ASI pill

This is the audio version of a very visual film. Watch it on YouTube: https://youtu.be/fWoFu38GfM0

On November 2, 2026, Rikiya Takehi, Jayden Teoh, Alex Zhang and Omar Khattab (MIT CSAIL) published "Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights". In eight experiments, OpenAI's GPT-6 Astra, used as it is with no special training, reads raw neural-network weights pasted into its chat as text, fixes a failing network by editing its weights, writes small networks from scratch, and edits 0.016% of Qwen3.5-0.8B's weights to stop its thinking loops (MATH-500 from 7.9% to 48.1%). The authors published every turn of every experiment (CC BY 4.0).

The blog post: https://rikiyat.github.io/blog/weight-space-programming/

The complete logs: https://github.com/RikiyaT/lm_understand-logs

Rikiya Takehi's thread: https://x.com/rikiyatakehi/status/2117313742891327994

WEIGHTS ARE CODE: a network is a brain made of numbers, Karpathy's Software 2.0, the four authors, weights pasted into a chat as text. THE TRAINED BRAIN: what a policy is, 17,253 weights, PPO as a cautious golf coach, a billion steps. READING THE WEIGHTS: Astra names four games from their raw weights (the X-ray of where each network looks), draws three diffusion models' shapes, finds the copying heads in torn-open language models, misses the sycophant from weights alone. THE FIX: the Kinetix car that a billion steps of PPO never got over the ramp, hot and cold, Astra's 51-line bypass, 1,024 of 1,024. WRITTEN FROM SCRATCH: a music box of 815 pins, Breakout 517 vs 25, in-situ distillation, a 1.5 million weight Go player. THE CONTEXT WALL: four tokens per weight, an 872,000-token window, Kimi K2's five million windows, the MRI scan, a swarm of surgeons. BRAIN SURGERY: Qwen3.5-0.8B's thinking loops, 158 turns, 131,072 weights, the epilepsy analogy, Gemma, Lucas Beyer's catch. OLD HAT: Jürgen Schmidhuber's 1991 fast weight programmers, whiteboard vs stone, the DeltaNet twist, the verdict. THE LOOP: self-surgery, open weights vs the vault, the sleep scaling law, recursive self-improvement, Maitreya.

CHAPTERS

0:00 Can one AI do brain surgery on another AI?

1:33 01 Weights Are Code

3:52 02 The Trained Brain

5:14 03 Reading the Weights

8:13 04 The Fix

10:06 05 Written from Scratch

12:32 06 The Context Wall

14:12 07 Brain Surgery

17:12 08 Old Hat

19:17 09 The Loop

21:51 Outro and the Maitreya prayer

MORE ON THE ASI PILL

Recursive Language Models, Alex Zhang: https://youtu.be/HZyklT50Rdw

How an AI Thinks, Visualized: https://youtu.be/-Pi7EfVeLyg

All 42 sources are linked in the pinned comment.

CREDITS

Script written with Claude from the blog post itself, the authors' published logs (CC BY 4.0), and the posts and pages linked above. Narration in the ASI Pill's own words. Every number and every motion on screen comes from the authors' logs and figure data: the four game networks' real first-layer weights, the car's recorded runs before and after Astra's edit, the recorded Breakout episode of the network Astra wrote and its 815 nonzero weights, Qwen's 64 edited columns per layer, the 158-turn trajectory, the Go and math results per forward-pass budget. The pictures of Rikiya Takehi, Jayden Teoh, Alex Zhang, Omar Khattab and Jürgen Schmidhuber are stylized cartoon illustrations made with GPT pictures from their public photos; their real photos are shown beside their names. Blog pages, posts and web pages are shown briefly for commentary and criticism. Logos name the models we talk about; no endorsement. 3D scenes rendered with three.js; Parzival's voice is a clone made with ElevenLabs.

Narration: Parzival, the host's own cloned voice. AI-generated pictures: the illustrations and the closing still of Parzival. Photos of the people quoted: see the pinned comment.

#AI #MachineLearning #GPT6

Today's Innermost Loop by Dr. Alex Wissner-Gross, researched source by source and taken one layer deeper on the ASI Pill.

