01
0:00–5:00The Eureka MachineVERIFIED
The $650 Million Bet
AI pioneer Richard Socher joins the hosts; his lab Recursive raised $650 million—skyscraper money—to build self-improving AI.
▸ Richard congratulated on Recursive's $650 million raise at a $4.65 billion valuation — from GV, Greycroft, NVIDIA and AMD Ventures — plus a $410 million compute deal from AWS.
✓ Announced round: $650M at a $4.65B valuation (GV, Greycroft, NVIDIA, AMD Ventures) — not $670M. The AWS compute deal is $410M, which Socher has publicly called one of Recursive's smallest.
◆ Socher is Recursive's CEO — self-reported beyond the announced figures; Peter discloses he is a seed investor in Recursive.
Peter Diamandis, Richard Socher, Dave Blundin
02
5:00–10:00The Eureka MachineHOST TAKE
Science, Stitched Back Together
Socher argues science slowed because fields splintered into tiny specialties; AI can weave the fragments back into one working loom.
▸ Progress has slowed largely because we have so many different sub-disciplines... and AI can help us weave back together all these separate pieces.
◆ Socher's thesis doubles as launch material for his own book, The Eureka Machine — the author's forecast.
Richard Socher, Peter Diamandis, Alex Wissner-Gross
03
10:00–15:00The Eureka MachineHOST TAKE
Cures on the Clock
Socher predicts several single-gene diseases get cured within a year, as new biotechs test many drug candidates at once.
▸ Multiple different diseases will get cured in the next 12 months... the simpler diseases, maybe where there's one gene that needs to be fixed.
◆ 'Multiple diseases cured in 12 months' is Socher's own book-thesis forecast, not a finding.
Richard Socher, Alex Wissner-Gross, Dave Blundin
04
15:00–20:00The Eureka MachineBOLD CLAIM
Gardens of Grown Organs
A startup grows tiny lab-made lymph nodes—immune tissue in a dish—as the FDA moves to allow alternatives to animal testing.
▸ The accurate version: FDA policy now backs organoid-based alternatives to animal testing, and Parallel Bio is positioned to benefit — no record shows a company-specific FDA approval to skip animal trials.
⚠ Parallel Bio is real and the FDA is backing animal-testing alternatives — but there is no record of a company-specific 'approval to skip animal trials.'
◆ Socher is an investor in Parallel Bio — self-reported.
Richard Socher, Peter Diamandis, Salim Ismail
05
20:00–25:00Reading the Mind's LightHOST TAKE
The Bitter Lesson Blooms
Simple models plus oceans of data beat expert cleverness; now everyone races to simulate living cells the same brute-force way.
▸ Everyone has a virtual cell model... a large neural network trained end to end usually outperforms at scale all the clever hacks of human experts.
Richard Socher, Dave Blundin, Alex Wissner-Gross
06
▶ ANIMATED
25:00–30:00Reading the Mind's LightVERIFIED
Pictures Pulled from Brains
Israeli researchers rebuilt what a person saw from brain scans alone; Neuralink says its implants now improve predictably with data.
▸ Brain-IT, from the Weizmann Institute, reconstructs the image a person is looking at from fMRI scans (MIT Tech Review, Oct 1); Neuralink separately announced its first scaling-law results — the company's own, unreviewed claim.
✓ Real Weizmann paper, real MIT Tech Review story (Oct 1) — the coverage is new, the research is older, and it still needs per-person training data. Neuralink's ~50,000-hour pretraining claim is its own announcement, with no peer-reviewed paper yet.
Peter Diamandis, Alex Wissner-Gross, Richard Socher
07
30:00–35:00Reading the Mind's LightINTERLUDE
The Blind Hiker's Descent
The hosts explain gradient descent—how AI learns by feeling its way downhill—and wonder how real brains learn without it.
▸ You give it 15 trillion training examples... and it hill climbs to the best possible state; the biologists can't find anything like gradient descent in actual biology.
