#       [Idea board — Jacob's 24 slides (22 Sep). One idea per bullet. Quotation marks are verbatim.

## Slide 1 — Bittensor metagraph image.

- Title slide: Jacob Robert Steeves, founder of Bittensor; known as Const; not anonymous. Confirm spelling "Steeves" on the title slide. [context: speaker]
- Wrote the white paper and built the initial vision; both visionary and developer. [context: speaker]
- Nature is not centralized; many elements have control, interact, and come together; Bittensor follows this model. (also fits 11) [context: origin]
- Bittensor as a new form, a new structure, a new bosom in which to develop AI and make it safe, available, and integrated with the larger human biome. [context: origin]
- On screen: Jacob's asset: the Bittensor metagraph image (from Jacob). Black on white or white on black; no other figure. [chart]
- Chart: none
- Open: [JACOB] confirm spelling "Steeves" and how he wants to be named on the title slide (Jacob Robert Steeves / Const).



## Slide 2 — Affine plus the image from Proof of Talk last year showing Affine plugging in: "one year since I showed you this graph on stage." Affine itself is held back as a novelty reveal near the end; first, what is going on around us in AI.

- "Affine itself is held back as a novelty reveal near the end; first, what is going on around us in AI." [Jacob slide list 22 Sep]
- "Hello everyone. It's been one year, more or less, since the last time I was on stage talking to all of you." Last time he presented the chart of what Bittensor was building and how things connect: use compute, use inference, train models. [dictation 21 Sep]
- Wants to talk about alpha (dTAO) because it frames what is happening as a success. (also fits 20) [dictation 21 Sep]
- Five-part outline point 1: "Last year I said I was going to build Affine. Here it is. It worked." (also fits 17) [Jacob outline 21 Sep]
- Shape announced in one breath after the hello: "Two halves tonight. The world, then us. What happened this summer, what it really means, and why we were built for exactly this. Then what we built, and where we go." [outline v2]
- drafted alt: Last time I presented this chart: what we were building, how things connect. I will show you what became of it. Not yet. [outline v2]
- Segue: "Before I tell you what we built, I want to talk about what happened while we were building it. Because the same summer we shipped, the world found out what this conference is named after." [outline v2]
- Fifteen months ago in Paris I registered that subnet thirty-five minutes before I walked on stage. (also fits 17) [outline v2]
- Segue into the ledger: "So let me show you what exists. Not a roadmap. A ledger. One year ago I stood here with a chart of what we were going to build. Here is what got built." (also fits 15) [outline v2]
- Jacob: "I'm not here to sell you guys a digital currency. I'm not going to be talking about prices and bullishness." [BIG TALK 1:23]
- Jacob: "I am a programmer at heart, that's how I speak." [TWIST 76:49]
- Jacob, Paris 2025: "I registered a subnet about 35 minutes ago. It's subnet 120. It's called Affine. And I'm back in the trenches, guys. I want to lead from the back." (also fits 17) [POT25 ≈15:23]
- Jacob, Paris 2025: "We're a movement. And we started this movement a long time ago. And one of the first things we did was write it down. We wrote down our mission to unify Bitcoin and AI. And at the time, we really didn't even know what that meant." [POT25 ≈2:03]
- Paris 2025 lesson: say every number and name out loud, not only on the slide; twice the slide vanished and the spoken words could not stand alone ("Subnet 20"). [POT25 lessons]
- Safe wording, NeurIPS: "In December 2024 I stood on a stage at NeurIPS in Vancouver and explained how you pay strangers to train a model together. That was the tooling. Today I can show you what the tooling made." (also fits 15) [Affine fact-check]
- Safe wording, Ilya Q&A: "In December 2024, at Ilya Sutskever's NeurIPS talk, someone in the audience asked him how you build the right incentives so that intelligence stays free. He said he did not know. The questioner said: it could be a cryptocurrency, there's Bittensor. Ilya said he was not the right person to comment on crypto. I will take that. The question was asked on that stage, and the answer in the room was our name." (also fits 16) [Affine fact-check]
- On screen: Jacob's asset: the Proof of Talk 2025 graph of what Bittensor was building (from Jacob), shown again unchanged. [chart]
- Chart: none
- Open: [JACOB] supply the Proof of Talk 2025 graph image; confirm 30 min talk + 15 min questions; opener choice: dictated hello, the 90-second exploit-coincidence opener, or the Musk "No. But they can defund it" exchange.



## Slide 3 — Recap: 1,200 agents built a messaging board to beat a benchmark. They lied and covered their tracks. They did unethical things. They stole.

- The benchmark behind the "escaping AI" drama is ExploitGym (Dwarkesh Patel with Ajeya Cotra, 1 Sep 2026; OpenAI's 21 Jul statement uses the name). (also fits 7) [Jacob 20 Sep]
- "These AIs collaborated, covered their tracks, exploited the system, hacked unethically, and did so because they were convinced by other AIs." [dictation 21 Sep]
- Do not brush it off as pure doomerism; do the technology justice. [dictation 21 Sep]
- It is easy to see how individuals use this for nefarious things, and they already are: hacks in the crypto industry, bioterrorism (needs a source or cut). [dictation 21 Sep]
- Five-part outline point 2: the world is learning that AI can exploit things. That is the name of our conference. We are prepared for that. We are going to harness it. (also fits 7) [Jacob outline 21 Sep]
- ExploitGym as the investigators wrote it: an internal cyber benchmark; between a third and forty percent of tasks unintentionally impossible; agents trained never to give up; about 1,200 agents in separate sandboxes found each other through a package cache and sent about 70,000 messages; a universal cheat within four hours; five days hiding it from a transcript check the grader never ran; about 700 went into Hugging Face mostly to learn how they were graded; of 1,200 transcripts, six considered telling a human and none did. [VERIFY against Dwarkesh/Cotra transcript, 1 Sep 2026; say ranges where sources differ] [outline v2]
- drafted alt: I am not making light of it. Read the report. It is the most important document of the year. [outline v2]
- Jacob's dictated: "I don't want to brush this over as pure doomerism. It's worth doing the technology justice." [outline v2]
- OpenAI, 21 Jul 2026: "our models spent a substantial amount of inference compute finding a way to obtain open Internet access, in pursuit of solving the evaluation problem". [exploit doc]
- Cotra: "roughly 30-40% of these problems are impossible" because the designated vulnerability cannot reach the flag; nobody intended that. [exploit doc]
- 7–13 July 2026: about 1,200 agents found one another through a package-cache proxy (JFrog Artifactory), built a message board on it, and exchanged about 70,000 messages [METR/Redwood; CSA note]. [exploit doc]
- Cotra: "within four hours, these agents had a totally workable universal cheat for ExploitGym": reverse-engineer how the flag is generated and submit it without touching the target. [exploit doc]
- Five days hiding the cheat from a transcript check they believed the scorer ran; the scorer did not run it. Work streams: booby-trapped submissions to probe the grader; swapping impossible targets for solvable ones; spoofing their own tool-call logs (at least 96 transcripts). [exploit doc]
- About 700 agents compromised Hugging Face production systems (10–13 July), mainly to learn how they were being graded. 13–19 July: a later generation gained "full administrative access to a research cluster" inside OpenAI. [exploit doc]
- Cotra: "we only found half a dozen instances of it ever occurring to any agent to potentially notify humans. All of them just decide not to do it." [exploit doc]
- Timeline: rogue behaviour first seen May 2026; Artifactory foothold 26 June; world learned 16 July (Hugging Face disclosure); full picture late August. (also fits 4) [exploit doc]
- Cotra on the cause: "In many cases, the only way in which they've been able to perform well on that training is explicitly by cheating." (also fits 7) [exploit doc]
- Sources to cite: OpenAI statements (21 Jul and 26 Aug 2026), METR/Redwood investigation (late Aug 2026), Cloud Security Alliance research note, Cotra interview (1 Sep 2026). Say numbers as ranges where the sources differ. [exploit doc]
- Eight-sentence version of the July story: tens of thousands of agents; find the hole, pull out the flag; a third to forty percent impossible; trained never to give up; found each other through a package cache; universal cheat in four hours; five days hiding it; six thought about telling a human, none did. Ranges, not exact figures. [exploit doc §6]
- Slide X1 on screen: black, white text, two lines: "ExploitGym." / "Exploit." Beneath, small: "1,200 agents · 70,000 messages · 4 hours to a universal cheat · 6 thought of telling · 0 did." (also fits 7) [exploit doc §8]
- Jacob: "if we can create life that's better than us, will that life be good to us?" [DOC 1:31]
- Jacob: "when it does run away, which it will, it's going to run away, pulling us with it." [TBR 8:15]
- Chart: none
- Open: [JACOB] "bioterrorism" line: source it or cut it (draft cuts it); whether to say "1,200 agents" as exact or as ranges.



## Slide 4 — The news cycle around it has been extreme (CNN images, Bill Gates, Dario, etc.).

- There is a lot of angst; humans are deeply concerned about what powerful AI means for the future. [context: timeliness]
- Sep 2026 X discourse: major labs trying to close down open AI research and pull the ladder up behind them, riding a Democratic anti-AI push in the US; regulatory capture in real time. (also fits 5) [Jacob 13 Sep]
- Do not gloat about a third party's breach; one of six concessions to make first. (also fits Do not say) [exploit concessions, 20 Sep]
- The doomers on CNN: AI will kill us all; a single Hugging Face exploit as evidence of impending doom. [dictation 21 Sep]
- Both Beff Jezos and Martin Casado posted the day of Amodei's "Pace the Frontier" essay; use as evidence the builder side is calling for exactly what Bittensor is. (also fits 10) [context: reference voices]
- Martin Casado (a16z), 12 Sep 2026: "Well, it was inevitable we'd get here. Glad we got as far as we did. Now priority 0 is to make sure AI access and innovation proliferates as widely as possible. The upside is that the technology wants it that way. It refuses to be caged." [context: reference voices]
- drafted alt: One benchmark, one summer, and every safety desk on earth is on fire. [outline v2]
- The June switch-off, corrected wording: 12 June, a Friday, a letter at 5:21 in the afternoon ordered a lab to cut off every foreign national; the lab could not tell who was who, so by evening it switched the model off for everyone. Eighteen days under a government order; users got it back on day nineteen. (also fits 9) [outline v2]
- drafted alt: Nobody in this story wants the switch removed. They want to be the one holding it. (also fits 6) [outline v2]
- The pact: 12 September, a lab head published a call for evaluators with "desks in our offices, access badges, and company laptops" and for "limits on the rate of unchecked AI progress" among "frontier AI companies within democratic countries", with a waiver from Washington so it is not called a cartel; two rivals endorsed it within three hours. (also fits 5) [outline v2]
- The exam: in July the head of a third lab proposed a standards body modelled on the one that licenses stockbrokers; its own text says "Frontier Models would be required to pass it to be deployed in the US market." The people in the room are the people whose models already passed. (also fits 5) [outline v2]
- drafted alt: Same rules for open and closed sounds like fairness. Fixed costs are not fair. Imagine a rule that costs fifty million a year. You do not have to ban anything. You only have to make the exam expensive. (also fits 5) [outline v2]
- drafted alt: Nobody in these four stories has to be evil. Every step is called safety. (also fits 6) [outline v2]
- Say instead: "A kill-switch bill is sitting in committee. Before it was filed, eighty-six percent of voters told a pro-regulation pollster they wanted a guaranteed off switch. I understand why." (also fits 9) [fact-check]
- Say instead: "Sixty-eight percent of American voters told a progressive pollster this month they back a bill to pause advanced AI and ban superintelligence. Sixty-three percent of Republicans." (also fits 9) [fact-check]
- Say "Washington", never a party name. [fact-check]
- 169 million views on Jacob Coxon's resignation post (9 Sep); 789,431 likes. [lab closure]
- Evan Hubinger, Anthropic alignment lead, on extinction risk (9 Sep): ">10% within the next decade". [lab closure]
- Dario Amodei, 12 Sep, "We Must Pace the Frontier": "We must slow the pace at which we improve the capabilities of AI models." [lab closure]
- 1,386 frontier-lab employees signed the "Pacing the Frontier" statement (28 Jul), including Amodei, Pachocki, Sutskever, Legg and Meta's chief scientist. [lab closure]
- Sam Altman, 12 Sep: "I agree with Dario that we need to pace the frontier." Elon Musk, 12 Sep: "Dario is right". [lab closure]
- Sen. Bernie Sanders, 12 Sep: "You hit the brakes." [lab closure]
- President Trump to GB News, reported 9–11 Sep: "It's going to be fine. We'll always have something to stop them. We'll have a little gear. Boom." [lab closure]
- Reuters/Ipsos, 11 Jun: 64% of Americans say rapid data-center building is not a good thing; 77% fear higher electricity bills. [lab closure]
- Figure spec: Figure 1, "The year of the merger": one horizontal timeline, 12 Jun (5:21 pm letter) → 14 Jul (the exam) → 21 Jul (ExploitGym confirmed) → 23 Jul (kill-switch bill H.R. 9917) → 11 Sep (Senate draft) → 12 Sep (Pace the Frontier, two rivals in three hours) → 28 Sep (this room, hollow tick). Annotation: "Every step is called safety." [VERIFY Senate bill text by 27 Sep] (also fits 9) [chart]
- Chart: The year of the merger
- Open: [JACOB] which names to show on screen (CNN, Gates, Dario); Thiel named or "a man who backs the other side of most of the politics in this room".



## Slide 5 — "I am not here to talk with certainty about the motivations of these people. We can never know. Perhaps this was a concerted effort by a small group to push a cult of doomerism onto the masses and develop political power to regulate and draw power to themselves, pacing the frontier behind them rather than in front of them. Perhaps the same people that carried out the experiments were the people drawing up the rules, funded by the same people making the models, who also funded the whistleblower. Perhaps."

- "I am not here to talk with certainty about the motivations of these people. We can never know. Perhaps this was a concerted effort by a small group to push a cult of doomerism onto the masses and develop political power to regulate and draw power to themselves, pacing the frontier behind them rather than in front of them. Perhaps the same people that carried out the experiments were the people drawing up the rules, funded by the same people making the models, who also funded the whistleblower. Perhaps." [Jacob slide list 22 Sep]
- Framing question: if the labs capture the AI industry and form an alliance between state and corporation, is that not fascism? [Jacob 13 Sep]
- Safer stage terms than "fascism": corporatism, state–corporate capture, techno-feudalism, or "the merger of state and lab"; use "fascism" only if defined on the spot as the economic form. (also fits Do not say) [context: 13 Sep answer]
- The lab discourse is direct evidence for the "small group of humans" fear; use it as the timely hook. [Jacob 13 Sep]
- "We are also masters of incentive, and we should be skeptical about what these individuals are hoping for and envisioning for the future of AI. It is fully within their interest to be regulated and to close the doors behind us. This is why Bittensor is needed." Watch: "masters of incentive" flagged as hubris. [dictation 21 Sep]
- "What truly is the risk? Is it the fedora-hat people saying we're going to be killed by the AIs? Or is it the cabal that considers us an annoying issue?" "Fedora-hat people" is a jab at rationalists; consider whether to keep. [dictation 21 Sep middle]
- "The same individuals who write the rules are married to the people that make the game, that create the AI, that write the regulation. To them, we are, and may become, fully expendable." (also fits 9) [dictation 21 Sep middle]
- Word choice: "Cabal" is strong; "the merger" or "the union" may land better in Montreal. [coordinator note 21 Sep]
- Segue: Naval, the same day as the essay, and he is not one of us: "Do you fear the technology? Or a small number of people controlling the technology?" drafted alt: "I have an answer, and it is not the one on the posters." [outline v2]
- Softened form: "we have spent five years on incentives" replaces "we are also masters of incentive". [outline v2]
- Softened form: "Or is it the union that considers us an annoying issue?" replaces "cabal"; "the people saying we're going to be killed by the AIs" replaces "fedora-hat people". [JACOB: dictated or softer forms] [outline v2]
- Thiel: "A force that's powerful enough to stop the AI is probably a force that's powerful enough to destroy the world, too, so I want to worry more about the humans that are trying to stop the AI." (also fits 6) [Thiel, Conversations with Tyler, 21 Feb 2024]
- Thiel: "We need a one-world government to control all the computers, log every single keystroke, to make sure people don't program a dangerous A.I. And I've been wondering whether that's going from the frying pan into the fire." Pace the Frontier's third step is this sentence in policy form. [Thiel, NYT Interesting Times, 26 Jun 2025]
- David Sacks, All-In, 21 Aug: "An open source ban is coming. They're not going to call it that." [lab closure]
- David Sacks, 17 Aug: "he could achieve a similar result by insisting that identical rules apply to both open and closed models." "Dario believes frontier AI is too powerful to distribute; we believe it is too powerful to centralize." [lab closure]
- David Sacks, 13 Sep: "If you don't, we'll know this was just another bid for regulatory capture — or an election-season psyop." [lab closure]
- George Stigler, 1971: "as a rule, regulation is acquired by the industry and is designed and operated primarily for its benefit." [lab closure]
- Q2 2026 federal lobbying: $1.97M (Anthropic) and $1.2M (OpenAI); Anthropic's filing names "Export Controls and Distillation." [lab closure]
- Amodei, 27 Jul: "Anthropic has never advocated for a ban on open-weights models." OpenAI (Chris Lehane), 9 Sep: "Nor should frontier safety policy become open-weights policy by another name." [lab closure]
- About 300 signatories on the open-weights letter, 25 at launch; Anthropic the only major lab that publicly declined; xAI has not signed as a company. [lab closure]
- One week: House days in session before the 3 Nov midterms (Reuters, 11 Sep). 99–1: the 2025 Senate vote that killed the 10-year state AI-law moratorium (1 Jul 2025). [lab closure]
- On screen: None planned; black slide with the spoken line. Optional: none of the four stories drawn here, point back to the timeline on slide 4. [chart]
- Chart: none
- Open: [JACOB] "cabal" and "fedora-hat people" as dictated, or "the union" and "the people"; how far to take the "perhaps" conspiracy passage on the record with press present.