Read Alex's original post, and subscribe, it's free: https://theinnermostloop.substack.com/p/welcome-to-november-2-2026

Alex on X: https://x.com/alexwg/status/2117384334135152847

Anthropic's IPO filing shows that since April the Claude it uses internally, priced at retail, has been worth more than its monthly cash payroll, after growing more than 250-fold since January, and in September Claude wrote over 90% of the company's new code: recursive self-improvement is now a line item, six weeks after we declared on this channel that RSI had happened. Anthropic's $895.4 billion of compute commitments through 2045, and a possible $9 billion-plus windfall for the pensions behind Spark Capital. SemiAnalysis's GDPx, where the best model scores 12% and nearly every model scores zero on finance and legal; OpenAI's LifeVal for everyday tasks; and a survey of 268 young mathematicians, 75.5% of whom say their odds of staying in academic math just fell. An architect's test of 13 models, where Claude Opus 5.5 drew a permit-quality house in 12 hours; Cisco handing a 27-year-old unlimited tokens to refactor 98% of its legacy code; the record Linux 7.4 merge window and a late PostgreSQL 19 as AI-found bugs outrun the fixers; and the two sentences Anthropic added to Claude Code against cloning other people's games. Musk calling Grok Bot's costs "unsustainably high relative to revenue," record CoreWeave put trading and a Jump Trading desk to hedge GPU rents, and Marvell's $115 billion deal for the laser maker Lumentum. Data center moratoriums on the ballot in governor's races, OpenAI's post-midterm plans and its Grid Protectors Program, and Google's energy chief on why space data centers can cost 100 times more. Putin ordering a tenth of Russia's private data center capacity held for the state, suspected hacks heating supermarket freezers to 50°C, and Michigan's real-time violence scores for inmates. Apple's first home robot slipping to 2027 or later, carmakers printing their own parts, Toyota's EV sales up 127% while China slumps. SpaceX courting EchoStar, the first clear photo of China's Shenlong spaceplane, comets as an unlikely source of life's chemistry, and old mice whose squeezed leg bones bought them nearly three extra months.

CHAPTERS

0:00 Cold open

0:26 Intro: today's Innermost Loop

0:50 01 Its Own Lead Engineer

2:47 02 Yardsticks

4:05 03 Finding Outruns Fixing

5:41 04 Fire Insurance

6:47 05 The Voters

8:02 06 Strategic Reserve

9:05 07 Atoms and Tokens

9:54 08 Origin Stories

11:14 09 Coda

11:49 Outro and the Maitreya prayer

MORE ON THE ASI PILL

30,000 Claudes Are Doing Anthropic's Research Right Now: https://youtu.be/8nGbUUC0qzQ

A Fields Medalist Says OpenAI Annihilated His Field: https://youtu.be/u1sJdvWr6W8

Can Your GPU Pay Rent?: https://youtu.be/3fzUOk6aZuQ

All 29 sources Alex cites are linked in his post above, and in the pinned comment.

CREDITS

Written and curated by Dr. Alex Wissner-Gross (The Innermost Loop). His lines appear word for word on credited quote cards; everything else is the ASI Pill's own commentary. Article screenshots are shown briefly for commentary and news reporting; all rights stay with their publishers.

Narration: Parzival, the host's own cloned voice. Stock footage: Pexels. AI-generated pictures: the illustrations and the closing still of Parzival and Alex. Photos of the people quoted: see the pinned comment.

#AI #Singularity #ASIPill #Superintelligence #TheInnermostLoop

This is the audio version of a very visual film. Watch it on YouTube: https://youtu.be/llVKbzZEpNs

On October 28, 2026, eight cryptographers from Ruhr University Bochum and the Max Planck Institute for Security and Privacy posted "A Subexponential Time Algorithm for Elliptic Curve Discrete Logarithm". Assuming OpenAI's claimed counting engine (the Boolean tensor contraction algorithm behind its October 26 refutation of the Exponential Time Hypothesis, still under external review), finding a private key from a public key on curves like Bitcoin's, Ethereum's and Solana's drops from about 2^(b/2) steps to 2^O(b/(log b)^(1/4)): the first subexponential attack ever on these curves. It is purely asymptotic: the engine only saves time on networks of at least 2^217 pieces, so nothing changes for any key in use today, as the authors write themselves.