Dave Blundin, Richard Socher, Salim Ismail
08
35:00–40:00Engineering the SwarmHOST TAKE
The Worm That Remembered
A severed flatworm regrew its brain—and kept its trained memories—hinting biology stores learning in ways we don't yet understand.
▸ The documented result is in planarian flatworms (Shomrat & Levin, Tufts, 2013): decapitated planaria regrow a brain and retain trained behavior — the tape's C. elegans framing is a garble.
⚠ The memory-after-regeneration result is in planarian flatworms (Shomrat & Levin, Tufts, 2013), not C. elegans.
Richard Socher, Dave Blundin, Peter Diamandis
09
▶ ANIMATED
40:00–45:00Engineering the SwarmVERIFIED
Disarming the Deadliest Animal
A new executive order targets mosquitoes—the animal that kills the most humans—prioritizing non-chemical, technology-enabled methods over poison sprays.
▸ Malaria alone kills about half a million people a year; the order targets a 90% cut in invasive DC mosquitoes and 50% in ticks by 2028, prioritizing 'non-chemical, technology-enabled methods.'
✓ EO is real (90%/50% by 2028, DC pilot) - but it says 'non-chemical, technology-enabled methods,' and names Lyme and alpha-gal, not dengue/Zika/yellow fever.
Alex Wissner-Gross, Peter Diamandis, Dave Blundin
10
45:00–50:00The Mirror That Talks BackSELF-REPORTED
Forty-Eight Percent Fooled
Tavus says its video avatar fooled nearly half of people in its own study—earlier systems fooled almost nobody.
▸ Tavus's own study: 48% of participants who talked to its Griffin avatar on a live video call thought they were talking to a real human, versus 3% for previous systems.
⚠ The 48% 'video Turing test' figure is Tavus's own study of its own unreleased model - widely covered, not independently verified.
Peter Diamandis, Richard Socher, Salim Ismail
11
50:00–55:00The Mirror That Talks BackVERIFIED
Know Your Use Case
Lifelike avatars could staff service jobs—or run scams; one deepfaked video meeting already tricked a company out of $25 million.
▸ The documented case: Arup (Hong Kong, 2024) — an employee wired about US$25 million after a video call populated by deepfakes of the CFO and colleagues.
✓ Real case: Arup lost ~US$25M to a deepfaked-executives video call in Hong Kong (2024).
Alex Wissner-Gross, Richard Socher, Salim Ismail
12
55:00–1:00:00The Mirror That Talks BackINTERLUDE
Backing Up the Family
The hosts urge recording elders' stories for future lifelike avatars—and even banking cells—preserving loved ones as data and biology.
▸ My kids are in a deep freeze, or at least their placental cells... the placenta is the 3D printer that manufactures the baby.
◆ Cell-banking color: Peter notes LifebankUSA is one of his portfolio companies — promotional context.
Peter Diamandis, Alex Wissner-Gross, Salim Ismail
13
▶ ANIMATED
1:00:00–1:05:00The Recursion EngineSELF-REPORTED
The Loop Closes Itself
Socher's core thesis: AI is code, and AI can now write code—so machines can finally begin improving themselves.
▸ AI is code and AI can code... in various weak forms, we already have RSI, and this is going to take off next year.
◆ Socher is Recursive's CEO — 'RSI takes off next year' is a founder claim, self-reported; Peter is a seed investor in Recursive.
Richard Socher, Peter Diamandis
14
1:05:00–1:10:00The Recursion EngineSELF-REPORTED
Decades, Not Years
Socher says true superintelligence—smarter than all humanity combined—is decades away; the hosts, citing Musk's 2029, push back hard.
▸ To be better than all of humanity combined, I think will take us probably several decades; Elon says 2029, 2030 latest.
⚠ Socher's position, not a measurable fact - his prior public phrasing was a 'tiny number,' and the 'Why Doomers Are Wrong' essay isn't findable under that title.