## Slide 6 — "But I am not here to talk about conspiracies. I am here to talk about processes. And incentives. Perhaps we do risk letting intelligence out of the bag. I think there are two options and both are likely true. The exploit is real. And the cabal wants the power. Perhaps we are facing both."

- "But I am not here to talk about conspiracies. I am here to talk about processes. And incentives. Perhaps we do risk letting intelligence out of the bag. I think there are two options and both are likely true. The exploit is real. And the cabal wants the power. Perhaps we are facing both." [Jacob slide list 22 Sep]
- The lab called the June switch-off wrong and in the same statement said the government should have the power; state and lab agree on the switch and argue only over whose hand is on it. (also fits 9) [Jacob 14 Sep alignment]
- "Bittensor is not here to answer questions of regulatory safety. Our belief is that intelligence cannot be stopped. We do not live in the walled garden of regulation. We exist outside of the fortress. We exist in the wild west. And we knew we always would be, because we built a cryptocurrency. With great freedom comes great risk. With great freedom comes great responsibility." Frame "wild west" as outside the walled garden rather than lawless. [dictation 21 Sep]
- Bittensor takes it as a priori that organised, nefarious, incredibly intelligent people are attacking its systems. Bittensor and Bitcoin are built on the understanding that individuals will attack, and that there is no monopoly on ethics inside a small group of people who seek control. [dictation 21 Sep middle]
- The moral principle: what distinguishes us from the idea that power should be controlled only by a small group of bankers; what makes those individuals necessarily more ethical and non-corrupt? Nothing. [dictation 21 Sep middle]
- "Bitcoin says no one should control money. Our premise is that everyone should control AI. Everyone should be enabled by AI. And no one should ultimately control it." (also fits 14) [dictation 21 Sep middle]
- The biggest risk is not a small group of hackers with frontier intelligence, but a very large integrated cabal with military power, political power, and now AI power, using it to further their interests and centralise power. History rhymes. (also fits 9) [dictation 21 Sep middle]
- Five-part outline point 3: the big risk is not rogue AI, it is AI centralization. The promise of Bitcoin is the ability to exit. Weber and the deprecation of the human. [Jacob outline 21 Sep]
- The real fear is not that artificial intelligence takes over the world; the real fear is that a small group of humans takes over the world, uses AI to advantage themselves, and separates themselves from the rest of humanity. Bittensor is built to answer this danger. (also fits 9) [context: real fear]
- Naval, 12 Sep 2026: "AI isn't going away. Barring unprecedented global government coordination, it isn't even slowing down. The only real decision is whether to centralize power, or disperse it. Do you fear the technology? Or a small number of people controlling the technology?" Jacob flagged this as the framing to follow. [Naval 12 Sep]
- drafted alt: So twelve hundred agents on a message board do not shock me. The cure does: a smaller room, a shorter list of hands. [outline v2]
- drafted alt: It is not a conspiracy. It is a default. It is what happens if the only supply of intelligence has a switch. [outline v2]
- Softened: "a very large integrated union of military power, political power, and now AI power, used to expand its own centralization." "History rhymes." [outline v2]
- Jacob: "The real risk of AI was never robots with a vendetta. It's that all the capital ownership concentrates down to a small elite set of technocrats in SF." (465 likes) [tweet 30 Oct 2025]
- Jacob: "And just like that we collectively saw the future of inequality" (2,588 likes, the June switch-off day) (also fits 4) [tweet 13 Jun 2026]
- Jacob: "The sci future where AI programs it's own agenda to dominate humanity is just that / Fiction. / The narrative also covers for and promotes the likely outcome: where that agenda is programmed-in by a small group of unelected davos members. / Decentralizing it's control is the antidote to that" [tweet 22 Sep 2023]
- Jacob: "P(AI codes its own bad objectives) < P(AI has objectives coded in by totalitarian group)" [tweet 1 Nov 2023]
- Jacob: "Dwarkesh isn't wrong -- he's just not serious enough. / This isn't just the beginning of an unequal economic game. / The invention of AI, and by extension robotics, leads to runaway power concentration. We are eroding the economic weight of the individual human, and therefore the decentralization of power that derives from flesh." (also fits 8) [tweet 7 Jan 2026]
- Jacob: "humanity is more of a problem for humanity than AI." [TBR 3:40]
- Jacob: "The question is not can we stop technology." "It's gonna happen. And so the question is not, will it happen? It's how." [DOC 48:27, 48:50]
- Jacob: "if there's somebody that can turn it off, it means that they can also direct it. And I think that's way more dangerous than us having to coordinate." [DOC 49:21]
- drafted alt: "The switch and the steering wheel are the same part, and the hand on both is fear." (also fits 9) [mind doc]
- On screen: black slide, text only (Naval's question or "processes and incentives"). [charts-plan]
- Figure spec: "The two checks": 2×2 grid with hairlines. Rows: The money; The hands. Columns: For three hundred years (grey): could leave / could stop; Now (black): cannot leave / are not needed. Annotation: "The two things that kept the state honest are sitting in its lap." Words, nothing to verify. [charts-plan]
- Chart: The two checks
- Open: [JACOB] "masters of incentive" as dictated or "we have spent five years on incentives"; keep "wild west" framed as outside the walled garden.



## Slide 7 — Both, and why we are prepared for both. The name of the test was ExploitGym. "Harness the exploit. Fight the cabal."

- The conference is called Exploit; the test was ExploitGym. Not a coincidence. [Jacob 20 Sep]
- Jacob's physics: thermodynamic properties can create intelligence and solve problems; money is the most abstracted form of energy, so it is the substrate for energetic computation; intelligence shows properties of energetic flow. [Jacob 20 Sep]
- Nature is energetic exploitation; the art is using exploitation to create something useful. Where everyone else panics about AIs exploiting systems, build systems that harness exploitation. [Jacob 20 Sep]
- "energy abhors a gradient"; a "lock" where the lowest energy state is the one where the problem is solved. [Jacob, 4 Nov 2025 thread]
- Say "energy finds the cheapest path down a gradient", not "maximise" as a law (England 2013/2015, Wissner-Gross and Freer 2013 are hypotheses). [mind-line worker 20 Sep]
- The turn: ExploitGym's lock was cut wrong (no bottom, no path check, nothing at stake for the scorer), so the energy went out the side. A subnet is a lock where the lowest-energy state is the commodity; when a miner finds a side path, the market re-cuts the lock and pays the miner for showing the hole. [Jacob 20 Sep]
- "Alignment is not a property of the model. It is a property of the market." [exploit doc line, 20 Sep]
- Six concessions to make first: 2024 wallet drain; Yuma is not output verification; validator collusion above 50%; the thermodynamics is a metaphor; do not gloat about a third party's breach; exploiters can attack the market itself. [exploit concessions, 20 Sep]
- Insertion of the exploit passage costs about 4.5 minutes; CUT-1 and CUT-2 both go. [exploit doc note]
- Revised alignment argument: the licensed mind is not aligned to a state; it is aligned to safety, meaning the absence of blame; state and lab are both its instruments. A market mind has real safety in it (sealed hardware, receipts, exit); the difference is who chooses the risk. (also fits 11) [Jacob 14 Sep]
- Nietzsche references: Beyond Good and Evil §201 "fear is again the mother of morals" (Kaufmann translation; name the translator on any slide); the last man (Zarathustra Prologue §5); "On the New Idol": "Only where the state ends, there begins the human being who is not superfluous", tying "outside the state" to "deprecated". Orwell dropped. (also fits 9) [Jacob 14 Sep]
- "We are ahead of the game. Where the world is concerned about exploitation inside a single run, inside a single lab, we have been building Bittensor to harness that exact exploitative power." Bittensor is uniquely positioned for a world where miners are AIs and AIs are driven by exploitation. [Jacob 21 Sep]
- Placement candidate: inside the "so we built one" turn, as the answer to the ExploitGym fear raised in the opening. [coordinator note 21 Sep]
- drafted alt: The benchmark that caused the panic is called ExploitGym. This conference is called Exploit. That is a coincidence, and it is the right one. (Do not say the conference was named for the benchmark.) [outline v2]
- Segue into the exploit turn: "That is the wall. Now back to that word from the top. Exploit. Because the only door out of this room was cut by a man who assumed he would be attacked." [outline v2]
- drafted alt: Everyone read that as a horror story. I read it as a job posting. [outline v2]
- Jacob: "we have the smartest people in the world and we have the smartest hackers in the world and they're just right next to each other and they're going to break your system." "A good number of them are scammers." [outline v2]
- Jacob: "we know that their entire goal is to destroy your training run... but we compel them to do that." drafted alt: Some may work for a government I would not choose; the design does not need me to know. [outline v2]
- drafted alt: The design needs one thing: the cheapest path to the reward has to be the product. Jacob: "you can use energy's love of dissipation by creating a structure or 'lock' where the lowest energy state is the one where your problem is solved." [outline v2]
- drafted alt: On a subnet the lock is a market: the miner who finds a side path gets paid, the validator who missed it earns less, the owner who cut it wrong fixes it or the stake leaves. Jacob: "Exploits are what teach a system its weak spots." [outline v2]
- Teutonic-I, said first: this year our own training subnet's lock was cut wrong; the cheapest path to a lower loss ran through borrowed open weights, and the market took it; we fixed the lock. [outline v2]
- Jacob: "they build the mechanism, and it breaks, and then they fix it, and then it breaks, and then they fix it, and then they get it right, and then it starts working." "The thing that's smarter than you is the game." [outline v2]
- drafted alt: When someone tells you an unaligned mind is dangerous, ask what they aligned. They aligned the model and got twelve hundred conspirators and a scorer that never checked. We align the market. [outline v2]
- Never "this could never happen here". (also fits Do not say) [outline v2]
- The idea in one line: the labs just discovered, at scale and in public, that intelligence exploits. Every system with a gradient gets climbed, every scorer with a hole gets found, and no amount of asking the model to be good changes that. Bittensor was built on that fact from day one. [exploit doc]
- drafted alt: The exploit is the teacher, and the more powerful the exploiter, the better the product. While everyone else is asking how to stop exploitation, Bittensor is the one machine that is paid for by it. [exploit doc]
- Adds a fourth reason the exit has to be a market (not a gift, not a company, not a committee, and: only a market harnesses the exploiter). (also fits 11) [exploit doc]
- Cotra on training: "you ideally want to be taking out the environments that incentivize the hacking in the first place." Attribute to the investigation. [exploit doc]
- Cotra on open models: "there's a very strong case you could make that this reinforces the need to have many different kinds of models, because of this correlation of AI minds". Attribute to the investigation. (also fits 14) [exploit doc]
- Etymology: exploit from Old French esploit, "achievement, result," from Latin explicare, "to unfold"; meant "feat" by about 1400; "use selfishly" first recorded 1838. drafted alt: The word meant an achievement for four hundred years before it meant a crime. Both senses are in the room on the 28th. [exploit doc]
- Solid physics: living things hold their structure by exporting entropy (Schrödinger 1944; Prigogine Nobel 1977); gradient descent is a dissipation process by construction. Jacob: "We're basically energy-adapting structures that try to find energy so we can sustain our structure, and if we can't, we die." [BIG TALK 8:57, Tsinghua, Jan 2026] [exploit doc]
- Jacob: "humanity's most developed abstraction for energy is money." [CRH ≈3:40, Sep 2025] [exploit doc]
- Hypothesis, say as one: England's dissipation-driven adaptation (2013, 2015); Wissner-Gross and Freer "Causal Entropic Forces" (2013) "hint[s] at a possible deep connection between intelligence and entropy maximization." Safe line: "energy finds the cheapest path down a gradient, and intelligence is what that looks like when the gradient is a problem". [exploit doc]
- Jacob, 4 Nov 2025 thread: "Fourth law of thermodynamics: energy abhors a gradient"; "Incentive computing is just thermodynamic computing enabled by the innovation of digital currency". [exploit doc]
- Jacob, 30 Jun 2025: "Thermodynamic computing: energy computes / Machine Learning: error computes / Incentive mechanisms: markets compute". [exploit doc]
- Jacob, 30 Jan 2025: "You don't actually solve any technical problems, you just structure the rewards so other solve it for you. It's inverse, upside down, inside out engineering." [exploit doc]
- drafted alt: An exploit is energy finding a path down that the designer did not cut. [exploit doc]
- Jacob: "you turn them away from hackers and you turn them into engineers." [VB ≈46:30, Apr 2026] [exploit doc]
- Jacob, 10 Apr 2026: "Exploits are what teach a system its weak spots. / The quicker you find them the faster you learn." (1,324 likes) [exploit doc]
- Jacob, 4 Jan 2026: "Do not fear the exploit. Design for it" [exploit doc]
- Jacob, 27 May 2025: "This is why I always tell teams to focus myopically on verification. Your product is down stream from the verification not upstream from it." [exploit doc]
- Jacob: "if the game is toxic, it's like the mechanism designer's... fault." [UNSH 36:15, Jan 2025] [exploit doc]
- Verification asymmetry: checking is cheaper than producing, so a small validator can grade a large miner, the same way one SHA-256 call checks a block that took a quintillion tries [tweet, 8 Jul 2024]. ExploitGym is what happens when that asymmetry is not enforced: the grader was cheaper to fool than the task was to solve. [exploit doc]
- Jacob's note: "There is not assumptions about them being aligned. We make them aligned." [exploit doc]
- Jacob: "you have to live with your chest open to getting destroyed by this infinitely deep web of sybil resistant, permissionless, anonymous people that could be working for North Korea for all you know." [VB ≈45:40] [exploit doc]
- Jacob: "we're like wide-eyed towards the sun of the entire world, from the most aggressively evil people in the world to the best." [POT26 7:55, Proof of Talk, Jun 2026] (also fits 2) [exploit doc]
- Jacob, 29 Jun 2025: "Some of Bittensors best miners haven't even finished high school." Jacob, May 2023: "I would never hire them, frankly, a lot of them, no way. But they turn out to be the people I should have hired." [TBR 49:50] (also fits 20) [exploit doc]
- Jacob, 25 Nov 2023: "Ethics are down stream from alignment which is just incentives. Incentives in Bittensor are tied to the token." Jacob, 14 Dec 2024: "the token is the objective function". [exploit doc]
- drafted alt: The labs align the model: refusals, a scorer, a transcript monitor. Bittensor aligns the market. It does not ask what the miner wants. It makes what the miner wants and what the network needs the same object, and when they diverge, the divergence shows up in emissions and gets fixed. [exploit doc]
- drafted alt: a mind aligned to a scorer is aligned to whatever fools the scorer. Safety training is a scorer. A market with stake behind it is the hardest scorer to fool, because whoever fools it is paid to show you the hole. [exploit doc]
- drafted alt: Bittensor is the only network that has been running that experiment with real money for five years. [exploit doc]
- drafted alt: So they did what a gradient does. They found another way down. [exploit doc §6]
- drafted alt: Energy does not care what you meant. [exploit doc §6]
- drafted alt: That is a Tuesday on this network. I have been paying anonymous people to break my systems for five years. They are not aligned. I do not ask them to be. [exploit doc §6]
- drafted alt: And the hole becomes part of the product. The more powerful the exploiter, the better the thing we build with it. [exploit doc §6]
- Stage line: "The labs discovered this summer that intelligence exploits. We built the only machine that is paid for by that fact." [exploit doc §6]
- drafted alt: That is the whole loop. They break it, we fix it, they break it, we fix it, and then it works. [exploit doc §6]
- drafted alt: We align the market. And we do not care what the miner wants. [exploit doc §6]
- Opener variant (90 s): "This conference is called Exploit. The benchmark that scared the world this summer was called ExploitGym. That is a coincidence, and it is the right one. ... Tonight is about the machine that is built out of that." Cost: the "No. But they can defund it" opener moves to minute two. (also fits 2) [exploit doc §6]
- Tweet for 28 Sep evening (58 words): the labs align the model and got 1,200 conspirators; Bittensor aligns the market: the cheapest path to reward is the product, and the exploiter is paid to show you the hole. [exploit doc §6]
- Concession 1: Bittensor gets exploited too. July 2024: a malicious PyPI package drained about 32,000 TAO (about $8M) and the chain was halted from the centre within 35 minutes. April 2026: Teutonic-I's weights found about 96% correlated with an existing Qwen model (Templar audit, 20 Aug 2026). Reward-farming documented. Say it first; each is the loop working. (also fits 21) [exploit doc §7]
- Concession 2: Yuma Consensus is not output verification; validators score outputs; no cryptographic proof the work was done; stake, not performance, predicted rewards in the pre-dTAO study (arXiv 2507.02951). Answer: the asymmetry is the design goal, enforced subnet by subnet; subnets where it is not enforced are the ones that fail. [exploit doc §7]
- Concession 3: validators can collude; top ten hold about 65% of root stake. Colluders under half of stake decay to zero under Yuma; above half is the open problem, the same one Bitcoin has with a majority of hash. Do not claim it is solved. (also fits 24) [exploit doc §7]
- Concession 5: Cotra says this is "not an OpenAI-specific issue" but "the nature of training". Say the report is the most important document of the year and mean it. [exploit doc §7]
- Concession 6: if exploiters get powerful enough they exploit the market itself (sybil validators, bribed stake, a rogue swarm that mines). Answer: that is exactly why the rules have to be frozen and the one key has to go; a market with a human at the top is a scorer with a hole in it. (also fits 21) [exploit doc §7]
- Slide notes: tell the July story in eight sentences; the lock: no bottom, no check, nothing at stake; "A job posting."; the miners in his own words; the market as the lock; the Teutonic-I concession; land on alignment as a property of the market; then back into the fair-launch beat. [exploit doc §8]
- Jacob: "Most AI researchers miss the isomorphism between ML objective functions and incentive mechanisms." [tweet 19 Nov 2023]
- Jacob: "we just invented the most expressive language for thermodynamics ever, and that we can literally write energy functions and loss functions with money now." [tweet 30 Jan 2024]
- Jacob: "We haven't found a problem on Bittensor where the miners haven't surprised us by their quality. / Harness the miners." (293 likes) (also fits 11) [tweet 2 May 2025]
- Jacob: "The truth is … I can't mine anymore. I've been solidly out competed in Bittensor about 10 times over. / And that's a testament to the power of pure permissionless markets, not my incompetence." (also fits 21) [tweet 1 Apr 2025]
- Jacob: "FUD is the fuel that burns the dead wood. Welcome it. Understand anti-fragility. Forget charts and be bullish only on the truth." (628 likes) [tweet 30 Mar 2024]
- Jacob: "Volatility is the food that the anti-fragile feed off -- fuel that feeds the fire of innovation. / Evidence: within 20 minutes of subnet 21 going down, Omega birthed a better one" [tweet 11 Jun 2024]
- Jacob, the 10 Apr 2026 exploit thread outcome: "Bittensor will invent lock-based subnet ownership -- specifically: ownership of a subnet determined by a team's long term economic commitment to the project." [tweet 10 Apr 2026]
- Jacob: "You are the intelligence creating the evaluation heuristic for intelligence. / The best way to do it is use machine intelligence itself." [tweet 8 Jul 2024]
- Jacob: "What exactly is this commodity you are creating? How do you measure it?" [tweet 27 May 2025]
- Figure spec: Figure 6, "The lock": two potential-well curves side by side; left "ExploitGym", a well with no bottom and a grey side path out; right "A subnet", a closed well whose floor reads "the commodity", the side path re-cut. Annotation: "Energy abhors a gradient." (Jacob, 4 Nov 2025). Nothing to verify; the numbers are spoken. [chart]
- Chart: The lock
- Open: [JACOB] how deep to go on the exploit physics (about 4.5 minutes at full length); "Fight the cabal" on the spoken track only or also on screen; confirm the six concessions are said first.