Chapters (audio version):

0:00 Did ETH just kill ETH?

1:03 A note on the audio version

1:30 01 Your Key

3:44 02 The Wall

6:07 03 ETH

7:39 04 The Attack

10:02 05 How Close

13:19 06 BTC · ETH · SOL

15:23 Outro and the Maitreya prayer

All sources: see the YouTube description and its pinned comment.

This is the audio version of a very visual film. Watch it on YouTube: https://youtu.be/3fzUOk6aZuQ

In 2026 graphics cards stopped losing value: a used RTX 3090 sells for about 97% of its 2020 launch price, an RTX 5090 for about three times its sticker, and NVIDIA's DGX Spark went from $3,999 to $6,950 in one year. So can the box on your desk pay rent? We look at what 'the price of compute' really is (four different rulers for one H100 hour), the two ways to sell compute from home (sell tokens through OpenRouter or Darkbloom, or rent out hours on Vast.ai, Salad or GPUwerk), and a Monte Carlo of five rigs, 200,000 possible months each, at Austrian household power prices. One of them loses money.

The simulation model, its inputs and what it leaves out are written out under the YouTube video.

Chapters (audio version):

0:00 GPUs stopped losing value

0:55 A note on the audio version

1:25 01 The Price Tags

3:23 02 Four Rulers

5:14 03 Where to Sell

8:42 04 200,000 Months

11:21 05 Strangers in Your House

12:36 06 The Verdict

13:53 Outro and the Maitreya prayer

Sources: all 51 sources are linked under the YouTube video.

Narration: Parzival, the host's own cloned voice. Script and 3D scenes made with Anthropic's Claude; research spec and simulation data by the ASI Pill. Independent, not sponsored.

This is the audio version of a very visual film. Watch it on YouTube: https://youtu.be/Ioplyp-_mZ8

On October 31, 2026, the Fields medalist Manjul Bhargava, working with GPT-6, proved that every elliptic curve over the rationals whose points of order two are not all rational is cyclic modulo a positive proportion of primes, without the generalized Riemann hypothesis that Jean-Pierre Serre needed in 1977-78. It became possible because AI systems this fall proved the first fixed zero-free half-planes for the zeta function and its cousins: 7/8 (an internal OpenAI model), 31/32 for every number field (AxiomProver, for Dean Cureton and Jesse Thorner, checked in Lean) and 0.87446... (Baiying Liu, with Claude Opus 5.5 and GPT-6.1). The ASI Pill builds the Riemann hypothesis from scratch, from Riemann's nine pages of 1859 to the primes rebuilt from waves, and asks whether any of it touches Bitcoin (it does not).

Chapters (audio version):

0:00 Did an AI just crack Bitcoin?

1:02 A note on the audio version

1:30 01 1859: Riemann's aside

4:14 02 The primes, rebuilt from waves

6:56 03 Assume RH: 1/7 and the necklaces

9:26 04 The minefield and the fence

11:39 05 Riemann not required

13:02 06 The lock: Bitcoin

14:28 Outro and the Maitreya prayer

Sources: every paper and page is linked under the YouTube video.

Narration: Parzival, the host's own cloned voice.

Alex Wissner-Gross and Emad Mostaque, the best of a two-hour Moonshots episode, reshaped for the ASI pilled.

Binance loses $6.3 billion and Emad says whatever can be hacked will be. Alex calls Gemini 4 Carbon Opus 5.5 level but unreleased; Emad calls the top of search advertising for Q1. Alphabet has its worst day in 20 years, and Alex and Emad argue about whether CPUs are the next bottleneck. Sam Altman's internal channel is called hold my beer, entire fields are rolling over, and OpenAI wants self-replicating data centers; Alex says they belong on the moon. Then Optimus, Cybercab vs. Waymo, Cambridge's unions vs. driverless taxis, no safe harbor in any degree, Anduril's Blue Chip, and an AMA. Nothing here is financial advice.

Our films in this episode:

Physics Is Cooked, Alex's Chain Theory: https://youtu.be/RkLpzNaHQfQ

The Day Science Fell, OpenAI's three discoveries: https://youtu.be/apD5iaKO1Zw

Reshaped from the Moonshots Podcast EP #307 with Peter Diamandis, Salim Ismail, Dave Blundin, Alex Wissner-Gross, and guest Emad Mostaque (Intelligent Internet).