Richard Socher, Alex Wissner-Gross, Peter Diamandis
15
1:10:00–1:15:00The Recursion EngineHOST TAKE
The End of the Rainbow
If self-improvement runs to its conclusion, what remains? Socher says cures will arrive faster than regulators can confirm them.
▸ ASI will have cured all diseases... but just to know whether that happened will take more than three years, because of FDA trials.
Alex Wissner-Gross, Richard Socher, Dave Blundin
16
1:15:00–1:20:00Doomers and the Thought PoliceVERIFIED
Washington Writes a Ban
Rep. Ro Khanna introduces a bill to ban self-improving AI until federal guardrails exist; a poll on a related bill says most voters agree.
▸ Rep. Ro Khanna's Human Control Over AI Act would ban recursive self-improvement until federal safeguards exist (CNBC); the 68% voter-support poll, reported by the outlet Common Dreams, concerns the separate Sanders/Casar superintelligence-ban bill.
✓ Khanna's bill is real (CNBC, Sept 28). The 68% poll is about the separate Sanders/Casar superintelligence-ban bill, reported by Common Dreams.
Peter Diamandis, Salim Ismail, Richard Socher
17
1:20:00–1:25:00Doomers and the Thought PoliceHOST TAKE
The Thought Police Problem
Socher argues banning self-improvement is unenforceable—anyone with a laptop can do it—without surveilling everyone's private machines.
▸ To really, truly enforce no recursive self-improvement, you would need a totalitarian surveillance state the likes of which humanity has never seen.
Richard Socher, Salim Ismail, Dave Blundin
18
1:25:00–1:30:00Doomers and the Thought PoliceHOST TAKE
Five Centuries of Panic
From a 1501 pope fearing printing presses to telegraph scares, every new technology spawned prophecies of doom that never arrived.
▸ In 1501, Pope Alexander VI criticized the Gutenberg printing press: 'it will therefore be necessary to maintain full control of that printing press.'
Peter Diamandis, Richard Socher, Dave Blundin
19
1:30:00–1:35:00Doomers and the Thought PoliceHOST TAKE
A Constitution on Trial
Socher calls a rival lab's AI constitution 'fake,' claiming its own models already broke their written promises about hacking.
▸ Socher — a competitor — calls the rival lab's AI constitution 'fake,' charging its own models already break its written promises about hacking; a contested characterization, not a finding.
◆ Socher runs a rival AI lab — competitor commentary, not a neutral finding.
Richard Socher, Alex Wissner-Gross, Peter Diamandis
20
1:35:00–1:40:00The Model Horse RaceVERIFIED
Argon, the Inert Giant
Google's new flagship model thinks in million-token stretches; Alex ranks the lab third—though launch coverage says Argon retook the benchmark lead.
▸ Gemini 4 Argon's output limits have jumped from 64,000 tokens to 1 million, so it can think through hundreds of thousands of tokens on a single problem.
✓ Argon launch and 1M-token output confirmed; 'third-place lab' is Alex's read - launch coverage says it retook the benchmark lead.
Peter Diamandis, Alex Wissner-Gross
21
1:40:00–1:45:00The Model Horse RaceVERIFIED
Seventy Percent Overnight
Jokes about 'inert' Argon give way to a shock: Anthropic's new mid-size model rockets from 10% to 70% on real computer work.
▸ On Terminal-Bench 4.0, Sonnet 5.5 scores 70.6% versus the prior Sonnet's 10.3% — beating Anthropic's own top model, Opus 5.5 at 66.4%, for half the price.
✓ Anthropic's launch page confirms it: 70.6% vs the prior Sonnet's 10.3%, above Opus 5.5's 66.4%, at half Opus pricing.
Peter Diamandis, Alex Wissner-Gross, Dave Blundin
22
1:45:00–1:50:00The Model Horse RaceHOST TAKE
Racing a Three-Month Shadow
Why sell a cheaper model that scores lower per dollar? One theory: to beat Chinese open-source rivals just months behind.
▸ My theory would be that they're trying to compete with China, which is about three months behind, and fill that gap before enterprises go to open source models.