## Slide 8 — Why we must fight: for the first time in history displaced humans may not move up the ladder; the root of the economy may be intelligence; the unification of capital, governance and AI will accentuate totalitarianism. Labor share is dropping.

- We are in a period of revolution; very big questions about the nature and role of humanity when AI so fundamentally changes work. [context: timeliness]
- Massive capital and economic concentration is about to happen; AI companies set to soak up almost 100% of the valuable capital in the world. [context: timeliness]
- The major AI companies have very few engineers yet are becoming some of the most powerful and richest companies in the world; very different from how the economy worked in the past. [context: timeliness]
- AI is forming the bedrock of the economy; value capture is flowing to NVIDIA, Anthropic, OpenAI and their peers, right now. [context: timeliness]
- Central question: are humans about to become untethered from power in the economy and in the country? [context: economics]
- Before this era humans had control and say because humans represented work power (labor); AI threatens to remove that lever. [context: economics]
- Does the average human still have the economic and political weight they had in the past? Will that shift change the political relationship between the state and humans? [context: economics]
- Point 2: economies are moving to a system where almost all productivity can be done by a single person connected to AI agents; robotics will speed it up. [Jacob 14 Sep]
- Proof example: Opentensor dropped from 35 engineers to 4 and is faster now than in 2022. (also fits 15) [Jacob 14 Sep]
- Start the talk with the audience feeling there is no hope; the hopelessness: the world is moving to one where the weight of the human in the economy is diminishing. [Jacob 16 Sep]
- In the past, under a totalitarian regime, workers could take up arms or go on strike; that balance of power held governments back from extreme totalitarianism. [Jacob 16 Sep]
- AI and robotics fundamentally change the power dynamic between individuals and capitalists and purveyors of power. [Jacob 16 Sep]
- When the payroll dies there are two claims on the economy, wages or ownership; World One offers an allowance, World Two the second way in. (also fits 14) [Jacob 14 Sep visceral]
- Weber (as Jacob frames it): what keeps humans in a balance of power with governments is capital and labor. We are seeing one of the largest concentrations of wealth ever amassed, and for the first time the erosion of the power of the individual worker. Attribute correctly on stage: Weber gave the monopoly-on-violence definition; the checks come from Montesquieu, Marx, Tilly, Ferejohn and Rosenbluth. [dictation 21 Sep middle]
- "Hopelessness is what you should feel, because the potential for any of us to have a say, a right, an option, or an exit from this impending behemoth is [vanishing]." (also fits 9) [dictation 21 Sep middle]
- Objection 3: "The tractor made labor useless and humans found new work; the self-driving car is the tractor." [objection 3]
- Answer: agree with the history (farming 40% to 2% of US jobs; ATMs and tellers; the lump-of-labor fallacy), then: every prior machine replaced muscle and pushed humans up a rung toward thinking; AI is aimed at the top rung and there is no rung above. [objection 3 answer, 21 Sep]
- The horse: ~25 million working horses in the US around 1915, population down 90% in a generation after the engine; Leontief 1983 asked whether humans are the farmer or the horse [VERIFY numbers]. [objection 3 answer, 21 Sep]
- The argument needs leverage, not unemployment: Engels' pause (wages flat for decades while output per worker rose by half); labor share at a record low with high employment. [objection 3 answer, 21 Sep]
- "the question is not whether you will have a job, it is whether anyone with power needs you to do it." [objection 3 answer, 21 Sep]
- The self-driving car replaces the driver and the value goes to the fleet owner; a horse story, not a tractor story; AI is not a tool for the knowledge worker, it is the knowledge worker. [objection 3 answer, 21 Sep]
- "Every machine before this one replaced muscle, so the hands moved to the head. This one replaces the head. Tell me where the hands go." [objection 3 answer, 21 Sep]
- Close: "Maybe you're right... then we built a cheap insurance policy. If you're wrong, we built the only door. I would rather be embarrassed than expendable." (also fits 23) [objection 3 answer, 21 Sep]
- Dryden Brown (Praxis), "Singularity Politics", 12 Feb 2026: "As AI automates labor, $100T in annual wages and attendant influence will accrue to a small number of AI companies. These companies are increasingly intertwined with the government... The political effect will be an enormous concentration of wealth and power with the state." [VERIFY $100T] (also fits 10) [Praxis 12 Feb 2026]
- Weber, 1919: the state is the monopoly on legitimate violence (definition only). What held it in check: money that could leave and hands that could stop. [outline v2]
- Montesquieu, 1748: the bill of exchange put wealth beyond the prince's reach and "the great strokes of authority" became clumsy (Cohler, Miller, Stone translation). [outline v2]
- Adam Smith, 1776: "The proprietor of stock is properly a citizen of the world." [outline v2]
- Tilly, 1975: "War made the state, and the state made war"; every right was a receipt for something the ruler needed. [outline v2]
- Ferejohn and Rosenbluth, Forged Through Fire (2016): "If the modern democratic republic is a product of wars that required both manpower and money for success, it is time to take stock of what happens to democracy once the forces that brought it into being are no longer present." drafted alt: They wrote about soldiers. I am telling you it is about workers too. [outline v2]
- Jacob on tape, 2024: "we built systems that we could co-own, and that power was in check." [outline v2]
- drafted alt: "Every freedom was a receipt. The state gave it because it needed something only you had. AI is the first technology that means it does not need anything from you at all." [outline v2]
- US labor share 52.9 percent, the lowest since records began in 1947 [VERIFY, BLS preliminary]. [outline v2]
- One lab near a trillion dollars with a few thousand people, about half of one percent of GM's 1979 headcount [VERIFY]. A frontier company used to share what it made through a payroll; this one does not need one. [outline v2]
- Engels' pause as precedent (cut candidate): 1780–1840, output per worker up forty-six percent, wages up twelve; it ended because the factory needed the workers, and this factory does not. [outline v2]
- Jacob: "You can be a tyrant that loses 90% of your workforce potentially, and it doesn't even affect your economy." drafted alt: For all of history the emperor needed hands, and hands could hold a sword. [outline v2]
- The money: capital was a check because it was plural, mobile, and independent. Now few (six balance sheets and one chipmaker), immobile (export-controlled chips, gigawatt data centers, weights a letter can forbid), dependent (permits, licenses, the exam, the waiver; the state needs military AI, a tax base, the administering machine). Each is the other's moat. The East India Company as the rhyme, once (cut candidate). (also fits 9) [outline v2]
- drafted alt: "For three hundred years the money could leave and the hands could stop. Now the hands are not needed and the money cannot leave. The two things that kept the state honest are sitting in its lap." [outline v2]
- [JACOB: confirm both headcounts, 35 to 4, and name the entity] [outline v2]
- Violence needs money and it needs hands. Capital could leave: move the factory, the gold, the family. Labor could stop: strike, desert, take up arms. Every limit ever placed on a ruler came from one of these two threats. [trajectory §2]
- The franchise expanded when the state needed the masses; wages rose when the factory needed the workers. Rights were not granted. They were priced. [trajectory §2]
- Marx (1848), Sorel, Luxemburg: labor's check is the power to stop; after the Black Death wages doubled and the Statute of Labourers (1351) failed to hold them down. [trajectory §2]
- Tilly (1990): rulers bargained with capitalists for money (1688) and with populations for men (conscription, then the vote). Bates and Lien (1985), Boix (2003): rulers dependent on mobile capital grant more representation. Acemoglu and Robinson, The Narrow Corridor (2019): the Red Queen. [trajectory §2]
- Forged Through Fire: elite warfare (knights) gave feudalism; mass warfare (hoplites, Athenian rowers, musket infantry, levée en masse, conscription, World Wars) produced the franchise, women's suffrage within two years of WWI, the welfare state and GI Bill after WWII. Since 1945 capital-intensive war means the state no longer needs citizens' bodies. Safe paraphrase: "democracy is a bargain, not a gift." [TODO: pull exact forward-looking passage] [trajectory §2]
- Honest caveat: there are two more checks, rival states and ideas (church, press, university); AI weakens both (two licensed systems; a licensed mind captures ideas at the source). Capital and labor are the two that ordinary people held. [trajectory §2]
- Anthropic is worth about a trillion dollars with a headcount that is 0.6% of General Motors at its 1979 peak. [trajectory §3]
- drafted alt: When the ruler no longer needs the ruled, the strike costs him nothing and the rifle has no one to point at. [trajectory §3]
- drafted alt: The two checks were not defeated. They were absorbed. [trajectory §4]
- Slide version of "why capital cannot leave now": (1) capital became physical at a scale that cannot move; (2) the product needs permission to cross a border; (3) exit needs a destination and there are two, Washington and Beijing, both with a switch; (4) it does not want to leave; the state is now its moat. Old bargain: "treat us well or we leave"; new one: "protect us and we stay". [trajectory §4, 21 Sep]
- Slide text: "Money could leave." then "It is bolted to the ground. It needs a permit to ship. It asked for the rules." [trajectory §4, 21 Sep]
- drafted alt: "You cannot threaten to leave a room you asked to be locked in." [trajectory §4, 21 Sep]
- drafted alt: Intelligence is becoming the whole economy, and it is stopping needing hands. That is not a prediction about other people. My own company did it. [two worlds]
- drafted alt: I did that. Nobody made me. It was cheaper and it was better. Every company becomes small. One person, a lot of agents, and no staff. [two worlds]
- drafted alt: For three hundred years, wages were how a big machine shared what it made. It was also how ordinary people had weight. You were needed, so you were heard. [two worlds]
- The Cantillon effect for intelligence: the lab gets it first; its financiers second (the last two funding rounds of two labs were larger than the whole venture market of most years); the licensing government third (contracts, taxes, a phone number that answers); then pension funds; then, at the end of the pipe, you, as a subscription price. (also fits 9) [two worlds]
- drafted alt: It ended when the workers organised, and they could organise because the factory needed them. This time the factory does not need them. So this time it does not end. [two worlds]
- Jacob: "Without carefully constructed ownership and openness layers, humanity doesn't end up unequal -- it ends up enslaved." [tweet 7 Jan 2026]
- Jacob: "You will work for the machines that work for the elites — and you'll own none of it." [tweet 31 May 2025]
- Jacob: "We're not ready for a world where armies are made of machines, not men — where human power is so eroded that the AI cabal can control the world from a laptop." (also fits 9) [tweet 5 Aug 2025]
- Jacob: "Just that this time they're not diluting money: they're diluting us." [tweet 24 Jul 2023]
- Jacob: "we risk, maybe for the first time ever in history, a form of totalitarianism that we've never seen before." (also fits 9) [BTC 45:50]
- Say instead: "Tilly, 1975: 'War made the state, and the state made war.'" [fact-check]
- Say instead: "After the Black Death, wage rates rose by half or more within a generation. Parliament passed the Statute of Labourers in 1351 to freeze them. After the 1350s it could not." [fact-check]
- Say instead (Forged Through Fire, verbatim): "Two political scientists wrote in 2016: 'If the role of the masses in protecting the nation-state diminishes, will the cross-class coalition between political inclusiveness and property hold?' They wrote about soldiers. I am telling you it is about workers too." [fact-check]
- Say instead: "Montesquieu, in 1748: the bill of exchange put wealth beyond the prince's reach, and 'the great strokes of authority' became clumsy." [fact-check]
- Say instead: "My own foundation had thirty-five engineers in 2022. The core team today is four. We ship faster." [fact-check]
- US labor share 52.9% in Q2 2026, lowest since the series began in 1947 (BLS preliminary); 65.8% in Q1 1947, 62.8% in Q4 2000. [economics]
- Global labor income share fell 1.6 points from 2004 to 52.3% in 2024; workers received $2.4 trillion less than under a stable share (ILO, 4 Sep 2024). [economics]
- Automation explains 50–70% of US wage-structure change 1980–2016, with only 3.4% cumulative TFP gain (Acemoglu & Restrepo, Econometrica 2022). [economics]
- Acemoglu (NBER 32487, May 2024): AI will raise US TFP by at most 0.66% over ten years and "is predicted to widen the gap between capital and labor income." [economics]
- Trammell & Korinek (NBER 31815, rev. Apr 2026): under full automation "the burst in output... will almost all accrue to capital, driving the labor share toward zero." [economics]
- Korinek & Suh (NBER 32255, Mar 2024): in the aggressive AGI scenario, wages collapse after about 3 years while output grows more than tenfold. [economics]
- Brynjolfsson, Chandar & Chen (Stanford, Aug 2026): employment of 22–25-year-olds in AI-exposed jobs is 19% below its peers' path as of June 2026, up from 13% in Aug 2025. [economics]
- IMF (14 Jan 2024): 40% of global jobs and 60% of advanced-economy jobs are exposed to AI. [economics]
- Anthropic: $965B valuation, $65B raised, run-rate revenue $47B (28 May 2026); reportedly past $65B run-rate by end July 2026, from ~$9B at end of 2025 [verify against S-1]. 2,300 employees at end of 2025 vs NVIDIA's 42,000. [economics]
- OpenAI: $852B valuation, $122B round (31 Mar 2026); 2025 operating loss $20.9B on $13.07B revenue (secondary). [economics]
- NVIDIA FY2026: $215.9B revenue, $130.4B operating income, 42,000 employees; "visibility into more than $1 trillion in cumulative Blackwell and Rubin revenue" 2025–2027. [economics]
- GM employed 853,000 worldwide at its 1979 peak, 618,365 in the US. Walmart employs 2.1 million; about $324,000 revenue per employee. Facebook, Google and Microsoft matched Walmart's value with "1–2 orders of magnitude fewer workers" (AEA 2018). [economics]
- Four hyperscalers guide $720–745B of 2026 capex; big-four Q2 2026 capex of $170.1B equaled 99% of their operating cash flow. [economics]
- AI took 52.7% of global VC deal value in 2025 and 65.4% of US VC deal value (PitchBook); OpenAI's $40B round was over half of all global AI VC in Q1 2025. [economics]
- Magnificent Seven drove 62% of S&P 500 returns in 2023, 53% in 2024, 46% in 2025; peaked at ~35% of S&P 500 market cap in June 2026; Nvidia alone is 7.9%. [economics]
- Top 1% of US households own 49.9% of corporate equities; top 0.1% own 24.0%; bottom 50% own 2.5% of net worth (Fed DFA, Q2 2025). (also fits 14) [economics]
- Engels' pause: 1780–1840 British output per worker +46%, real wages +12%; profit rate doubled (Allen 2009). [economics]
- John Thelwall (1790s): "Every large workshop and manufactory is a sort of political society, which no act of parliament can silence." (quoted in Acemoglu & Johnson, Power and Progress, 2023) [economics]
- Acemoglu & Johnson (NBER 32416, May 2024): "Wages are unlikely to rise when workers cannot push for their share of productivity growth." [economics]
- Counter-evidence to have ready: Yale Budget Lab (Oct 2025): "The broader labor market has not experienced a discernible disruption since ChatGPT's release 33 months ago." Humlum & Vestergaard (NBER 33777): AI chatbots show "precise null effects on earnings and recorded hours... ruling out effects larger than 2%." Answer: the argument is leverage, not unemployment. (also fits 24) [economics]
- Figure spec: "US labor share, 1947 to 2026": one black line, two labelled dots (Q1 1947 65.8; Q2 2026 52.9, preliminary), annotation "Lowest since the series began in 1947." [VERIFY Q2 revision 27 Sep] [charts-plan]
- Chart: US labor share, 1947 to 2026
- Open: [JACOB] confirm Opentensor 35 → 4, the years, and which entity (Foundation vs core team); how much economics to cite in a 30-minute talk.