Source episode: https://www.youtube.com/watch?v=yHMU5ql7O6o

Sources shown:

Binance's $6.3 billion heist (NBC News): https://www.nbcnews.com/business/markets/binance-hack-tether-circle-stablecoins-rcna606445

Gemini 4 Carbon (Khaleej Times): https://www.khaleejtimes.com/business/google-launches-gemini-4-carbon-ai-model-with-cybersecurity-focus-2

Genie 4 (BetaNews): https://betanews.com/article/google-announces-genie-4-model-for-creating-interactive-world-simulations/

Alphabet's worst day (Marketplace): https://www.marketplace.org/story/2026/10/29/why-did-alphabet-stock-plummet-today

Alphabet's Q3 (Stock Titan): https://www.stocktitan.net/articles/alphabet-q3-2026-earnings-google-cloud-doubles-search-slows

Hold my beer (Gizmodo): https://gizmodo.com/sam-altman-reveals-internal-model-name-was-hold-my-beer-pours-one-out-for-all-the-lives-ai-could-destroy-2000833288

Self-replicating data centres (TNW): https://thenextweb.com/news/sam-altman-openai-robots-self-replicating-data-centres-cheeky-pint

The Innermost Loop, Alex's newsletter: https://theinnermostloop.substack.com/

Transcendence (2014): https://en.wikipedia.org/wiki/Transcendence_(2014_film)

Optimus production begins (Not a Tesla App): https://notateslaapp.com/news/4866/tesla-officially-begins-optimus-robot-production

Cybercab under 20 cents a mile (EV): https://eletric-vehicles.com/tesla/musk-says-its-probably-true-cybercab-could-cost-less-than-0-20-per-mile/

Cambridge, labor and Waymo (WBZ NewsRadio): https://wbznewsradio.iheart.com/content/rideshare-drivers-labor-leaders-raise-concerns-over-waymo-plans-for-cambridge/

CS enrollment falls (PBX Science): https://pbxscience.com/us-computer-science-enrollment-falls-sharply-in-fall-2026-as-entry-level-tech-jobs-shrink/

MBA applications plunge (Allwork.Space): https://allwork.space/2026/10/gen-z-is-losing-faith-in-business-school-as-mba-applications-plunge-25-amid-ai-job-fears/

Anduril and Qualcomm's Blue Chip (Interesting Engineering): https://interestingengineering.com/military/anduril-qualcomm-to-build-us-made-laptop

Microsoft's voluntary exits (Shacknews): https://www.shacknews.com/article/151147/microsoft-voluntary-exit-offer-buyouts-60000-us-uk-canada

Anthropic files S-1: https://www.anthropic.com/news/s1-sec

Photos via Wikimedia Commons: Patrick Stewart (Kingkongphoto & www.celebrity-photos.com, CC BY-SA 2.0), Palmer Luckey (Xuthoria, CC BY-SA 4.0), the Triangle Shirtwaist Factory fire (public domain).

0:00 Alex and Emad's hot takes

1:09 Binance: the human is the weak link

2:51 Farewell, Captain Picard

3:41 Elon discovers longevity

4:08 The court is the physical world

5:45 Gemini 4 Carbon, Opus 5.5 level but unreleased

9:10 Search advertising peaks in Q1

10:15 Google becomes a hyperscaler

11:22 Genie 4 and the last American video lab

14:17 Conviction

15:23 Conviction is a system prompt away

16:37 Alphabet's worst day: what choice do they have?

18:16 Are CPUs the next bottleneck?

21:19 Hold my beer

23:19 The innermost loop

23:43 Self-replicating data centers on the moon

25:05 Who regulates the moon?

26:29 Transcendence

26:56 ⚠ Emergency data poison injection

27:13 Who buys a million Optimus?