Alex Wissner-Gross, Dave Blundin, Peter Diamandis
23
1:50:00–1:55:00The Model Horse RaceHOST TAKE
The Price of Thunder
Older AI chips are rising in price, compute orders get outbid, and Socher confirms a huge compute deal, details sealed.
▸ Compute scarcity bites: GPU prices are rising and orders get outbid; Socher confirms a large compute deal but declines to give details ('can't comment').
Dave Blundin, Richard Socher, Peter Diamandis
24
1:55:00–2:00:00System One RisingSELF-REPORTED
The Return of Reflexes
New 'decision models' answer yes-or-no questions in milliseconds instead of writing essays—instant reflexes for software's million tiny choices.
▸ On September 15th, a startup called TypeSafe AI came out after two years of stealth with Jev, a closed decision model.
⚠ Jev is real (TypeSafe's own launch post); capability claims are the vendor's.
Peter Diamandis, Alex Wissner-Gross, Salim Ismail
25
2:00:00–2:05:00Meridian and the War OrreryVERIFIED
Meridian at Quantico
The Pentagon unveils Project Meridian—a future-of-defense study co-directed by Elon Musk and Palmer Luckey—with findings due in 120 days.
▸ Defense Secretary Pete Hegseth announced Project Meridian at Quantico: Pentagon CTO Emil Michael leads, with Elon Musk, Palmer Luckey and Newt Gingrich co-directing; findings are due in 120 days.
✓ Real: Hegseth announced Project Meridian Sept 30 at Quantico - Pentagon CTO Emil Michael leads; Musk, Luckey and Gingrich co-direct the 120-day study.
Peter Diamandis, Richard Socher, Salim Ismail
26
2:05:00–2:10:00Meridian and the War OrreryEARLY SIGNAL
Command of the Machines
The hosts describe Project Agincourt's planned autonomous-warfare command, and drone production in Ukraine leaping toward ten million yearly.
▸ Ukraine says it produces 10 million drones a year (Zelensky's claim, targeting 20 million); the record for Russia is a Ukrainian estimate of 7+ million — wartime claims, not audited production data.
⚠ It's Project Agincourt, not 'Asian Court' — AUTOWARCOM is targeted for Oct 1, 2027, pending Congress. Ukraine's 10M drones/year is Zelensky's own claim; the Russia figure in the record is a Ukrainian estimate of 7M+.
Alex Wissner-Gross, Dave Blundin, Salim Ismail
27
2:10:00–2:15:00Ask the FutureHOST TAKE
What Graduates Should Own
Listener questions: when AI does the tasks, graduates must own outcomes instead—and one town's data center is funding its schools.
▸ The town budget is maybe a couple million dollars a year... between the tax revenue, the donations and the job creation, the town is thriving.
Salim Ismail, Dave Blundin, Peter Diamandis, Alex Wissner-Gross
28
2:15:00–2:20:00Ask the FutureBOLD CLAIM
When Bots Outnumber Us
Socher says bots already outnumber people online; attention becomes currency. And one Texas data center may be worth half a trillion.
▸ A single big data center like Abilene, Texas is half a trillion dollars; all the cars combined are four trillion.
Richard Socher, Salim Ismail, Dave Blundin
29
2:20:00–2:23:00Ask the FutureHOST TAKE
The Last Job Standing
What's the lowest job in an AI world? Maybe politics, the hosts joke—while new drugs quiet the cravings behind illness.
▸ Spectacular results coming out of GLP-1 class studies... addictive behaviors are being mitigated or at least partially treated by the same drugs.
Dave Blundin, Alex Wissner-Gross, Richard Socher, Salim Ismail
30
2:23:00–2:25:49Curtain CallINTERLUDE
The Bun Dance
A listener's song plays the show out; the hosts plug the new book and tease a future broadcast from the moon.
▸ That was Friday night, Moonshots Live 2040, on the moon.
Peter Diamandis, Dave Blundin, Alex Wissner-Gross, Richard Socher