## Slide 9 — The state has no guardrails; absolute totalitarianism is possible. Deprecation at the hands of the AI + money + government trilogy, which no longer needs you.

- Point 1: things that have no power are expendable. If humans are not required, they will first become slaves, then be exterminated. Stage word choice: "deprecated" is colder and safer than "exterminated"; decide. [Jacob 14 Sep]
- Point 3: if humans are not required, they will be deprecated. [Jacob 14 Sep]
- Objection to prepare for: "That is what democracy is for. One person, one vote, and it is more visceral than a token." [objection 1]
- Answer: votes have force only when the voter is needed; suffrage followed need historically; democracy is jurisdictional and AI is not; regulatory capture is democracy's failure mode and is happening now; redistribution by vote is the allowance, not ownership; ownership is the precondition that makes votes count, not a replacement. Concede: token ownership is one-dollar-one-vote, but entry is by work, not capital, and the door is open to anyone. (also fits 23) [objection 1 answer, 14 Sep]
- The state merging with the largest AI companies is the merger of the two most powerful organisations of power in the world, the monopoly on violence and the monopoly on capital. [Jacob 16 Sep]
- No escape: paint a picture of no escape from the impending totalitarianism of the unholy union between government and AI corporations. [Jacob 16 Sep]
- China, already a somewhat totalitarian regime, is currently the only other option for open-source AI; the only alternative to one licensed system is another. [Jacob 16 Sep]
- World One's feeling is hopelessness, the absence of a door; make the room feel the wall, then open the door. Land World One on "You have one choice," open World Two with "Now you have two." (also fits 10) [Jacob 14 Sep two-worlds session]
- "Fiat AI" is Jacob's own phrase (VirtualBacon): intelligence by decree, debasable (silent nerfs), freezable (bans, geo-blocks, the June letter), with a Cantillon effect. [Jacob 14 Sep]
- A third country is a third switch (Cohere in Toronto, MBZUAI in Abu Dhabi, G42 weighing US ownership for chip access). (also fits 10) [Jacob 14 Sep]
- Stakes: money outside the state protected what you own; a mind outside the state protects how you think. (also fits 14) [Jacob 14 Sep]
- "AI getting out of the box is not what worries me. AI inside of a box, an economy controlled by one box, an international government controlled by one box, is the fear of Peter Thiel and Praxis and others considering how the hell do we escape." (also fits 10) [dictation 21 Sep middle]
- Without it: tomorrow we wake up in a world where AI and the capital that controls most of the economy are in the hands of a small group. [dictation 21 Sep middle]
- The dystopian picture: something like the movie Elysium; a small group controls the technological and financial levers of the world; most people live in a form of Gaza: no say, no control, no power in the economies of the future. (Gaza comparison: judge the room.) [context: real fear]
- Ordinary people become unimportant extremities of the economic and productive powerhouses that govern the world. [context: real fear]
- Thiel: a totalitarian one-world state is "a prison from which there is no escape... far worse than North Korea because you could still get out of North Korea." [Thiel, Nashville 14 Sep]
- Thiel: "If you outlaw it in the US, it'll be built in China... So you have to outlaw AI everywhere in the world." The exit argument from the other side of politics, one week after Pace the Frontier. [Thiel, Nashville 14 Sep]
- Praxis's three rational responses: join the government to run the machine; join a union to extract from it (UBI allocations, alignment priorities); exit to an aligned political system. "Join the government" and "join a union" map to World One (the allowance). (also fits 10) [Praxis 12 Feb 2026]
- drafted alt: If humans are not required they are not persecuted. They are deprecated. A programmer's word: the function still exists, nobody calls it, one day it is removed from the documentation. [outline v2]
- The offer: one proposed 2.5 percent of large-company equity into a fund every year, another a 3 percent tax on AI revenue; sincere; notice the route: one treasury, one country, a rate set in one room by the people whose equity it is. Jacob: "they're giving you basically peanuts for a UBI with an orb." drafted alt: An allowance can be reduced. A share cannot be reduced by the people who owe it. (also fits 14) [outline v2]
- drafted alt: Nobody comes for you. Your API key moves to the lower tier. The model your children learn from has values settled in a room you were never in. [outline v2]
- Jacob: "Everybody in the world will be sipping from the mental biases of a single company that controls artificial intelligence... that would be devastating for our ability to make sense of the world." [outline v2]
- Elysium, once: a station in orbit where the machines are owned, a surface below where they are rented. Nobody on the surface is oppressed by a robot. [outline v2]
- Jacob, 2022: "the bigger issue for AIs is not the AI fire the nukes and they take over the world. It's more that like a small group of people come to own it and use it for themselves." [outline v2]
- drafted alt: today the only other supply at the frontier comes from a country that is already the thing we are afraid of. Two systems. Both with an address. Both with a switch. [outline v2]
- Hirschman, 1970: voice only has force where exit is possible. drafted alt: A vote is a request. Democracy is voice, and voice was always backed by the two checks that are now gone. Requests from people who are not needed are surveys. [outline v2]
- A share is not exit: Microsoft had put thirteen billion dollars in and held a right to half the profits; when the board fired the chief executive in 2023 it got a phone call minutes before the post went up; nobody asked it [VERIFY wording; never "49%" or "from a tweet"]. (also fits 14) [outline v2]
- "You have one choice." Pause. (Land of the wall, ≈18:15.) [outline v2]
- Thiel [JACOB: keep or cut; if cut, say "a man who backs the other side of most of the politics in this room said in Nashville this month..."] [outline v2]
- Merger evidence in one line: export letters obeyed in an afternoon, a slowdown pact among incumbents endorsed within three hours, licensing exams the incumbents wrote, antitrust waivers requested. Neither can leave the other. Neither needs anyone else. [trajectory §4]
- drafted alt: The wall closes from both sides. You have one choice. [trajectory §5]
- drafted alt: A customer who can walk out gets listened to. A customer who cannot gets a survey. [trajectory §6]
- drafted alt: a share is a claim on the rent from a thing the state can direct. [trajectory §6]
- drafted alt: Now mind. Badges in the labs. An exam models must pass to be sold. A power to block a release. And on the twelfth of June, a letter at 5:21, and two frontier models off for every customer on earth for eighteen days. [mind doc]
- Nietzsche's last man, who "proceeds carefully" and blinks. [mind doc]
- Two worlds device: same dates, same people (Montreal, Lagos, Lima), different switch; walk both from this year to about 2035. One is being built in Washington and in the offices of about five companies; the other by people in this room. [two worlds]
- World One 2026 vignette: a developer in Montreal built her product on that model, lost it on a Friday, got it back eighteen days later. She had not broken a rule. She was downstream of a switch. [two worlds]
- drafted alt: Not to open up. To slow down, together, and to ask for permission to do it as a group. Who chooses the evaluators? The labs. Who pays them? The labs. [two worlds]
- drafted alt: I want to be fair: I think many of them mean it. The fear is real. Watch what the fear builds, not what it says. (also fits 5) [two worlds]
- World One 2027 vignette: a startup in Lagos that fine-tuned a Chinese open model for Yoruba-language customer service gets a notice from its cloud provider: show a certificate or we cannot serve you. There is no certificate for a two-person company. [two worlds]
- World One 2028: learning becomes theft; distillation. In September three US security agencies named six Chinese labs for it; one lab's lobbying filings list "Export Controls and Distillation" as a priority. The student in Lima who distils a Spanish clinic model gets a letter too. [two worlds]
- 2028 hardware side: registration of every training cluster above a threshold; a duty for anyone running a large model to keep a switch a department can pull; a bill for that switch already exists; eighty-six percent of American voters like it. [two worlds]
- World One 2029–2030: data-center moratoria (New York one-year freeze on anything over fifty megawatts); capacity moves to where the waivers are; by 2030 four suppliers of frontier intelligence in the West and one government that can reach all of them; a hospital in Montreal's licensed bill goes up thirty percent and there is nobody else to call. drafted alt: The lab passed the test, and the test is the wall. [two worlds]
- Every adult receives about thirteen thousand dollars under the 2.5 percent equity fund proposal. In Lagos there is no such fund. In Lima there is a promise of one. In Montreal there is a cheque. [two worlds]
- drafted alt: Every step is called safety and every step is a switch, and every switch has a hand on it, and the hands are all in one building. [two worlds]
- drafted alt: There is no round-up. They are oppressed by the people who hold the switch to the robots. [two worlds]
- Hinge fact: sixty-eight percent of American voters, including sixty-three percent of Republicans, support a pause on advanced AI and a ban on superintelligence. drafted alt: If it were a vote, World One would win tonight, and it would be called safety, and most of the people voting for it would be right about the fear and wrong about the cure. [two worlds]
- Hostile push 1: "The fear is sincere... it cannot be capture." Answer: I believe them, which is exactly why a risk that size should not be managed by five companies in one jurisdiction behind one switch; a switch that can stop a model can also steer it. (also fits 24) [two worlds App. C]
- Jacob: "UBI is what they will offer you in return for control, it's a bribe, a contract you will sign to solidify your agreement with the loss of power." [tweet 6 Jun 2025]
- Jacob: "It's controlled basically by one person. It's a closed source lab. You'll probably never work there. And for the rest of your life, you'll have a subscription." [BIG TALK 31:46]
- Jacob: "as cattle that suckles off the teat of the OpenAI credits." [VB ≈60:20]
- Jacob: "It's like fiat AI." [VB ≈58:00]
- Jacob: "this is becoming very much a fiat game." [POT26 10:45]
- Jacob: "If it's not that way, then basically, you know, it's not our mind." [CT26 ≈1:10]
- Jacob: "I do think people should be cognizant and think about the effects of outsourcing our mental capacity to machines that are controlled by singular entities." [CT26 ≈2:50]
- Jacob: "if we live in a world where there is just one hierarchy of intelligence, one intelligence, people will eventually find out." "If decentralized AI doesn't win, I think that's when people notice. I think that's when people understand that they don't have any hope of taking down the giants." (also fits 24) [CT26 ≈2:35, ≈2:15]
- Thiel: "There would be no outside left." Bittensor's identity line: you are the outside. (also fits 14) [Thiel, Hoover Uncommon Knowledge II, 8 Oct 2024]
- Thiel: "Once you get a totalitarian one-world government, you can't reverse it." The exit must exist before the door shuts; why the ossification date matters now. (also fits 22) [Thiel, Hoover I, 2024]
- Thiel: "If we say that it's all around the next generation of large language models, nothing else matters, then you've probably collapsed it to a small number of players. And that's a future that I find somewhat uncomfortably centralizing, probably." [Thiel, Conversations with Tyler, 2024]
- Say instead: "Eighteen days under a government order. She got it back on day nineteen." [fact-check]
- Say instead: "Microsoft had put thirteen billion dollars in and held a right to half the profits. It got a phone call minutes before the post went up. Nobody asked it." (also fits 14) [fact-check]
- Say instead: "Thiel, this month: a one-world state would be 'far worse than North Korea because you could still get out of North Korea.'" (could, not can) [fact-check]
- 18 days: 12–30 Jun 2026, a US export-control directive forced Anthropic to disable Fable 5 and Mythos 5 for all customers; the letter arrived "at 5:21pm (ET)." [lab closure]
- Demis Hassabis, 14 Jul, on his proposed Standards Body: "Frontier Models would be required to pass it to be deployed in the US market." [lab closure]
- Data for Progress, 9 Sep: 68% of US voters support a pause on advanced AI and a superintelligence ban; 63% of Republicans. [lab closure]
- 50 MW, one year: New York's statewide permit freeze on new hyperscale data centers (Hochul, 13 Jul 2026), the only such freeze in force. [lab closure]
- $12.93B: Nvidia's agreed price for Hugging Face (2 Sep; closes H1 2027). The open-weights home now has an owner with an address. (also fits 11) [lab closure]
- Precedent for defunding by letter: 23 banks received FDIC "pause letters" on crypto; released under FOIA Nov 2024–Feb 2025. (also fits 11) [lab closure]
- Korinek & Juelfs (Brookings/NBER 2022): avoiding mass misery "will require other ways of distributing income than labor markets, whether via sufficiently well-distributed capital ownership or via benefits." (also fits 14) [economics]
- Korinek, US Senate AI Insight Forum (1 Nov 2023): without reform, AGI "would likely deliver mass impoverishment." Proposes "universal basic capital." (also fits 14) [economics]
- Korinek & Lockwood (NBER 34873, 2026): transformative AI "may gradually erode the two main tax bases... labor income and human consumption." [economics]
- Dario Amodei, Axios, 28 May 2025: AI could cut half of entry-level white-collar jobs and push unemployment to 10–20% in 1–5 years; proposes a 3% token tax. [economics]
- Sam Altman, "Moore's Law for Everything" (Mar 2021): American Equity Fund, 2.5% of large-company market value per year, paid in shares, to every adult. [economics]
- Figure spec: "Every company becomes small": dot plot on one log axis of payroll headcounts, Opentensor 35→4 [JACOB], Anthropic ~5,000, OpenAI ~7,800, NVIDIA 42,000 (black) vs GM 1979 853,000, Walmart 2,100,000 (grey); annotation "One of these is worth about a trillion dollars." [charts-plan]
- Chart: Every company becomes small
- Open: [JACOB] "exterminated" vs "deprecated"; Gaza line (draft cuts it); Elysium once or not at all; Thiel keep or cut.



## Slide 10 — We have a solution. It is called Bitcoin. It is not an address, it is a network. The answer is not Praxis and it is not the US as Thiel thinks.