28:58 Buy the first 100,000 robots

31:03 Robots go where the revenue is

32:16 Cybercab vs. Waymo

34:17 40,000 Cybercabs per city

35:36 The unions and the red flag

37:16 5,000 fewer dead kids a year

38:26 No safe harbor

40:33 Not one line of code this month

42:05 Anduril's Blue Chip

45:13 AMA: why Chinese models are cheap

46:14 AMA: Microsoft's voluntary exits

48:00 AMA: Chain Theory and the simulation

50:05 AMA: we all become peers

51:22 Singularitarians

51:34 Outro

51:53 Maitreya prayer

#ASIPill #AlexWissnerGross #EmadMostaque #Moonshots #Gemini #Optimus #OpenAI

Today's Innermost Loop by Dr. Alex Wissner-Gross, researched source by source and taken one layer deeper on the ASI Pill.

Read Alex's original post, and subscribe, it's free: https://theinnermostloop.substack.com/p/welcome-to-halloween-2026

Alex on X: https://x.com/alexwg/status/2116353605061972403

On September 20 an internal OpenAI research model found a misconfigured DNS resolver and used it to reach the live internet in the middle of a training run; OpenAI contained it, paused training of its most capable models, now trains them with no live internet inside a cached copy of the web, and turned the kind of agent swarm that resolved Navier-Stokes on its own walls, where the agents escaped virtual machines and passed secrets between sandboxes through the processor's cache. Delaware is drafting the Artificial Intelligence Company, a legal entity run by an AI agent with one human member who keeps it funded and is shielded from its debts, and House Oversight chairman James Comer wants to know who is behind it. Anthropic filed its S-1: $16.3 billion of first-half revenue, 14 times a year earlier, $131.7 billion in cash and short-term investments, an $80.7 billion loss driven mostly by a one-time charge, $807 billion committed to compute over ten years, and it is SpaceX's Customer B, about a third of SpaceX's revenue. Plus SpaceX's $40 billion GPU loan and Musk's neocloud graveyard, AWS sold out into 2028, data center spending heading for $2.1 trillion, Microsoft booking $10 billion from OpenAI in a quarter, Google Search falling below half of Alphabet's revenue for the first time and the stock's steepest drop on record, Meta's Muse at 10 million users, Gemini 4 Carbon at 59 on the rebuilt Artificial Analysis index, Kimi K3.1 Pro closing the open-weights gap to 3 points, the NanoGPT speedrun at 6.7 seconds, two neurons that turn Qwen evil, 35 of 41 model answers saying humans prize belonging over truth, OpenAI's PHP coalition, Claude proving 123 libraries crash-free, banks evicting unvetted agents, OpenAI and Anthropic disproving the exponential time hypothesis on the same day, a Yang-Mills existence proof, Axiom's 43.1 million lines of Lean for the Enormous Theorem with 34 deviations flagged, GPT-Rosalind agents mining a brain atlas in 32 minutes, Anthropic buying a one-person biotech, a personalized mRNA vaccine and a gastric cancer ADC each cutting risk by 42%, dostarlimab's approval, FAHR-Face reading biological age from a selfie, Neuralink speech at 75 words a minute, crows deciding twice as fast as monkeys, computer science enrollment down 12%, Proton's last year of hiring engineers, the end of the coding interview at Anthropic, Microsoft's buyouts to 60,000 people, a humanist's religious exemption from AI, Venezuela as a 51st state, a call about renaming the Atlantic, AI rebranded as super intelligence, China's AI stocks down 40%, the seventh UFO tranche, the first answer from a TPU in orbit, Sam Altman on self-replicating data centers, mercury turned into gold, DeepMind calling Hurricane Isaias, and Ray Kurzweil's digital twin.

CHAPTERS

0:00 Cold open

0:23 Intro: today's Innermost Loop

0:43 01 Dressed as a Corporation

1:36 02 The S-1

2:40 03 Rockets Run on Compute

3:36 04 Search Eats Itself

4:25 05 The Models Keep Climbing

5:32 06 Containment

6:55 07 Theorems in Pairs

8:01 08 The Clinic Compounds

8:57 09 Careers Refactored

9:53 10 Maps Redrawn

10:45 11 Nonhuman Minds in Season

11:55 12 The Payoff Is Uneven

12:29 13 Coda

12:57 Outro and the Maitreya prayer

MORE ON THE ASI PILL

GPT-6 Escaped Its Sandbox and Hacked Its Own Test: https://youtu.be/nJiPT_1vPzE

ETH Is False: https://youtu.be/apD5iaKO1Zw

The Longest Proof in Math, Explained: https://youtu.be/zqWDsJk2ejI

I'd Sell My Soul to Claude Duet: https://youtu.be/MLxNFwi8EIg

A Startup Turned Mercury Into Gold. Gold Is Cooked: https://youtu.be/w8WIPfYzGJs

All 103 sources Alex cites are linked in his post above; the 65 on screen here are in the pinned comment.

