 · ▶ watch

“Their target is 100,000 robots between now and 2030.”

Your call:

open Peter Diamandis: 1X Technologies will produce 10,000 robots in 2026 and 100,000 robots in 2027.

EP #? · · due: 2027 · ▶ watch

“1X Technologies production goal is 10,000 robots this year. 100,000 in 2027.”

Your call:

open Elon Musk: Up to 10 billion humanoid robots will exist by 2040 (prediction echoed by Brett Adcock; Alex Wissner-Gross calls it 'probably right').

EP #? · · due: 2040 · ▶ watch

“The prediction that comes both from Elon and from Brett Adcock, up to 10 billion humanoid robots by 2040. It's probably right.”

Your call:

open Elon Musk: Tesla will produce 1 million Optimus humanoid robots by 2030.

EP #? · · due: 2030 · ▶ watch

“Elon predicts 1 million Optimus production by 2030.”

Your call:

open Alex Wissner-Gross: The number of humanoid robots will cross the number of wheeled robots in the early 2030s, and humanoid robot count will surpass the number of humans on Earth by the end of the 2030s.

EP #? · · due: late 2030s · ▶ watch

“You find that in the early 2030s, the number of humanoids is predicted to cross the number of wheeled robots... it's not going to be the case that humanoids pass humans by the end of the 2030s.”

Your call:

open Alex Wissner-Gross: By 2040 there will be many trillions of direct cellular nanorobots in our solar system, making the '10 billion humanoids by 2040' figure look as quaint as 1950s predictions of atomic vacuum cleaners.

EP #? · · due: 2040 · ▶ watch

“I do expect by 2040 we're going to get our nano robots... I would expect many, many trillions of those to be in our solar system.”

Your call:

open Salim Ismail: Within roughly 2-3 years, society will reach a point where for-profit vs. nonprofit structuring stops making sense for moonshot goals as we move into a post-capitalist economy.

EP #? · · due: 2028-2029 · ▶ watch

“I think we're going to enter a point at some point, I'll say 2, 3 years, where it won't make sense anymore.”

Your call:

open Peter Diamandis: Colossal will bring back the bluebuck, extinct for 200 years, via genome editing and lab testing.

EP #? · · due: 2028 · ▶ watch

“They expect to be able to bring back the bluebuck in 2028.”

Your call:

open Peter Diamandis: Eli Lilly's retatrutide, a triple-agonist longevity drug, will be released and FDA approved.

EP #? · · due: mid-2027 · ▶ watch

“It's an extraordinary drug that is expected to be released and FDA approved by mid-2027.”

Your call:

open Alex Wissner-Gross: Longevity escape velocity will likely arrive by the early 2030s, if not sooner, primarily via the GLP-1 drug class.

EP #? · · due: early 2030s · ▶ watch

“When we speak about longevity escape velocity, which I do think is likely for the record, perhaps sometime by the early 2030s, if not earlier.”

Your call:

open Dave Blundin: The compute shortfall relative to consumer demand for immersive AI experiences (like Link's holodeck) will not be solved by 2028, but should ease by 2030-2031 once new chip fabrication ('terrafab') capacity comes online.

EP #? · · due: 2030-2031 · ▶ watch

“There's no way that's going to get solved by 2028. I'm hoping 2030, 2031 after the terrafab has a chance.”

Your call:

open Dave Blundin: An entire generation will have 70-80% of their conversations with AI companions rather than with other people.

EP #? · · due: unspecified (framed around 2028) · ▶ watch

“I suspect that an entire generation will have 70-80% of their conversations will be with AIs and not with other people.”

Your call:

open Alex Wissner-Gross: By the end of this decade, many people will be swallowing ingestible computers (pillbot-type devices).

EP #? · · due: 2029 · ▶ watch

“I'd be sorely disappointed if by the end of this decade we don't have many people swallowing computers.”

Your call:

open Alex Wissner-Gross: Extrapolating the historical shrinkage of computers needed to hit one gigaflop of compute (as seen in Apple Watch, iMac, and iPhone launches), by the mid-2040s a gigaflop-class computer will be approximately the size of a human (eukaryotic) cell, enabling cell-sized nanomachine AI.

EP #? · · due: mid-2040s · ▶ watch

“By... the mid-2040s, around the time Ray says we're going to hit his version of a technological singularity, the size of a computer hits approximately the size of a eukaryotic human cell.”