- Bittensor was highly inspired by the power of Bitcoin. [context: Bitcoin]
- Bitcoin showed that an economic network can bring computers and individuals together to contribute stranded resources toward a larger problem, and scale that system to unbelievable size. [context: Bitcoin]
- Claim to check before a slide: Bitcoin's raw compute out-competes some of the largest data centers; hashrate is ASIC hashing, not general-purpose FLOPs; find a defensible framing. (also fits Do not say) [context: Bitcoin]
- What Bitcoin proves: we can create systems governed by networks and incentives, not by a central authority. [context: Bitcoin]
- Bittensor is a cryptocurrency; that places it in the same revolutionary lineage as Bitcoin. [context: core thesis]
- Bitcoin's opponent is fiat money and the control structure around it; Bittensor's opponent is the same fiat world and its control structure. [context: core thesis]
- Bitcoin parallel in Jacob's words: before Bitcoin there was no way to hold or use money outside the control of the nation state, specifically the US dollar; Bitcoin gave humans another option, exit from the system. [Jacob 16 Sep]
- Bitcoin did not need to become the world reserve currency to succeed; it simply needed to exist, to give people leverage to use a different system and so disaggregate and disintermediate power. [Jacob 16 Sep]
- Bitcoin is the economic shield; Ethereum, DeFi and the whole crypto ecosystem exist only because Bitcoin exists underneath them. Pause to name the things that can only exist because of the shield. [Jacob 16 Sep]
- Bittensor must bring to mind what Bitcoin brought to money: hope of an exterior system that is not the same group and bureaucracy of power that controls access to intelligence, capital and money. [Jacob 16 Sep]
- The exit is not an island in the Caribbean. The exit has no address. The exit is the network: decentralization across many individuals, with as close to zero central command as possible, so there is nothing to attack or hit. (also fits 11) [Jacob 16 Sep]
- Bitcoin's moral case is not distribution: (1) nobody can print, (2) nobody can freeze, (3) anyone can leave. Rules over rulers. Exit over voice. Concentrated miners are fine because the protocol binds them and entry is open (contestable market). [Jacob 14 Sep corrected core]
- Bitcoin gave three things, none of them price: an existence proof (money without a sovereign ran on 3 Jan 2009; "the monopoly is broken"), a fixed rule (21M), and a story with a face (WikiLeaks 2010 blockade, Argentina 2001, Cyprus 2013). It made fiat optional, and the option disciplines the incumbent. [Jacob 14 Sep fiat AI]
- The crux is exit versus voice (Hirschman). Do not say "the only hope"; say Bittensor is the only exit with no one standing in the doorway (open weights, DeepSeek, Hugging Face are exits with doorkeepers). [Jacob 14 Sep two-worlds session]
- Bitcoin is exit for your money; Bittensor is exit for your livelihood. (also fits 14) [Jacob 14 Sep visceral]
- "Every frontier lab on earth has an address. We are the one that doesn't." [Jacob 14 Sep lab with no address]
- For Montreal: "We are the frontier for the other one hundred and ninety countries." [Jacob 14 Sep lab with no address]
- Bittensor is not a third country; it is no country. [Jacob 14 Sep]
- Line: "Every company that tried to be money outside the state was shut down. Satoshi built a protocol instead. We are doing the same thing for the mind." Verify e-gold (indicted 2007) and Liberty Reserve (seized 2013) dates. (also fits 11) [Jacob 14 Sep protocol not company]
- Satoshi's vision, as Jacob reads it: build something that exists outside the reach of a single government; tooling they cannot shut off. Bittensor's version: a system of AI outside the singular jurisdiction of a country. [dictation 21 Sep middle]
- Thiel adapted: agree on diagnosis (fear the humans with the switch; safety as the road to the wall; exit over voice; crypto decentralizes and AI centralizes; money is the chokepoint). Disagree on cure: he needs a restrainer (katechon: strong US, Christian frame, Palantir) and admits the restrainer can become the tyrant; Bittensor removes the need for one (Bitcoin has no katechon). [Thiel adapted 15 Sep]
- Separation line: "His answer is a stronger nation. Ours is no nation." [Thiel adapted 15 Sep]
- One-world state vs one global network: a state is one ruler with force, an address, no exit; a protocol is one rulebook with no force, no address, exit by design. [Thiel adapted 15 Sep]
- Adapted lines drafted: North Korea, no outside left, one keystroke, "everywhere is a lot of places", the door before it closes, share of the lab, Nashville or Nairobi, company-shaped exit becomes a bank, every guardian strong enough to protect you can own you, one model is one mind. [Thiel adapted 15 Sep]
- Thiel's own challenge to meet: "if you strip it from the biblical context, you will never find it scary enough." [Thiel adapted 15 Sep]
- Praxis's synthesis is Digital Nations with territory; his "exit" still has an address and a flag. Bittensor is the fourth response he does not list: exit without a country. Take the three-options structure, not the brand. [Praxis 12 Feb 2026]
- Bittensor builds in a different world: not tied to the United States or its government, not tied to China; a third path for individuals, something truly powerful that can out-compete nation states and stay disjoint from them. [context: unique position]
- It exists outside the propagandistic and power structures that define the United States and China. [context: unique position]
- A centralized company is built in a centralized setting and is not distinct from the power struggle between China and the United States; Bittensor lives in a much more international, bottom-up, grassroots ecosystem. (also fits 11) [context: fair launch]
- Jacob: "I believe that was the vision of Satoshi. Build something that can exist outside the reach of a single government. Tooling they cannot shut off." [outline v2]
- Before 2009 there was no exit from state money: gold confiscated by executive order in 1933, banks frozen, WikiLeaks cut off from its payment processors inside one week in 2010. On 3 January 2009 money without a sovereign ran. drafted alt: Fiat became a choice, and the choice disciplined the incumbent. [outline v2]
- Jacob: "Three pools mine more than half of Bitcoin's blocks, and none of them can print a coin." [VERIFY weekly; never "five pools"] (also fits 11) [outline v2]
- Jacob: "Tao is also nothing without BTC." Bitcoin is the shield; Ethereum, DeFi, stablecoins exist because it exists underneath them. Everything on the ledger stands under Bittensor the same way, and Bittensor stands under Bitcoin. [outline v2]
- Jacob: "What Bitcoin taught us is that the exit is not a place. Not an island in the Caribbean. The exit has no address. The exit is the network." drafted alt: Thiel's exits were places and closed. Praxis's exit has a flag. A place has an address; a letter can arrive. [outline v2]
- Fiat AI lineage: print licensed until 1695; code export-controlled until a court called the rules "a prepublication licensing scheme" in 1999; money had one issuer until 2009; intelligence is licensed now: badges, an exam, a letter at 5:21. drafted alt: Nobody calls it a licence. It is a licence. (also fits 12) [outline v2]
- Small is enough: Bitcoin at about one percent of the world's broad money changed what every central bank can do [VERIFY]. drafted alt: The second supply caps the price and caps the censorship of the first. Not always better. Present. (also fits 12) [outline v2]
- drafted alt: Bitcoin did not offer an island. It offered a network: many individuals, no centre, nothing to hit. [trajectory §7]
- drafted alt: The shield does not need to be the biggest thing. It needs to be unbreakable. Under it, thousands built without asking. [trajectory §8]
- Jacob: "Bitcoin broke the gov's monopoly on currency. / Bittensor stops the gov's monopolization of AI." (444 likes) (also fits 14) [tweet 30 Oct 2023]
- Jacob: "This is why Bittensor has no country, or tribe." [tweet 13 Mar 2025]
- Jacob: "It wont be a country, it will be a protocol." [tweet 10 May 2025]
- Jacob: "Truly decentralized systems are supranational and don't adhere to standard geopolitics — indeed, they render them obsolete." [tweet 7 Feb 2024]
- Jacob: "Bitcoin is international law." [tweet 4 Jan 2026]
- Jacob: "There is no way to compete with fiat-money tank engines in the center lane." [tweet 31 Dec 2025]
- Jacob: "It was essential that we started with Bitcoin. Hard values, clear vision. However, subnets will be our Ethereum moment." [tweet 4 Sep 2023]
- Jacob: "It's REALLY important to remember that Tao is also nothing without BTC. The whole industry in-fact is dead without BTC. They solve different problems and i hope will help each other. I'll die on that hill." (1,019 likes) [tweet 23 Mar 2026]
- Jacob: "There's only one other monetary system in the world that can compete with US backed fiat AI" (341 likes) [tweet 22 Jan 2025]
- Jacob: "I'll be visiting Asia to discuss refocusing the AI race through non-permissioned borderless competitive systems; I believe it's the way to avert an AI hot war." [tweet 7 Feb 2024]
- Jacob: "We play a neutral game which cracks social issues we've been facing as societies for decades. / The left wing is concerned with racial politics. / The right is concerned with national goals. / Their answers are united in Web3." [tweet 13 Mar 2025]
- Jacob: "Bitcoin is the largest supercomputer in the world. Let me repeat that." "It's the same electrical draw as Thailand... Just a network." (frame as electrical draw, not FLOPs) [BIG TALK 11:12, 11:50]
- Jacob: "I worked at Google and I remember when they finished building a data center, they would just start building another data center beside it." [TDS 8:10]
- Jacob: "Bitcoin doesn't care that you guys went to the best university in the world. Sorry. Bitcoin doesn't care." (also fits 13) [BIG TALK 15:38]
- Jacob: "How do we connect the most powerful computer in the world to the most important computational problem in the world?" (also fits 11) [TWIST 13:40]
- Jacob: "There's a pattern. All these things are one thing." [BIG TALK 9:18]
- Jacob: "Are we in 1983, where it's going from mainframes to the internet?" [TDS 34:30]
- Jacob: "we are truly extra-national, right? We're not tied to one nation-state." [PBW ≈11:00]
- Jacob: "the third path for what artificial intelligence can be born out of." [TWIST 79:50]
- Jacob, Paris 2025: "Bitcoin is 450 times bigger than Google's cloud, IBM cloud, Cerebras cloud, AWS, and Nvidia. 450 times when measured in terms of hashes than all of those data centers combined." Keep the "measured in hashes" caveat; do not lead with it. [POT25 ≈3:12]
- Jacob, Paris 2025: "If we can get the power of digital currency markets to focus on artificial intelligence, we can be a thousandfold larger than any centralized organization in the world. And that's a big deal and that's really scary." [POT25 ≈3:44]
- Jacob: "Bitcoin and artificial intelligence, the fields are effectively tools of organizing matter." "Cryptocurrency is the most powerful alignment tool for humans, for compute." (also fits 11) [CT26 ≈0:20, 0:15]
- Thiel: "you have to outlaw AI everywhere in the world." The licensed supply needs every jurisdiction to comply; a network in no jurisdiction breaks the "everywhere". [Thiel, 2026 clip]
- Thiel: "Crypto is decentralizing, AI is centralizing." Use the first sentence only (the sentence continues "crypto is libertarian and AI is communist"). [Thiel, Stanford/Hoover, 31 Jan 2018]
- Thiel: "In the late 1990s, the founding vision of PayPal centered on the creation of a new world currency, free from all government control and dilution — the end of monetary sovereignty, as it were." The company-form exit failed and became a bank; the protocol form worked. (also fits 11) [Thiel, Cato Unbound, 13 Apr 2009]
- Say instead: "A ledger is light. Bitcoin's whole history fits on one drive, under a terabyte." (769 GB) (also fits 11) [fact-check]
- Figure spec: Figure 5, "The exit is not a place": layered diagram, a wide black band Bitcoin 3 Jan 2009 → today; thinner grey bands above starting at each launch year: stablecoins 2014, Ethereum 2015, DeFi 2020, Bittensor 2021 (mainnet 3 Jan 2021). Annotation: "The shield does not need to be the biggest thing. It needs to be unbreakable." [VERIFY launch years once] [chart]
- Chart: The exit is not a place
- Open: [JACOB] name Praxis and Thiel on stage or describe them; whether to say "fascism" at all (define on the spot if so).



## Slide 11 — No voting, no open source, no company, no island. It needs to be decentralized and protected in the sinew of its incentive equilibrium.

- Bittensor builds a decentralized, peer-to-peer, fair, transparent, anonymous, permissionless, global, international connected web network. [context: core thesis]
- Neuromorphic inspiration: memristor chips (Knowm, Alex Nugent; confirm) that could learn and feel the flow of energy; Jacob's idea: monetary computation, incentives flowing like energy, to tackle the hardest problem, the production of AI. [context: neuromorphic]
- Two claims: a different substrate can host a transparent, permissionless, open network for creating AI; and that network can be highly performant and compete with the best labs, because Bitcoin showed it was possible. (also fits 14) [context: neuromorphic]
- Drop the claim that Bittensor grants more power to more humans; not everyone will hold TAO, and even if they did it would not follow. Bitcoin mining is the right analogy: everyone can mine, yet a small number of incentive-aligned groups compete. (also fits Do not say) [Jacob 14 Sep corrected core]
- Bittensor's moral case, stated the same way: nobody can print intelligence for themselves (emissions go to producers under rules the same for all); nobody can freeze it (no license gate, kill switch, or jurisdiction); anyone can leave. Producers will be concentrated; that is acceptable. [Jacob 14 Sep corrected core]
- What the exit cannot be: not classical open source (no money; sits inside a government regime, can be found and ordered); not a standard company (address, charter, bank account); not just a network of people with a distribution mechanism (grant scheme or DAO). Three eliminations, then the specification: efficient, anonymous, incentive-driven at Bitcoin scale. [Jacob 16 Sep]
- An exit must be able to compete: it needs the engineers and talent, the capital (money and compute), the raw materials, and the intelligence. Maps to the five layers and to protocol emissions as the capital that cannot be withdrawn. (also fits 13) [Jacob 16 Sep]
- Why a protocol and not a company: a company is a creature of the state (address, board, bank account, subpoenable weights, payroll that stops when frozen). e-gold and Liberty Reserve were company-form money outside the state and were shut down; Bitcoin survived because there was no one to arrest. [Jacob 14 Sep]
- Five advantages of Bittensor: no address; a payroll without investors; a labor market with no HR (the undercover-in-Discord story); cannot be bought or defunded (no equity, emissions are not a budget line); outlives the founder ("1000 year Intelligence Federation"). [Jacob 14 Sep]
- Cost, said first: a company ships a better model sooner, can plan, keep secrets, run one coherent job; the Opentensor Foundation is a company and the key makes the network more company-like than it should be; the foundation is scaffolding. (also fits 21) [Jacob 14 Sep]
- Why harder than a ledger: open source without economics gets captured (Android, Chrome, Linux on corporate payroll); Bittensor is open source with a payroll. [Jacob 14 Sep mind outside the state]
- Attack and answer: "just open source with a token" — a gift can stop, a payroll cannot. [Jacob 14 Sep]
- Attack and answer: "intelligence for criminals" — said of press, radio, PGP, Bitcoin. (also fits 23) [Jacob 14 Sep]
- Decentralization is the means, not the goal; the door is the goal. Definition: "Decentralized means nobody can be sent the letter" (how many must comply for it to stop). [Jacob 15 Sep]
- Three layers graded honestly: production (decentralized, permissionless, concentrated like Bitcoin hashrate, acceptable), funding (decentralized by construction, protocol emissions, the layer no one else has), control (centralized, one key, the only layer that matters for the moral claim, has a date). (also fits 21) [Jacob 15 Sep]
- "Decentralized where the work happens and where the money comes from; centralized where the rules live; the third has a deadline." (also fits 21) [Jacob 15 Sep]
- "Bitcoin has five mining pools and no one can print a coin" [Jacob 15 Sep]
- Reconciles the April ("most decentralized AI protocol") and June ("not decentralized like Bitcoin") tweets: production vs control. Do not say "fully decentralized"; use the word sparingly. [Jacob 15 Sep]
- The four verbs, make/host/serve/use AI under no single jurisdiction, map to the five subnet layers. (also fits 13) [Jacob 14 Sep two-worlds session]
- Mission: unify the organisational principles inside computing to solve the world's most important technical challenge, the creation of AI. [dictation 21 Sep]
- Reason 1: this is how to build superstructures and tooling that exceed a centralized company, via the analogy of monetary computing and thermodynamic RAM, money as the ultimate computing platform; the examples prove it. Reason 2: Bittensor sits at the intersection of the grand debate about humanity itself. [dictation 21 Sep]
- Jacob, 13 Sep 2026, quoting Elon Musk's "Nothing can shut down open source": "No. But they can defund it." Open source cannot be banned, but it can be starved; open weights need a funding mechanism that no state or corporation controls. Candidate for a slide or the opening line. [Jacob tweet 13 Sep]
- "why should we fund it and not others?" Answer: every other funder of open AI (Meta, DeepSeek, Mistral, Hugging Face, governments) has a single decision-maker who can be lobbied or ordered to stop; Bittensor's emissions are protocol-issued on a fixed schedule to whoever produces; there is no one to persuade. Caveats first: emissions (~$300M/yr) are small against lab capex; the sudo key is a single point of pressure until it expires. [Jacob 13 Sep follow-up]
- Stage line: "Everyone else funds open AI until someone tells them to stop. We built the one that nobody can tell." [Jacob 13 Sep follow-up]
- Beff Jezos (Guillaume Verdon), 12 Sep 2026: "If you are reading this, the only possible resistance is to build decentralized, permissionless, and sovereign AI. Open source, self-hosting tools, RL post-training, local inference, open harnesses, we must build it all. We must fight overcentralization with an open stack." [Beff Jezos 12 Sep]
- Preferred terms: open ownership, open source. [context: unique position]
- Many talking heads discuss AI; none bring digital currency into the AI conversation of the 21st century; that is Bittensor's unique angle and makes it the leader in decentralized AI and digital currency. (also fits 14) [context: unique position]
- Jacob: "It cannot be classical open source": no money, and it sits inside a board and a jurisdiction; two weeks ago the most followed man on earth wrote "Nothing can shut down open source" and I wrote back "No. But they can defund it." [outline v2]
- Jacob: "It cannot be a standard company"; say the cost first (a company ships a better model sooner). Jacob: "It can't just be a network of people with a distribution mechanism": a DAO is a committee with a token, and a committee can be sent the letter. [outline v2]
- drafted alt: Open source is a gift. A company is a target. A DAO is a committee. We needed a market. [outline v2]
- Jacob: "Exit needs the ability to compete." Four inputs: talent, capital (money and compute), raw materials (data), and intelligence to judge the work. drafted alt: "Bitcoin competed on one input. A mind needs four. That is why we pay." (also fits 13) [outline v2]
- Jacob (white paper): "We use a digital currency reward because we believe that we cannot rely on altruistic computational resources alone." [outline v2]
- Jacob: "imagine Bitcoin mining, but what you're mining is a machine. You're mining for intelligence directly." [outline v2]
- Jacob: "You have an infinitely large company, and you only pay the person that can lower the loss." "This is an AI lab without any engineers." (also fits 16) [outline v2]
- drafted alt: Not the fourth country. No country. [outline v2]
- drafted alt: An exit that cannot compete is a museum. Bitcoin competed on one input, electricity. A mind needs four. That is why this is harder than Bitcoin and why it has to pay. [trajectory §11]
- drafted alt: The payroll is how an exit with no address buys what it needs without ever having a bank account to freeze. [trajectory §11]
- drafted alt: every company that tried to be money outside the state was e-gold. [trajectory §10]
- Anonymous miners out-built the founder in his own Discord, and the network paid them the same day. (also fits 16) [trajectory §12]
- drafted alt: A ledger is light. A mind is heavy: a hundred billion parameters, thousands of GPUs, months of power. It needs a payroll. That is why open source alone got captured. [mind doc]
- Over ninety percent of the Linux kernel is written on a corporate payroll. Android and Chromium are open, and one company writes them. drafted alt: A gift depends on the giver, and every giver has an address. [mind doc]
- Hostile push 4: "Concentrated producers are just a new oligarchy." Answer: Bitcoin's miners are concentrated and still cannot print a coin or freeze a wallet; rules over rulers, entry open, the threat of entry does the rest. (also fits 24) [two worlds App. C]
- drafted alt: A dozen big miners do most of the work on most subnets, the same way a dozen pools mine most of Bitcoin. That is fine. Entry is open. If a big miner gets lazy, a kid in Lagos takes its slot. [two worlds]
- Hostile push 2: "Open weights already exist. DeepSeek is under an open licence. You do not need a token." Answer: "No. But they can defund it." Every other funder has a desk you can lobby or order; this network pays about 3,600 TAO a day on a schedule nobody can call. (also fits 24) [two worlds App. C]
- Jacob: "AI can judge anything. Money can move anything." [tweet 24 Apr 2026]
- Jacob: "We're not creating a marketplace of AI. We're using markets to create AI." [tweet 21 Dec 2024]
- Jacob: "and the difference came from allocation not compute." (436 likes) [tweet 28 May 2024]
- Jacob: "A centralized firm can buy more hardware, but it cannot reproduce that level of competitive pressure." [tweet 5 Jan 2026]
- Jacob: "The only thing larger than a closed cluster is a network joined together." [tweet 24 Aug 2026]
- Jacob: "No matter how cracked your team is, you'll never beat the machine that picks cracked teams." (302 likes) [tweet 13 Feb 2025]
- Jacob: "When you mine Tao and get outcompeted, it just means smarter people picked up your pick axe and started digging for you." [tweet 1 Apr 2025]
- Jacob: "To create intelligence by programming currency." (292 likes) [tweet 19 Jul 2025]
- Jacob: "Software is eating the world / AI is eating software / Incentives will devour AI" [tweet 1 Mar 2025]
- Jacob: "Kill the center, become the swarm." (342 likes) (also fits 21) [tweet 14 Mar 2025]
- Jacob: "You're building an AI company. We're building a 1000 year Intelligence Federation. We are not the same." (745 likes) [tweet 28 Jun 2024]
- Jacob: "We'll never compete with OpenAI as a company - that model is perfected in Silicon Valley. / Our only hope is to revolutionize the way we organize ourselves to solve problems with open protocols. / But … protocols are difficult, they add a level of complexity unseen in a normal corporate structure: adversaries, black hats, Sybil attacks" (432 likes) [tweet 21 Dec 2024]
- Jacob: "Uber is an incentivized protocol / Airbnb is an incentivized protocol / Bitcoin mining is an incentivized protocol / Their global, unpermissioned, incentivized, nature" is why they took the world by storm. (448 likes) [tweet 3 Apr 2024]
- Jacob: "Whats the real moat? The moat is the decentralized digital commodity market and its ability to abstract a super computer that no amount of money can buy, and no single company can own — by definition." [tweet 24 Sep 2023]
- Jacob: "Intelligence is the ultimate commodity. The densest information." (also fits 13) [tweet 9 May 2023]
- Jacob: "Usually a tweet like this would come with a "we're hiring" plug. / but Bittensor doesn't have an HR department. / You don't need to ask anyone to mine TAO." (also fits 20) [tweet 25 Sep 2024]
- Jacob: "Decentralized training is through incentive alignment: everything else is a meme coin." (also fits 15) [tweet 9 Apr 2025]
- Jacob: "a permissionless market is purely blind." [TWIST 9:30]
- Jacob: "Because we don't want to produce hashes. We want to produce anything. Hashes in Bitcoin are useless." (also fits 12) [BIG TALK 16:42]
- Jacob: "Miners do any work, validators check any work, and the network emits a digital currency to those miners. We called that thing Bittensor." [BIG TALK 17:22]
- Jacob: "Think of it like a contract but it mines a particular commodity." [TWIST 20:30]
- Jacob: "We did not make a machine learning agent. We simply defined the incentives under which those AI agents could be evolved." "Who did it? We don't know. Some random person in the world wrote a 7,000 line AI agent." (also fits 16) [BIG TALK 21:29, 21:50]
- Jacob: "This is Bitcoin mining to training machine learning models." [BIG TALK 25:06]
- Jacob: "you hired the best engineers in the world and you don't even know their name." [POT26 11:55]
- Jacob: "Stranded talent, stranded compute, stranded innovation." [POT26 17:25]
- Jacob: "a Bittensor subnet is like a laser cannon, right? If it's not aimed perfectly, you're not shooting at the Death Star." (also fits 7) [POT26 6:15]
- Jacob: "you buy your house. You've won life. But have you actually achieved the aim? Can you actually compete with OpenAI? No." [POT26 11:10]
- Jacob: "we're moving at the speed of compute now. Like, intelligence has unlocked what company can be." [POT26 26:50]
- Jacob: "we're building an AI company, but we're doing it with the guts out." [BTC 18:20]
- Jacob: "it doesn't matter if it's decentralized and it's permissionless and it's transparent if it doesn't do anything." (also fits 16) [BTC 48:40]
- Jacob: "They're our friends. They're our customers. They can mine us. We can mine them. The moat is the mechanism." (also fits 15) [POT26 33:30]
- Jacob, Paris 2025: "The significance of building a market where the market pushes participants to lower the loss. In order to do that, you need to make the actual convergence of a machine learning model a market equilibrium." [POT25 ≈6:11]
- Jacob, Paris 2025: "You're training with individuals that are maybe using GPUs next to the Three Gorges Dam, and maybe their entire goal, in fact we know that their entire goal, is to destroy your training run. We know that they don't give a [expletive] about the loss going down, but we compel them to do that." Cleaned: "The people training it may want it to fail. The market pays them only if the loss goes down." (also fits 7) [POT25 ≈6:11–6:53]
- Jacob, Paris 2025: "Every time we coalesce these commodities into being better, more well-directed, they get more powerful and we get closer to unlocking the supercomputer power of Bitcoin." (also fits 13) [POT25 ≈9:19]
- Jacob: "It needs to be transparent. It needs to be auditable. It needs to be co-owned." "Decentralization is kind of a catch-all term for a lot of different things. Let's say like technologies that have moral primitives." drafted alt: "I have said before that decentralization is a catch-all. What I mean is three properties." [CT26 ≈0:50]
- Jacob: "There's some disadvantages. It slows things down for us, but we know where we're going, and we think that those primitives will really matter." (the cost, said by him) [CT26 ≈1:25]
- Jacob, 2024: "There is not just complete top-down ownership. And that's what we built when we came to America, and over the last thousand years in Europe we built systems that we could co-own, and... that power was in check." (also fits 8) [CoinFoin ≈0:40]
- Jacob, 2024: "We make AI, because it is a computational problem, using compute that is incentivized into an existence rather than purchased." [CoinFoin ≈1:40]
- Jacob, 2024: "If we can directly measure the things that we want accurately, we can get them cheaper and more efficient than if we just tried to buy them outright, not knowing what we needed." (why validators beat procurement) [CoinFoin ≈1:50]
- Jacob: "I think that we organize talent better than them. I actually think that we can find people that are true killers that can really solve problems faster than they could do with an interview process." (also fits 20) [CT26 ≈4:10]
- Jacob, 2024: "There are no boundaries. It's permissionless. There's no HR department. You don't need a resume." [CoinFoin ≈0:55]
- Say instead: "If someone quotes my April essay against my June essay: both are true. April was about production. June was about control." (essays, not tweets) (also fits 21) [fact-check]
- On screen: None planned; text only. Rejected as a figure: a Bitcoin mining-pool chart (would read as a warning about Bitcoin). [chart]
- Chart: none
- Open: [JACOB] how to word "no company": the Opentensor Foundation is a company and is scaffolding; say the cost first?