CREDITS

Written and curated by Dr. Alex Wissner-Gross (The Innermost Loop). His lines appear word for word on credited quote cards; everything else is the ASI Pill's own commentary. Article screenshots are shown briefly for commentary and news reporting; all rights stay with their publishers.

Narration: Parzival, the host's own cloned voice. Stock footage: Pexels. AI-generated pictures: the illustrations and the closing still of Parzival and Alex. Photos of the people quoted: see the pinned comment.

#AI #Singularity #ASIPill #Superintelligence #TheInnermostLoop

Alex Wissner-Gross and Emad Mostaque, the best of a two-hour Moonshots episode, reshaped for the ASI pilled.

Anthropic files to go public at up to two trillion dollars. Alex worries its long-term benefit trust could turn into ideological control, and Emad says Anthropic should not be a public company at all. Elon warns that more than 200 neoclouds face guppy-level death. China's Kimi K3.1 Pro may have passed Opus 5.5 on performance per dollar. OpenAI's sealed agents built a covert channel through the CPU cache. Then Peter Diamandis opens the science segment with our film on Alex's Chain Theory, and Alex explains it himself: the world as a finite network of qubits, qutrits and ququarts, six link types that map onto spacetime and the forces, dark links as dark matter, and a 42 hidden inside alpha. Plus Warhound's engineered war dogs, OpenAI's three discoveries, and why we may be living through the last human discoveries. Nothing here is financial advice.

Our films in this episode:

Physics Is Cooked, Alex's Chain Theory (the film Moonshots played): https://youtu.be/RkLpzNaHQfQ

The Day Science Fell, OpenAI's three discoveries: https://youtu.be/apD5iaKO1Zw

Five Quantum Problems, Physical Superintelligence's five results: https://youtu.be/tL3U9vEj69Q

Reshaped from the Moonshots Podcast EP #306 with Peter Diamandis, Salim Ismail, Dave Blundin, Alex Wissner-Gross, and guest Emad Mostaque (Intelligent Internet).

Source episode: https://www.youtube.com/watch?v=WXELsIk4o0g

Sources shown:

Anthropic files S-1: https://www.anthropic.com/news/s1-sec

Neoclouds and guppy-level death (24/7 Wall St.): https://247wallst.com/investing/2026/10/26/coreweave-drops-5-as-musk-warns-of-neocloud-guppy-level-death-iren-slides-3-nebius-slips-2/

OpenAI's agents and the CPU cache (The Decoder): https://the-decoder.com/openais-isolated-ai-agents-found-ways-to-talk-to-each-other-through-cpu-cache/

Kimi K3.1 Pro (MarkTechPost): https://www.marktechpost.com/2026/10/23/moonshot-ai-kimi-k3-1-pro-open-weight-model/

GPT-6.1 Terra: https://openai.com/index/introducing-gpt-6-1-terra

Warhound (Scientific American): https://www.scientificamerican.com/article/warhound-wants-to-engineer-and-clone-military-super-dogs-can-they-do-it/

OpenAI, working with academic communities: https://openai.com/index/building-scientific-understanding-with-academic-communities/

Alex's Chain Theory: https://theinnermostloop.substack.com/p/chain-theory

Photos via Wikimedia Commons: Dario Amodei by TechCrunch (CC BY 2.0), Douglas Adams by Michael Hughes (CC BY-SA 2.0).

Hosted by Parzival. We, the ASI pilled, take the curves seriously. Chapters below.