Your call:

Numbers that matter

Figure AI increased robot production from 1 robot/day to 1 robot/hour Evidence cited for exponential scaling of humanoid robot manufacturing.
Elon's SpaceX comp package: 200 million super-voting shares (10:1 voting power) vesting when a Mars colony reaches 1 million+ people, at a $7.5 trillion market valuation, paying out roughly $500 billion SpaceX board-approved compensation package, described as unprecedented in corporate history.
Second tranche of Elon's comp package: 60.4 million restricted shares for bringing online 100 terawatts of space compute (up from an earlier stated target of 100 gigawatts) Compute-based milestone in the same SpaceX comp package.
Elon is seeking $150 billion in damages from OpenAI, plus reversion to full nonprofit status and removal of Sam Altman and Greg Brockman Terms of Musk's lawsuit against OpenAI.
Polymarket gives Elon a 33% chance of winning the OpenAI trial, down from 50/50 a couple weeks earlier Betting-market odds tracked as a live indicator of the lawsuit's trajectory.
OpenAI recently received a $122 billion investment Raised as a rhetorical question about how such an investment could be reversed if Musk's suit succeeds.
Figure AI has raised funding at a valuation of $30 billion plus Cited as evidence Figure is one of the best-funded humanoid robotics companies.
1X Technologies opened a new 58,000 sq ft manufacturing facility in Hawthorne, California, for its Neo robot Production scale-up supporting 1X's 2026-2027 robot output targets.
Roughly 1.6 billion cars are on the road today, with about 100 million manufactured per year Alex Wissner-Gross's baseline comparison for extrapolating humanoid robot production curves.
iPhone production grew from 1.3 million units in year one to about 250 million iPhones produced per year now Second comparison used to extrapolate the humanoid robot production ramp.
Hangzhou tech worker 'Zhao' earned 25,000 yuan/month and was offered a demotion to 15,000 yuan/month after AI automated part of his role; the court sided with him Chinese court ruling that AI-driven efficiency is a business choice, not grounds to unilaterally alter an employment contract.
China is roughly 80-85% optimistic about AI, versus about 25% optimistic (majority pessimistic) in the US Contrast cited to argue AI sentiment, not just capability, differs sharply by country.
Species die-off rates are roughly 1,200 times greater than historical baseline Steven Kotler's concern that de-extinction news could understate the severity of the ongoing environmental crisis.
GLP-1 drugs (Ozempic and Mounjaro) generated 2.4 times more revenue than OpenAI and Anthropic combined in 2025 Headline stat framing GLP-1s as health-span drugs competing economically with AI labor investment.
The GLP-1 drug market is estimated at up to $500 billion, but is capped by manufacturing capacity, not demand Both GLP-1 drugs and AI compute are described as fully sold out / supply-constrained rather than demand-constrained.

Worth digging into

🕳️ Demis Hassabis's compressing AGI timeline

Alex Wissner-Gross reports Hassabis went from believing 5+ major breakthroughs were needed for AGI (a decade ago) to zero or one now, and Hassabis is separately quoted giving AGI a 50/50 chance of needing one more breakthrough in world models -- a rare, trackable, first-person timeline shift from one of the field's top researchers.

🕳️ Elon Musk's SpaceX comp package as a new corporate governance category

Alex Wissner-Gross frames this as the seed of a novel corporate structure -- companies that exist purely to achieve moonshots and compensate proportionally -- distinct from C-corps and public benefit corporations; if it becomes a template, it could reshape executive comp for ambitious ventures.

🕳️ Musk v. OpenAI trial and the nonprofit-to-PBC pattern

The panel draws a direct parallel between OpenAI's and Anthropic's founding stories (nonprofit/alignment-lab intentions overtaken by the need for capital), suggesting this is now a structural inevitability for any AI moonshot -- a legal precedent here could shape how future AI labs are chartered.

🕳️ Humanoid robot production race and the 10-billion-by-2040 thesis

Figure AI, 1X, and Tesla all published concrete, near-term production targets (10,000 to 100,000 units by 2026-2027) that make Elon/Brett Adcock's 10-billion-by-2040 figure directly checkable against real manufacturing data.

🕳️ Retatrutide and the GLP-1 longevity escape velocity thesis

Alex Wissner-Gross names the GLP-1 drug class as his single likeliest path to longevity escape velocity by the early 2030s, backed by extraordinary phase-trial data (37 lb weight loss, 80% liver fat reduction) and a stated mid-2027 FDA approval target.

🕳️ China's Hangzhou AI-employment court ruling

This appears to be a first-of-its-kind legal precedent determining whether AI-driven productivity gains accrue to the worker, the company, or the state -- a question the panel frames as having very different societal outcomes depending on the answer.