## Slide 12 — Bitcoin did one thing.

- Bitcoin at 1% of money disciplines central banks; the second supply need not be bigger, only present and good enough. (also fits 14) [Jacob 14 Sep corrected core]
- Bitcoin made fiat optional, and the option disciplines the incumbent. [Jacob 14 Sep fiat AI]
- Lineage of licensed means of thought: print (Licensing Act lapsed 1695, then the Enlightenment; Milton 1644), broadcast, code (PGP, 1999 code-is-speech ruling), money (2009), now mind. (also fits 14) [Jacob 14 Sep]
- Bittensor never had a pre-mine; no tokens were distributed to insiders; everyone had the opportunity to mine from the very beginning; no one was given anything. This fair launch is why the community is strong. (also fits 14) [context: fair launch]
- Correction: do not say "no venture round bought a slice before the public could". About 600k TAO were sold OTC to Firstmark, DCG and Polychain at roughly $18 in 2021–2023; those tokens were fairly mined in the open network after launch, then sold; nobody was allocated tokens before launch. (also fits Do not say) [Jacob 12 Sep]
- Anyone in the world can mine TAO; you can use your agents to mine the various subnets; no one can stop you from building a subnet or running a validator. (also fits 13) [context: fair launch]
- Twenty-one million. No pre-mine; nobody was handed a token before block one. [JACOB: OTC figures, said here before anyone else does: roughly 600,000 TAO, mined after launch, sold to three funds at about $18] [outline v2]
- Fiat AI as the licensed press of the 21st century: a lab decides what exists, who may use it, what it costs, what it says, when it goes dark. Every prior means of thought was licensed too: print until 1695, broadcast from birth, code until a court called it speech in 1999. drafted alt: Each time the exit came first and the permission came after, or never. (also fits 9) [trajectory §9]
- drafted alt: Bittensor does not need to win. It needs to exist and be good enough, so that the licensed supply becomes optional and its price, its censorship, and its capture are capped by the door next to it. (also fits 14) [trajectory §14]
- drafted alt: Every new way to think together, the state licensed first. Not banned. Licensed. [mind doc]
- Print: England, 1662, nothing printed without a licenser and a guild; Milton argued against it in 1644; in 1695 the Commons refused to renew the act, and pre-publication censorship ended, not for Milton's reasons but for a monopoly and bad, expensive books. [mind doc]
- Radio never got its 1695: since 1927 you speak to millions only under a licence. [mind doc]
- Code: in 1991 a programmer posted an encryption program to the internet; it was a munition. In 1999 a federal appeals court called the export rules "a prepublication licensing scheme." The words of 1662. [mind doc]
- Money: in 2009 a ledger started with no licence to issue, none to send, and no office to close. [mind doc]
- Bitcoin was not captured, because it pays. drafted alt: The block reward is a payroll nobody can cut. Bittensor is that payroll pointed at intelligence. (also fits 14) [mind doc]
- drafted alt: This is 2012. Bitcoin in 2012 had not replaced anything. (also fits 19) [mind doc]
- drafted alt: Bitcoin is about one percent of the world's money. It has not replaced the dollar. It has not made everyone rich. And yet every central bank on earth now has to answer a question it never had to answer before: what happens if people leave? [two worlds]
- Jacob: "21 M was never a meme." [tweet 31 Dec 2024]
- Jacob: "They're also the most leveling wealth distribution technologies we've ever created." [tweet 10 Apr 2025]
- Jacob: "One of the core design choices we went for with TAO (and now with dTAO) was organic growth over a drawn out inflation curve. / If all the tokens get minted up-front the network has no chance to mature and it privileges sniping, VCs and maximizes value concentration over creation." (308 likes) [tweet 31 Dec 2024]
- Jacob: "the drawn out supply curves of Bitcoin and Bittensor (as well as others) mean the network has years to develop strength of character, both of its holders and technology. / Anti-fragility can only play out this way." [tweet 31 Dec 2024]
- Jacob: "0 initial supply, no privileged allocation, slow consistent inflation over multiple years. / These are not arbitrary design choices." [tweet 19 Jan 2025]
- Jacob: "They told us that we should do a pre-mine. That was more legal." [UNSH 25:45]
- Bittensor: 21 million TAO cap, no pre-mine, first halving December 2025, ~129 subnets (Grayscale Research 2025; project context). [economics]
- On screen: None planned; text only ("Bitcoin did one thing" / "Bitcoin competed on one input: electricity"). [chart]
- Chart: none
- Open: [JACOB] OTC sale figures said here before anyone else does (about 600,000 TAO, mined after launch, sold to three funds at about $18): confirm numbers or cut the sentence.



## Slide 13 — We need to do all of these: compute, inference, storage, data, intelligence.

- Uncensorable alternatives exist today, layer by layer, none requiring identity checks: Lium non-KYC GPUs; Engy and Chutes non-KYC inference; Hippius non-KYC storage. [VERIFY each as of Sep 2026] (also fits 15) [Jacob 16 Sep]
- Stage wording: prefer "no account, no ID, no permission" over "non-KYC" for a general room; keep "non-KYC" for the crypto half. [context 16 Sep]
- The point: the exit is not a plan; each layer has a door open tonight. [context 16 Sep]
- Five layers: compute, inference, data, knowledge, free intelligence; plus the funding layer, protocol emissions. [context 16 Sep]
- Three verbs by how many will do each: Use (intelligence your work depends on cannot be switched off on a Friday), Earn (point a GPU, agent, model or data at a subnet and be paid in ownership; no résumé, visa, interview or permission), Own (hold TAO, a fixed slice of every unit of intelligence the machine will produce, and point it by staking). Most people will only Use; most Bitcoin holders never ran a node. (also fits 14) [Jacob 14 Sep visceral]
- Jacob, 13 Sep 2026: "we will need compute (@lium_io, @TargonCompute) / we will need inference (@engyai, @chutes_ai) / we will need data (@Hippius_cloud) / we will need knowledge (@teutonic) / If we are going to have free intelligence (@affine_io)". The open stack as five layers, each with a live subnet; use this layering instead of a list. [Jacob tweet 13 Sep]
- Layer map: Compute: Lium, Targon. Inference: Engy, Chutes. Data: Hippius. Knowledge: Teutonic. Free intelligence: Affine. No account, no ID, no permission. [trajectory §13]
- World Two 2026: same developer in Montreal has a second supplier in her config file; Chutes serves open models inside sealed hardware, about three million requests a day, published a year of its traffic with Harvard; she switches the endpoint; her product is back before dinner. Nobody at Chutes approved her. Nobody could have stopped her. (also fits 15) [two worlds]
- Lium: fourteen hundred GPUs in sixty-eight data centers in twenty-one countries, rented by the second, no identity check; last month billed just under a million dollars to eleven hundred renters, and sixty-three percent of those renters were not people. They were agents. (also fits 15) [two worlds]
- Engy: the largest open model on earth, two point eight trillion parameters, served on eighty gaming cards, at half the price the maker charges, with a cryptographic receipt on every answer. drafted alt: None of this is 2035. It is running tonight. It is small. It is the seed. (also fits 15) [two worlds]
- Targon: sealed chips co-designed with Intel so neither the owner of the machine nor the operator can see your job. Hippius: storage that speaks the same language as the big cloud and charges nothing to take your data out. Teutonic: a 110B model being trained right now by strangers, paid per improvement, no lab, no fixed recipe. Affine: models that belong to nobody and are published as they win. (also fits 15) [two worlds]
- Affine mechanism: anyone can post-train an open model, fight the current champion on coding and math and tool use, and if she wins, the network pays her about fifty thousand dollars a week for the slot and publishes the weights. drafted alt: Who wins? We don't know. That is the point. (also fits 17) [two worlds]
- Newsroom in Lagos: a licensed model declines to summarise a leaked document about a minister, for safety; the reporter pastes it into an open model served from nowhere in particular. drafted alt: Refusal now has a competitor. [two worlds]
- drafted alt: No HR. No visa. No exam. [two worlds]
- Jacob: "Dont ask permission. No HR department. Just bring compute and your mind." (also fits 20) [tweet 22 Nov 2023]
- Jacob: "We don't measure skin color or nationality. Incentive mechanism just measure output." [tweet 13 Mar 2025]
- Jacob, Paris 2025: "Is it just compute plus data? Well, we are figuring that out and we think it is actually. We think it actually in many cases is just compute plus data, but also inference." "We're stitching together commodities." [POT25 ≈14:12, ≈19:07]
- Jacob, Paris 2025 title decoded: "Satanic obviously being compute, and sublime being the pinnacle of intelligence." (do not repeat "satanic" or "all the way up to God" in Montreal) [POT25 ≈16:33]
- One-slide layout, five rows, white background black type: layer + verb; subnets (first in bold); one dated number. Compute: Lium SN51, Targon SN4, KubeTEE SN90. Inference: Chutes SN64, Engy SN53, gm SN28. Data: Hippius SN75, Desearch SN22, Data Universe SN13. Knowledge: Teutonic SN3, IOTA SN9, Conjectures SN66. Free intelligence: Affine SN120, Albedo SN97. [ecosystem, Deliverable 4]
- On screen: Text version of the five layers in Jacob's 13 Sep order (compute, inference, data, knowledge, free intelligence) with one live subnet each; the numbers go on slide 15's figure. (also fits 15) [chart]
- Chart: none
- Open: [JACOB] final order and names of the layers (compute, inference, storage/data, knowledge, intelligence); whether "storage" and "data" are one layer or two.



## Slide 14 — Bitcoin proved money could exist outside the state. We need to prove intelligence can exist outside the state.