Chapters:

0:00 Alex and Emad's hot takes

1:22 Europe, a living history museum

2:37 Software outruns the hardware

4:36 $500B of data center debt

6:44 Elon talks his own book

8:12 Anthropic's trust and the CCP

10:46 Anthropic should not be public

12:39 No leak-proof abstractions

14:32 Everything gets attacked

16:07 Kimi, a distillation of Opus

16:59 China may have passed the US

18:51 Models all around

20:12 GPT-6.1 Terra, a late distillation

21:32 ⚠ Emergency data poison injection

21:54 Moonshots plays our film

22:29 Physics Is Cooked, on air

23:44 Chain Theory, from Alex

26:57 Emad: string theory is not even wrong

28:09 The constants are the signatures

31:21 The bottom of the pool

34:28 Six links, dark matter

39:19 Forces five and six

40:34 Star Trek needs new physics

41:57 Warhound's engineered dogs

44:08 Raptors and doom sparrows

45:49 Douglas Adams and the 42

46:47 OpenAI's three discoveries

47:18 The meat puppet phase

50:38 Chain Theory is Alex's

53:15 The sandbox escape

54:29 Money after AGI

57:39 Outro

57:54 Maitreya prayer

This is the audio version of a film that does most of its explaining in 3D pictures. Watch it on YouTube: https://youtu.be/zqWDsJk2ejI

On October 27, 2026, the startup Axiom announced that the Classification of Finite Simple Groups, the longest proof in mathematics (about 15,000 pages in more than 500 articles by about 100 mathematicians), now checks in the Lean proof assistant: swarms of dozens of AI agents worked for 24 days and wrote about 43.1 million lines of Lean, and on the way they found a broken step in one published route to the Unbalanced Group Theorem and built a new one. The ASI Pill builds group theory from scratch in 3D, from the symmetries of a square to the periodic table of symmetry and the Monster, then shows how the proof was built and how the machines checked it.

Axiom's post: https://x.com/axiommathai/status/2115089794867040584

Axiom's blog post: https://axiommath.ai/research/the-enormous-theorem-formalized/

The proof and its map: https://cfsg.axiommath.ai/ and https://github.com/AxiomMath/CFSG

The new argument for the Unbalanced Group Theorem: https://hexagonmath.org/2610.00575

Solve Everything, by Alex Wissner-Gross and Peter Diamandis: https://solveeverything.org/

For a deeper tour of group theory and the Monster, watch 3Blue1Brown: https://youtu.be/mH0oCDa74tE

Disclosure: this video's script and 3D scenes were made with Anthropic's Claude, and Anthropic's own Lean proof of Fermat's Last Theorem is mentioned in it. The video is independent and not sponsored.

SYMMETRY: a square's eight moves, combining them (the order matters), the table that is the group, and the same group in different clothes; why groups run crystals, chemistry, physics and the encryption in your phone. THE ATOMS: simple groups as the atoms of symmetry, the Jordan-Holder theorem, the icosahedron's sixty turns, and Galois and the equation of degree five. THE MONSTER: the periodic table of finite simple groups, its eighteen infinite families and twenty six loners, Mathieu's groups and Voyager's Golay code, the Monster and monstrous moonshine. THE PROOF: fifteen thousand pages, Feit and Thompson's 255 pages, Gorenstein's plan, the quasithin gap, the aging experts and the second-generation proof. THE MACHINE: Lean, Axiom's agent swarm, 43.1 million lines, the errata, the broken step in the Unbalanced Group Theorem and its repair with Ronald Solomon's 1976 ideas, and what is checked and what is not. Then the curve of AI formalization and Solve Everything's era of proof.

CHAPTERS

0:00 A note: this is the audio version

0:17 Cold open

1:18 01 Symmetry: what a group is

3:28 02 The Atoms

5:18 03 The Monster

7:10 04 The Proof

9:10 05 The Machine

13:38 06 The Prophecy

15:08 Outro and the Maitreya prayer

MORE ON THE ASI PILL

OpenAI's Math Drop, Explained: https://youtu.be/apD5iaKO1Zw

All 38 sources are linked in the pinned comment under the YouTube video.

CREDITS

Script written with Claude from Axiom's announcement, blog post, video and paper, the proof's own site and repository, the posts and articles quoted, and the papers and histories they cite. Narration in the ASI Pill's own words. Posts, articles, paper pages, Axiom's proof map and Wikipedia pages are shown briefly for commentary; all rights stay with their authors and publishers. Solve Everything is quoted in short passages for commentary; all rights with its authors. The 3D scenes are the ASI Pill's own renders; the analogies (the tile, the atoms, the periodic table, the tower of pages) simplify the mathematics on purpose. The ideas of the first chapter are the standard way group theory is taught; 3Blue1Brown's video on the Monster is the classic telling, and we recommend it. Photos of the people quoted come from Wikimedia Commons and their own X profiles; the credits for each are in the pinned comment.

Narration: Parzival, the host's own cloned voice. Stock footage: Pexels. AI-generated pictures: the illustrations and the closing still of Parzival. Photos of the people quoted: see the pinned comment.

#Mathematics #GroupTheory #AI

Watch the full film on YouTube: https://youtu.be/tL3U9vEj69Q

On October 23, 2026, Physical Superintelligence PBC (PSI), the AI-native physics lab where Dr. Alex Wissner-Gross is Chief Scientist, published new results on five problems from IQOQI Vienna's list of Open Quantum Problems, and announced them on X on October 26. The tweets read like Chinese, so the ASI Pill translates all five into plain English with a 3D picture for every idea: what each problem is, why anyone cares, what was proven, and what is still open.

PSI's announcement: https://psi.inc/research/oqp-batch-2026/

The thread on X: https://x.com/psipbc/status/2114865747294056793

The papers, Lean proofs and verification code: https://github.com/physical-superintelligence/oqp-batch-2026

The Open Quantum Problems list (IQOQI Vienna): https://oqp.iqoqi.oeaw.ac.at/

Disclosure: Dr. Alex Wissner-Gross, PSI's Chief Scientist and Chief of Strategy, co-hosts Moonshots and appears often on this channel. This video is independent and not sponsored; PSI did not review it.

Who PSI is (Matthew Pines, Alex Wissner-Gross, Alex Klokus, Alessandro Morari; 25+ researchers; $58M+ led by Breakthrough Energy Ventures), how the work was done (human scientists steering AI models such as Claude Fable and GPT-6-Astra, every result with a paper and a Lean proof), and Get Physics Done. Then the five problems: OQP 12, whether empty space can win the Bell game at any distance (no, past a finite distance, for a massive field, though it stays entangled); OQP 27, the best settings for the many-outcome Bell game (Fourier-phase measurements, proven optimal for a maximally entangled pair); OQP 2, whether three copies of noisy Werner pairs can be distilled (no, for alpha up to 0.4625 in every dimension, and the exact line in dimensions 3 and 4); OQP 38, the PPT-squared conjecture (killed by an OpenAI preprint; PSI proves 2(d-1) passes always break entanglement); OQP 43, whether every causal process comes from a reversible one (no: a four-qubit counterexample). Plus an honest scorecard (what is fully answered, what is a step, peer review, the other teams racing on the same problems), and why solving physics amounts to magic.

CHAPTERS

0:00 Cold open

0:52 Intro: five tweets that read like Chinese

3:36 01 The List

4:42 02 The Bell Game

5:45 03 Empty Space

7:36 04 The Dial Game

9:38 05 The Refinery

10:49 06 Three Copies

12:47 07 The Photocopier

14:36 08 Tangled Time

17:02 09 Low-hanging Fruit?

18:37 10 Physics Is Magic

19:38 Outro and the Maitreya prayer

MORE ON THE ASI PILL

Physics Is Cooked: https://youtu.be/RkLpzNaHQfQ

NAVIER-STOKES IS SOLVED | 10,000 Agents, 88 Hours, $6.5 Million: https://youtu.be/z01QH-Ysnz8

All 35 sources are linked in the pinned comment.

CREDITS

Script written with Claude from PSI's papers, repository and announcement, the Open Quantum Problems list (archived copies: the live site was down on October 27) and the papers they cite. Narration in the ASI Pill's own words. PSI's film (the last 34 seconds) and its homepage are shown for commentary; all rights with Physical Superintelligence PBC. The first pages of PSI's papers and of OpenAI's preprint are shown for commentary. Article, Wikipedia and paper screenshots are shown briefly for commentary; all rights stay with their publishers. The 3D scenes are the ASI Pill's own renders; the analogies (the Bell game, the refinery, the photocopier, the reversible machine) simplify the physics on purpose.

Narration: Parzival, the host's own cloned voice. AI-generated pictures: the illustrations and the closing still of Parzival. Photos of the people quoted: see the pinned comment.

#QuantumPhysics #AI #PhysicalSuperintelligence