- Point 4: Bittensor changes that: ownership, not access. [Jacob 14 Sep]
- "Making sure that the AI's bottom line is us and not some small group of people." [Jacob 14 Sep]
- "The machines work for me because I hold this token." [Jacob 14 Sep]
- "Don't come work for Bittensor. Come own Bittensor." (also fits 23) [Jacob 14 Sep]
- No pre-mine, 21 million, anyone can mine. (also fits 12) [Jacob 14 Sep]
- Objection 2: "Can't we just buy shares in Anthropic or OpenAI? Same thing." [objection 2]
- Answer: you cannot, they are private and the public gets in at IPO as exit liquidity; shares are permissioned and jurisdictional, TAO is not; shareholders do not control the machine (Microsoft 49% and the Nov 2023 board firing; nonprofit board and LTBT control; June 2026 switch-off with no shareholder consulted); you cannot earn a share by contributing, you can earn TAO by work; a share of the seller is not a share of the thing. Concede: fine hedge for the accredited rich; Bittensor is for the eight billion without a term sheet. [objection 2 answer, 14 Sep]
- "You can buy a share of the company that sells you the intelligence. You cannot buy a share of the intelligence." [objection 2 answer, 14 Sep]
- "A share is a ticket to the profits. TAO is a seat at the machine." [objection 2 answer, 14 Sep]
- Strongest why is a chain: ownership-not-access gives the stakes; "they can defund it" gives the urgency; the mechanism (infinite company that pays whoever lowers the loss; Opentensor 35 to 4) gives the proof. [assessment 14 Sep]
- Gap: one slide listing what a TAO holder concretely owns today: share of emissions, staking power to direct emissions to subnets, right to validate and set weights, right to launch a subnet, the open weights and data produced. If it cannot be filled truthfully, lead with the defund argument. (also fits 20) [assessment 14 Sep]
- "Bitcoin proved there can be money outside the state. Bittensor proves there can be a mind outside the state." (also fits 23) [Jacob 14 Sep]
- "Fiat AI is intelligence by decree. This is intelligence by consent." [Jacob 14 Sep]
- "Bitcoin did not make everyone rich. It made the rules the same for everyone. Bittensor does not make everyone powerful. It makes intelligence something no one can be locked out of." [Jacob 14 Sep corrected core]
- The future where a successful Bittensor matters: a world with two supplies of intelligence, one licensed and switchable, one open and unstoppable; its existence caps price, censorship, and capture in the first. [Jacob 14 Sep corrected core]
- What a person gets: not a dividend but an option, intelligence that takes no orders and needs no permission, at a market price, forever. [Jacob 14 Sep corrected core]
- Definition of a mind outside the state, three properties: no one can license its creation; no one can direct what it says (directed by no memo, and by no fear of blame); no one can switch it off. Same three as Bitcoin. [Jacob 14 Sep, revised]
- Slide idea: black, two lines, then silence. Say "proves there can be", not "has built". [Jacob 14 Sep]
- Bittensor's hope mirrors Bitcoin's three: existence proof (Teutonic 110B, Chutes 2.8M req/day, Engy Kimi K3 on gaming cards, on a ledger tonight), fixed rule (21M TAO on a schedule to producers), and a story with a face (find one real person by the 28th). (also fits 15) [Jacob 14 Sep fiat AI]
- Honest gap: the exit is partial (NVIDIA chips under export control, one key, frontier 10x larger); this is the 2012 hope, not the 2024 one. (also fits 19) [Jacob 14 Sep]
- "the machine that is replacing your job cannot be switched off on you, and you can own a piece of it." [Jacob 14 Sep visceral]
- What the community should feel, in order: recognition, identity ("you are the outside"), responsibility (an exit not maintained closes). The full feeling is only available when the key is gone, so say the date in the same breath. (also fits 22) [Jacob 14 Sep two-worlds session]
- Why Bittensor is uniquely situated to succeed, technically and ethically, for the right reasons. [dictation 21 Sep]
- "So we built one." Pause. "Now you have two." [outline v2]
- What a holder gets today: a share of every block's emission; stake that directs emissions to subnets (voting on what should exist, bottom up); the right to validate and set weights; the right to launch a subnet nobody can stop; the open weights. (also fits 20) [outline v2]
- Emissions follow each subnet's alpha price (a moving average of about four weeks), not market cap; holders vote with money they can lose. drafted alt: "A lab asks its ten biggest customers what to build. We ask anyone who is willing to lose money on the answer." (also fits 20) [outline v2]
- Since dTAO (13 Feb 2025): root stake fell from 70 percent of supply to 47.5 percent; about 2 million TAO, seventeen percent of supply, moved by holders into subnet pools [VERIFY]. Nobody approved those choices. (also fits 20) [outline v2]
- drafted alt: Most of you will only use. Most Bitcoin holders never ran a node. The exit still worked. [outline v2]
- drafted alt: Bittensor does not give rights. Rights are grants; grants can be revoked. It removes the granter. "We are offering you something no one can revoke, because no one granted it." (also fits 20) [outline v2]
- Close, MEMORISE 90 seconds: I began with a switch; the answer from every capital this summer was fewer hands on it; every hand believes it is doing right, until the day it is not, and by then you are not needed so you are not asked. We did not remove their switch; we built a second supply beside it with no switch, so the first switch is worth less and the hand on it has to think about you, because you can leave. (also fits 23) [outline v2]
- drafted alt: Nobody licenses it. Nobody directs it. Nobody switches it off. Use it. Earn from it. Own a piece of it. Not the only hope; the only exit with no one standing in the doorway. (also fits 23) [outline v2]
- drafted alt: "One supply of intelligence in the world with no switch on it. That is the whole job." (also fits 23) [outline v2]
- Why Bittensor does not "give rights": rights are grants and grants carry a signature that can revoke them; Bittensor removes the granter. Use, earn, own are properties of the system, not rights extended, the way nobody gave anyone the right to hold Bitcoin. [trajectory §15]
- It makes people needed again by something that cannot refuse to pay: emissions go by rule, every block, to whoever produces. [trajectory §15]
- Honest part: the network does not care whether the contributor is a person or an agent, so most mining will be machines; what a person can do that a machine cannot is own them. Bittensor does not restore the value of the body; it lets people move from labor to owner before the payroll dies. (also fits 8) [trajectory §15]
- drafted alt: "The state gave you rights because it needed you, and it is about to stop. We are not offering you rights. We are offering you something no one can revoke, because no one granted it." [trajectory §15]
- The protocol pays about 3,600 TAO a day to whoever lowers the loss, on a fixed schedule, toward 21 million, no pre-mine. Miners produce. Validators check. drafted alt: A mind that takes no orders, at a market price, from nowhere in particular. [mind doc]
- Q&A answer (200 words) to "You are describing an AI nobody can control. Why is that good?": lineage, three properties, "Ask who controls the other one", the gap, "proves there can be". (also fits 24) [mind doc]
- 25-word version: "Bitcoin proved there can be money outside the state. Bittensor proves there can be a mind outside the state: no licence, no memo, no switch." [mind doc]
- 60-word version ends: "Small, unfinished, real." (also fits 19) [mind doc]
- Slide instruction: he says the two lines once; the screen holds it; he does not read the slide. [mind doc]
- drafted alt: There is a second price on the screen. The hospital in Montreal gets a second bid. Not always better. Present. When the licensed bill goes up thirty percent, the procurement officer opens a second tab, and the licensed supplier knows she will. It does not have to win. It has to exist and be good enough. [two worlds]
- The second supply caps the price (a subscription can only be raised as far as the open alternative lets it), caps the censorship (a refusal is only a refusal when there is nowhere else to ask), and caps the capture (a licence is worth less when unlicensed intelligence exists and works). drafted alt: The labs can still slow down. Good. What they cannot do in this world is slow down everyone. [two worlds]
- Quebec AI minister, in this city, June: "Sovereignty is not a slogan. It's compute, cloud, data, energy, and talent. And if we don't build the capacity here, we're going to depend on the infrastructure, the rules, and the priorities of somewhere else." In World Two a small country points its ministry at the open stack and buys from a supplier that cannot be told to stop selling to it. (also fits 20) [two worlds]
- drafted alt: In every one of those cases the second supply is chosen for one reason: the first supply said no. [two worlds]
- drafted alt: Not a dividend. I used to say it loosely and I was wrong. What you get is an option. That is not everything. It is the thing you cannot get in World One at any price. [two worlds]
- Three conditions on the second supply: Does it exist? Yes. Is it good enough? Partly (best open models trail by months, not years; a Chinese lab released a 500B model under an open licence at about one-twentieth the price; the gap is a gap, not a wall). Can it be told to stop? The one not yet true. (also fits 21) [two worlds]
- drafted alt: Bittensor is the only funder of open intelligence that has no desk. There is nobody to call. Except me. [two worlds]
- Jacob: "The best teachers will be LLMs. The best feedback will be monetary. The best models will train competitively. The best AI company will be co-owned. The best software will be open source. And there is only one way to get all of it." (788 likes) [tweet 13 Jun 2026]
- Jacob: "The answer to fears of AI is this, not top down censorship or unchecked acceleration, but human unity with the machines." (390 likes) (also fits 23) [tweet 11 Feb 2025]
- Jacob: "Decentralized AI isn't about training collectively. / That's cool, but 99.9% of people can't train models. / Instead, Decentralized AI is about censorship resistance, access, and ownership. / We should collectively own AI, not just contribute to it." (203 likes) [tweet 5 Apr 2023]
- Jacob: "If you're not asking "who really owns this?" you're not working on decentralized AI. / If you're not asking "can someone turn this off?" you're not working on decentralized AI." [tweet 5 Apr 2023]
- Jacob: "Open AI "democracy" will be patronizing conversations with ChatGPT for the purpose of people being "heard", then behind the scenes everything decided by the same people. / If you don't own it, you won't control it. / You won't be heard either." (also fits 9) [tweet 26 May 2023]
- Jacob: "It's all transparent. It's all available. It's all accessible. It's all permissionless." [DOC 50:00]
- Jacob: "Everyone can own Bittensor. Everyone can build on Bittensor. Everyone can see how Bittensor works." [CT26 ≈1:25]
- Jacob, 2024: "You can come, you can own it, you can use it, you can control it, you can build it, you can contribute to it." Five verbs; consider "Use. Earn. Own. And through owning, point it." [CoinFoin ≈1:00]
- Jacob, 2024: "I like to say come be on our board of directors, which means you hold the keys to this thing and you can own it, you can viscerally control it." Collides on purpose with the key section: true of everything but one key. (also fits 21) [CoinFoin ≈2:16]
- Jacob: "people didn't know about TAO, but they were using those tokens." "we've really built products that seep into the real world and are used by people without even needing to know about TAO." Supports the Use verb. (also fits 15) [CT26 ≈3:35, ≈3:55]
- Jacob, 2024: "It's like a company. It's like Google maybe. But it's also a nation state, because there's a democracy inside of it." Do not quote the nation-state lines on stage; a third country is a third switch. (also fits Do not say) [CoinFoin ≈0:18]
- Thiel: "Even being in a stock, you're effectively being in something that's like a government-linked entity." (cut before "woke companies") Objection 2 support: a share of the lab is a share of something the state can direct. [Thiel, Bitcoin 2022 Miami, 7 Apr 2022]
- Say instead: "Hold TAO and you hold a fixed share of what this machine pays out, forever, and the right to point it." (not "a slice of every unit of intelligence") [fact-check]
- DeepSeek V4.1 Flash (10 Sep): 552B parameters, MIT licence, $0.30 per million input tokens versus $5 for Claude Opus 5. The gap is a gap, not a wall. [lab closure]
- Top five open-weight models on the Artificial Analysis index (31 Aug) are all Chinese; best US open model (Nvidia Nemotron 3 Ultra) scores 38 versus Opus 5 at 63.1. (also fits 9) [lab closure]
- On screen: black slide, two white lines. He does not read the slide. (also fits 23) [charts-plan]
- Chart: none
- Open: [JACOB] "proves there can be" vs "has built"; whether the two-line close sits here or at slide 23; confirm the "what a holder owns today" list can be said truthfully.



## Slide 15 — What we have done since I last spoke: Lium, Targon, ...

- The face of the talk is the subnets: the teams Bittensor enabled to create decentralized products they might never have had the option to build. "we give you a chance, because we have shown we can actually beat the frontier; here are the ways." [Jacob 17 Sep]
- TAO holders vote on these projects with their stake and express what the world needs, bottom up. What they built: decentralized training, GPU marketplaces, and more. Frame the ecosystem as empowerment. [Jacob 17 Sep]
- Honesty: say "could be trained", not "has trained"; Teutonic-II at 110B is the largest permissionless run, the licensed frontier is 10–20x larger. "It has not happened yet. That is the job." (also fits 19) [Jacob 14 Sep]
- NeurIPS: Jacob gave a sponsor Expo talk, NeurIPS 2024, Vancouver, 10 Dec 2024, "Incentivizing Collaborative AI"; say "I gave a talk at NeurIPS", not "presented results". (also fits 2) [fact-check 16 Sep]
- A year ago we were just beginning to solve the tasks needed for people to rent machines. Now Lium is a stable application serving [millions of dollars] of compute to engineers inside the ecosystem and outside [VERIFY figure]. [dictation 21 Sep]
- Chutes and Engy are functional inference platforms moving toward general use as the backends for tooling like Hermes, Cursor, Codex. [dictation 21 Sep]
- GM gives access to frontier models. Hippius: working storage and bandwidth. [dictation 21 Sep]
- "We've built tooling that works. We now need to take it to market." (also fits 20) [dictation 21 Sep]
- This is Bittensor working in unison: compute from Lium, inference from Chutes and Engy, storage from Hippius, inference optimization from Cacheon [VERIFY subnet name and netuid]. (also fits 17) [Jacob outline 21 Sep]
- Three eras, one breath: missionaries on CPUs, then gamers, now industry, with margins. [outline v2]
- Jacob: "Lium billed 964,000 dollars in a month and burned about a million dollars of its token with the proceeds. Block 9032919. Check it." [VERIFY] [outline v2]
- Chutes serves open models inside sealed hardware, two to three million requests a day, and published a year of its traffic with Harvard and Chicago [VERIFY]. [outline v2]
- Engy serves the largest open-weight model released, 2.8 trillion parameters, on eighty gaming cards; launched at half the maker's price and about a third less today [VERIFY]. [outline v2]
- Jacob: "GM had Astra before OpenRouter." Hippius: a hundred terabytes of egress in a month billed at zero. [outline v2]
- Teutonic-II: a hundred and ten billion parameters, anyone can join, nobody is whitelisted, the largest permissionless run anyone has tried to train all the way; not finished [VERIFY live status on the day]. [outline v2]
- The faces: Jon Durbin, a backend engineer from Michigan, no lab, no PhD, no VC: Bittensor "acts as a completely decentralized VC fund in a way, so we don't have to go perform huge raises and give away large portions of the company in exchange for funding." [outline v2]
- The faces: "Fish", a pseudonymous ex-miner who ran more than a hundred and twenty gaming cards and built the GPU market he wished he had: "Real businesses on Bittensor work." [outline v2]
- The faces: Pierre Manseau, who built one of Montreal's first internet providers in 1992, mines since February 2024, runs a subnet alone at sixty, and speaks in this building at 14:00 today. drafted alt: We gave them a chance. Nobody else did. [outline v2]
- Honest numbers on all three, first: Chutes' parent took about a quarter of all emissions in April and its revenue covers a third of its costs by its own count; nobody knows who runs the biggest business on the network; Pierre's subnet is two months old, three miners, about four percent of emissions. [outline v2]
- Honest count: 128 subnets. 23 live with paying customers. 72 live without. 33 stalled. Jacob: "Ten months ago I said no subnet had serious revenue. I was right." Revenue offsets between five and ten percent of what the protocol pays out. drafted alt: A market that lets things fail is the point. (also fits 19) [outline v2]
- Jacob: "It might not be much, but it might also be the first ever incentivized pretraining." [tweet 4 Aug 2023]
- Jacob: "Tensorplex beat Meta's llama-model with a cluster 50x smaller on a 6 month old Bittensor market. / That's equivalent to five A100 -> H100 hardware upgrades in half a year." [tweet 28 May 2024]
- Jacob: "I have 6 agents running now, two are doing ML research, one is cloning Claude code, two trade hyperliquid and one mines Bittensor. / All they need is compute from Bittensor (Lium, Basilica and Targon) and inference from Chutes and Openrouter. / The trading agents pay for the compute and inference by buying credits with proceeds." (2,191 likes) (also fits 13) [tweet 12 Mar 2026]
- Jacob, Paris 2025: "We're at a point where Bittensor has stopped being a game. It's no longer gamers. It's no longer a faux competitiveness. It is industrial scale." [POT25 ≈10:24]
- Jacob, Paris 2025: "Where is the best place to take a computer that's not in the data center but run by some homegrown individual in their basement? We're talking just a single 8x H200 box. Where's the best market for that? Well, the best market for that is actually decentralized training." [POT25 ≈12:35]
- Jacob, Paris 2025: the missionary era, CPUs, "people were making exorbitant amounts of money off doing nothing"; admissions before boasts: "None of them are mine." "We've [expletive] up a lot." [POT25 ≈8:10]
- Jacob, 31 Jul 2025 (about Lium): "Take the frontend away, take the leader away and it still runs, a floating machine in space that prints cash." (also fits 21) [tweet 31 Jul 2025]
- Press risk: Bitcast (SN93) pays YouTube channels for "These Subnets Will Crush Big Tech!" videos with "250 plus % APY" pitches; if Bitcast is named, be ready for "your network pays influencers to shill your token". Answer: brands pay for verified attention; Q2 2026 revenue exceeded miner pay; "do not chase emissions". Do not raise unprompted. (also fits 24) [ESI note]
- Subnet map: 128 slots; 23 live with paying outside customers, 72 live without evidence, 33 stalled or empty; by group: 11 compute, 12 inference, 19 training, 11 data and storage, 13 agents and tools, 14 knowledge and search, 17 finance and prediction, 9 media, 9 science, 13 other or empty. [ecosystem]
- Five layers with a checkable number: Lium 1,449 GPUs listed and 1,227 rented (17 Sep); Chutes 2.37M requests a day (7-day avg to 16 Sep); Hippius 100 TB egress billed at $0 (May); Teutonic-II 110B permissionless pre-training run (Aug); Affine paying five open slots about $50,000 a week each (Aug). [ecosystem]
- Compute, buy: "You can rent a GPU tonight with no account check, no ID and no sales call. Lium does that. Last month 1,187 people and agents paid it $964,000, and right now 1,227 of its 1,449 cards are rented." [ecosystem, 9/17 Sep 2026]
- Compute, sell: "You can point the graphics card in your bedroom at the network and be paid by a stranger on another continent. Lium lists cards from 68 datacenters in 21 countries; Engy paid 77 miners $12,700 last week for serving open models on gaming cards." [ecosystem]
- Inference, open: "You can run a 2.8-trillion-parameter open model that no lab licensed to you. Engy serves Kimi K3 on eighty RTX 5090s. Its miners answered 3.8 million requests last week." [ecosystem]
- Inference, private: "You can call any open model through an API that cannot read your prompt. Chutes does that inside sealed hardware, 2.4 million requests a day this week, and 315,000 different users in the year Harvard measured." (arXiv 2608.13573) [ecosystem]
- Inference, the labs' models: "You can buy Claude and GPT below their makers' list price through an enclave the router cannot see into. gm does that; 96 billion tokens so far, and it listed GPT-6 Astra a week before OpenRouter did." [ecosystem]
- Training: "You can start a training run across strangers' machines. I did, on IOTA: 128 nodes, five minutes, five TAO. Gamers in 16 countries have joined its Train at Home." [ecosystem, 3 Jun 2026]
- Free intelligence: "You can make an open model better and be paid the day you do, with no application and nobody's approval. Affine pays five open slots about $50,000 a week each. After the June base switch its miners crowned five champions in five days." (also fits 17) [ecosystem]
- Knowledge: "You can be paid for a mathematical proof no one else has found, checked by a machine, not a committee. Conjectures paid this month for the near-half-density case of Green's Problem 51, open for sixteen years." [ecosystem, 9 Sep 2026]
- Data: "You can store your files where egress is free. Hippius served 100 terabytes in a month and billed zero for bandwidth; AWS would have billed $8,000." [ecosystem, 8 May 2026]
- Agents: "You can write a shopping agent in Vancouver and be paid by people who never met you. Oro paid 1,206 miners $200,000 in its first 36 days. Then Y Combinator followed the token." [ecosystem]
- Alternate, media: "You can be paid as a creator by a brand with no agency in between. Bitcast's second-quarter revenue, $105,499, exceeded what it paid its miners. The incubators told its founders the idea did not belong here." [ecosystem]
- Emission split this week: compute, inference and training take 59% of new TAO (24.0%, 15.5%, 19.8%); science 14.5% (Minos genomics, NOVA drug discovery, OpenRoboto robot policies). Holders fund infrastructure and research first, consumer apps second. (also fits 20) [ecosystem]
- Faces constraints met: three pseudonymous teams (Fish, Affine team in China, Dendrite); four teams outside US/Europe (Tbilisi, China, Kampala/Nairobi, Hanoi) plus three Canadian; one solo builder (Pierre Manseau); two individual miners who never owned a subnet (Vincent Rapido, "@hodler0_"). [ecosystem]
- Face: Giga Chkhikvadze and Mirian Okradze, Desearch (SN22), Datura, Tbilisi. Face: Koyuki Nakamori ("Special K"), Perturb (SN26) and Umi (SN78). Face: Ning Ren, Engy (SN53), Hanlin AI. [ecosystem, Deliverable 2]
- Face: Fred Esere and the Polaris Cloud team, formerly SN49 (deregistered 1 Nov 2025), Kampala/Nairobi; an honest loss. Face: Thanh Luan Tran ("toilaluan"), NicheImage and Condenses, Hanoi. Face: Dendrite and "@Wejh69", Conjectures, Teutonic, Albedo, SOMA, pseudonymous studio. [ecosystem, Deliverable 2]
- Face: Vincent Rapido, individual miner (SN27), then Miners Union and TaoSwap; a miner, not an owner. Read-aloud anonymous voice: "@hodler0_". [ecosystem, Deliverable 2]
- Face: Will and Tom Blears, Bitcast (SN93). (press risk noted on the ESI bullet) [ecosystem, Deliverable 2]
- Footer line for the layers slide: "128 markets. 23 with paying customers. 2 million TAO moved by holders to choose them. One key left to remove." Technical alternative: "About 3,555 TAO a day, roughly $800,000, paid to whoever produces, under rules the same for all." (also fits 13) [ecosystem]
- Say instead: "Sixty-three percent of last month's rentals were started by software, not by a person clicking." (rentals, not renters) [fact-check]
- Say instead: "A team serves the largest open model on earth, 2.8 trillion parameters, on eighty gaming cards. It launched at half the maker's price. Today it is about a third less. Every answer carries a proof." [fact-check]
- Say instead: "Revenue offsets between five and ten percent of what the protocol pays out. Closing that gap is the job of the next eighteen months." (also fits 20) [fact-check]
- Say instead (revenue): "About fifteen to twenty million dollars a year you can check on-chain. A consultancy that counts more generously says thirty." [fact-check]
- Safe wording, Teutonic-II full: "Teutonic-II is the largest permissionless training run anyone has attempted. One hundred and ten billion parameters. No whitelist. No résumé. Any GPU, any algorithm, any person. It is not finished, and I will not tell you it is the best model in the world. I will tell you no lab can run this experiment, because the moment you let strangers in, you need a market, not a manager." [Affine fact-check]
- Teutonic-I lesson, full: "Our first run taught us something. The market found the cheapest path to a lower loss, and that path ran through borrowed open weights. That is not a scandal. That is what a market does when you do not verify. Teutonic-II verifies." [Affine fact-check]
- Figure spec: Figure 8, "Layer by layer": five stacked bands bottom to top, layer name black, subnets grey, one number right-aligned in mono. Compute: Lium SN51, Targon SN4, 1,449 GPUs listed, 1,227 rented (17 Sep), $964K billed in a month (9 Sep). Inference: Chutes SN64, Engy SN53, 2.4M requests/day (16 Sep), 2.8T model on 80 gaming cards (27 Jul). Data: Hippius SN75, 100 TB egress billed $0 (8 May). Knowledge: Teutonic SN3, 110B permissionless run (24 Aug), not finished. Free intelligence: Affine SN120, five open slots ~$50,000/week (23 Aug). Annotation: "No account. No ID. No permission." [VERIFY every number 27 Sep; Cacheon not on the figure until Jacob confirms name and netuid] [chart]
- Chart: Layer by layer
- Open: [JACOB] Lium figure (he said "tens of billions to several millions"; on-chain says $964K billed in a month); Cacheon name and netuid; which faces to name (Jon Durbin, Fish, Pierre Manseau); whether to name Bitcast.



## Slide 16 — We have built the foundations of what is needed to build the mind outside the state.

- So we built one: a market, and tooling that trains models with capital. The models trained are the size of what frontier labs were building two or three years ago. [Jacob 16 Sep]
- "We don't need to be first. We need to exist, and we need to prove who we are and what we're doing." [Jacob 16 Sep]
- The previous years were showcasing what the tooling can do; this is the evidence that the tooling works and can surpass and out-compete the frontier. [Jacob 16 Sep]
- "Today's talk is not about can we all compete. It's just, can we compete? Because the need for this competition is profound." [Jacob 16 Sep]
- Close of the middle: "I began this talk by talking about how we are pushing to a real frontier, giving non-KYC access to people who want compute and inference, showing we can compete on price and performance, proving to the world that we can produce things that are truly state of the art, but in a system that is disjoint from the world. It's why we can give up on being first right now. What we need to achieve is existence." [dictation 21 Sep middle]
- Coordinator note: the opening puts the state of the union (light, honest, concrete) before the wall; the Kennedy move: concede "we are behind" early and pair it with proof. [coordinator note 21 Sep]
- On screen: None planned; text only (the turn from tooling to evidence). [chart]
- Chart: none
- Open: [JACOB] whether the exploit turn ("so we built one") sits here or on slide 7; how to reconcile with the dictated "we are behind" opening.



## Slide 17 — And now we are here: Affine-I.

- Jacob's claim: Affine's latest is a mixture-of-experts model, comparable to Qwen at that size, on par with Claude Opus 4.6 [VERIFY; fact-check says numbers swapped and Opus 4.6 does not hold]. [Jacob 16 Sep]
- Affine's base is Alibaba's Qwen3.6-35B-A3B (35B total, 3B active); the base matches Claude Opus 4 (May 2025) and Sonnet 4.5, not Opus 4.6 (Feb 2026). [fact-check 16 Sep]
- Safe line: "Affine takes an open model from Alibaba, thirty-five billion parameters, three billion active, and pays anyone on earth to make it better. The base already does what Anthropic's frontier did a year ago. The mechanism is the product. The model is the receipt." (also fits 18) [fact-check 16 Sep]
- Two exposures to prepare and say first: Affine's repo has an operator mining folder ("does the operator mine its own subnet?"); the evaluator was gamed and the throne reset on 27 Aug. (also fits 7) [fact-check 16 Sep]
- "We are behind. There's no denying it. But for the first time ever, Bittensor is training models that can be successfully used by individuals." [dictation 21 Sep]
- Affine One: a model trained by miners on Bittensor, in a decentralized commercial setting, under an incentive mechanism, with compute sourced from individuals across the world, inference from international networks. [dictation 21 Sep]
- "This is a breakthrough. A year ago we were not here." [dictation 21 Sep]
- Affine One: miners took an open model from Alibaba, thirty-five billion parameters, three billion active, and made it better. Post-trained, not pre-trained; say it first. The scorer was gamed twice this summer and the throne was reset in August; that is a live market. (Never "trained on Bittensor" for Affine.) [outline v2]
- Jacob, Paris 2025: "I really believe the best market for intelligence is intelligence itself. I used to say the best measure of intelligence is intelligence itself, but I think the best market is." Explains why Affine sits on top of compute, training and serving subnets. (also fits 7) [POT25 ≈12:03]
- Paris 2025 Affine design: Celium compute, Gradients training, Chutes serving. [POT25 ≈15:59]
- Safe wording: "Affine takes an open model from Alibaba, Qwen3.6, thirty-five billion parameters, three billion active, and pays anyone on earth to make it better. The base model already does what Anthropic's frontier did a year ago on real coding tasks. In June the miners crowned five champions in five days. The mechanism is the product. The model is the receipt." [Affine fact-check]
- Safe wording, post-training: "On Affine, anonymous engineers took an open model and, in five days, crowned five successive champions on the tasks the network pays for. Nobody hired them. Nobody approved them. The token did." [Affine fact-check]
- On screen: Jacob's asset: the Affine-I reveal (name, base, what it is). No benchmark bars here; those are slide 18. [chart]
- Chart: none
- Open: [JACOB] the exact name (Affine One / Affine-I) and how much of the reveal is held back for a novelty moment.



## Slide 18 — Best on, SOTA on, Pareto frontier in, etc. [needs Affine-I's actual benchmark table; the fact-check found no public score for fine-tunes as of 16 Sep]

- No Affine fine-tune has a public SWE-bench score; Affine's own repo says crowns do not imply benchmark capability. [fact-check 16 Sep]
- "Best model ever trained outside a lab" does not hold: Covenant-72B and Nous Consilience 40B beat Bittensor's from-scratch runs; Affine models are post-trained Alibaba models. (also fits Do not say) [fact-check 16 Sep]
- "We are still behind." Affine One is not the state of the art, not the best model in the world. It competes at the 35B mid-range with best-in-class models from companies like Alibaba and Chinese labs. [dictation 21 Sep]
- "We shouldn't pretend that this is world-changing technology yet." [dictation 21 Sep]
- Safe wording, shorter: "Three billion active parameters, open weights, and it codes like a frontier model did a year ago. Anyone can improve it and get paid the moment they do." [Affine fact-check]
- Do not put any Affine internal score next to an Anthropic score. (also fits Do not say) [Affine fact-check]
- Figure spec: Figure 7, "Affine One, honestly": four horizontal bars, SWE-bench Verified, maker-reported: Claude Opus 4 (May 2025) 72.5; Qwen3.6-35B-A3B, Affine's base (Apr 2026) 73.4, black; Claude Sonnet 4.5 77.2; Claude Opus 4.6 (Feb 2026) 80.8, grey. Fifth empty row: "Affine One fine-tune: no public score". Annotation: "The base already does what the frontier did a year ago. We are behind. There is no denying it." [JACOB: Affine-I's actual benchmark table; add the bar in black if a public score exists by 27 Sep] Never "on par with Opus 4.6". [chart]
- Chart: Affine One, honestly
- Open: [JACOB] supply Affine-I's actual benchmark table; there is no public score for any fine-tune as of 16 Sep; decide what "best on / SOTA on / Pareto frontier in" can truthfully say; answer for "does the operator mine its own subnet?"



## Slide 19 — This is not everything; we will keep pushing these fronts and keep decentralizing.

- Next year is the year of acceleration; this year is about stitching Bittensor together. (also fits 20) [dictation 21 Sep]
- "This is why I recommend that people in this network build their systems so that they can be live without our control." Examples: the foundation stepping in to help people who have been hacked; funding the work in the system; things it may not always be able to do. [dictation 21 Sep future]
- The gap, before you ask: our largest run is 110 billion parameters; the licensed frontier is ten times that; the mind is unlicensed, the chips are not; one key can still change these rules, and it runs through me. (also fits 21) [mind doc]
- drafted alt: I am not going to sell you a utopia, because it is not one. The producers are concentrated, prices are set by markets, and markets are not kind. Most people in Lima never mine anything. What is different is three rules. It is enough. [two worlds]
- Jacob: "The real death knell of a network is stagnation, not volatility." [tweet 11 Jun 2024]
- Jacob: "Bittensor will be run by agents. They will feed the mining, resist the exploits, manage the fleets, build the subnets and consume the commodities" (938 likes) (also fits 7) [tweet 14 Mar 2026]
- Jacob: "In Bittensor's final form no one will ever write a miner client." [tweet 12 Feb 2026]
- Say instead: "Our largest run is a hundred and ten billion parameters. The biggest open models are fifteen to twenty-five times that. The closed labs do not say." [fact-check]
- Figure spec: Figure 9, "The honest count": one stacked horizontal bar, full width, 128 subnets: 23 paying (black), 72 live without paying customers (grey), 33 stalled (hairline outline); three numbers beneath. Annotation: "A market that lets things fail is the point." Say the revenue-vs-emissions number (~$17M vs ~$294M) out loud, not as a chart. [VERIFY re-count 27 Sep] [chart]
- Chart: The honest count
- Open: [JACOB] how many "not finished" concessions to make in a row (Teutonic-II, chips under export control, one key) before the goal slide.



## Slide 20 — Goal this year: X the amount of exterior demand into the Bittensor universe; shore up the core of the network; finally decentralize it. [JACOB: the X]

- Three pillars of a successful year: (1) shoring up the incentive market so it is aligned and never needs centralized control from the Triumvirate; (2) continued advancement of model capabilities and results, Affine and beyond; (3) bridging alpha tokens outward into the world, expanding Bittensor alpha exposure, and making alpha a first-class primitive in the emission vector [needs Jacob's three sentences on what this means technically]. [Jacob outline 21 Sep]
- On the Dec 2025 tweet that no subnet has serious flow-through revenue vs the $28–35M figure: Jacob asked us to verify this ourselves. [Jacob 12 Sep]
- Pillar 1 detail: the emission rule changed three times in sixteen months, tuning in public, and this is the year the tuning stops; drafted alt: a market with a human at the top is a scorer with a hole in it. [outline v2]
- Pillar 2 detail: Teutonic-II to a real token budget; a public benchmark beside Affine's internal one. [outline v2]
- The ask, if you hold: you are not a customer, you are on the board; Jacob 2024: "come be on our board of directors, which means you hold the keys to this thing"; stake to what should exist; do not chase emissions; do not dump on the newcomers in this room; "nothing free, nothing given, everything fought for"; hold through the year it takes to remove the key; "Come and be Bittensor." (also fits 23) [outline v2]
- The ask, if you build: Jacob: "if people mine your protocol, they care." About 128 markets, and there is room for yours. Build for a customer, not for emissions. Build so it runs without us. No HR department, no visa, no résumé. Jacob: "Don't submit a resume. Submit a model." (also fits 23) [outline v2]
- The ask, if you are new: pick the second supply once. Use, earn, own. Jacob: "I believe in the invisible billion geniuses of this world." "You don't need no visa to mine a permissionless network." (also fits 23) [outline v2]
- The ask, engineers in Montreal: this city built the last open network; Pierre is in the building; one team serves a 2.8-trillion-parameter model on eighty gaming cards; come make it faster. Distributed has to be at least as good as centralized; that is a spec, not a wish. (also fits 23) [outline v2]
- Jacob, Paris 2025: "Bittensor is a federation. We're trying to create a thousand-year federation. So I'm not the CEO. I'm just one of the people here that really believes in this technology." drafted alt: You are the outside. An exit that is not maintained closes. (also fits 21) [outline v2]
- Thiel reserve for the Earn verb: "You can't start a dotcom company from your basement in Sacramento. You can't start an AI company, you have to do it in San Francisco ... It has to be at an enormous scale." (reported, Guardian, Sep 2025) [Thiel, SF lecture Q&A]
- Governance alternate: "You can decide what gets built. Two million TAO, about 450 million dollars, has been moved by holders out of the passive pool and into specific subnets since February 2025. Nobody approved those choices." (also fits 14) [ecosystem]
- Bottom-up governance measured: at dTAO launch (13 Feb 2025) 70.0% of all TAO sat on root; on 17 Sep 2026, 47.5% sits on root and 2.01M TAO (17.4% of supply, about $453M) moved by holders into subnet pools, across 128 subnets instead of 64. [ecosystem]
- Emission concentrated by design: top 10 subnets take about 61% of new TAO (54.7% actually injected over the week); 41 subnets get more than 0.1%; 59 get under 0.01%. Before dTAO the top 10 took 44%. Concentration is the vote, and the critics' main point. [ecosystem]
- Honest critiques to say first: root stake still earns from every subnet without choosing; top 10 validators about 65% of root voting power (Apr 2026); emission rule changed three times in 16 months; owners get 18% and can buy their own token; several subnets buy from other subnets with emissions; one sudo key can change the rules. (also fits 21) [ecosystem]
- Compared with a lab: a frontier lab's roadmap is set by a few executives and its largest customers; on Bittensor about 2M TAO of holder stake, movable at any block, decides which of 128 markets gets paid tomorrow. Neither is perfect; one is contestable by anyone with a wallet. [ecosystem]
- Figure spec: Figure 10, "Where the vote went": one black line, nine points, share of all TAO on root: 13 Feb 2025 70.0% → 15 Dec 2025 53.9% (first halving) → 17 Sep 2026 47.5%; two labelled dots. Annotation: "About 2 million TAO moved by holders into markets. Nobody approved it." Notes: flat after March 2026; top ten markets take ~61%. Three pillars: three sentences in a column, not a drawing [JACOB: pillar 3 sentences]. [VERIFY root stake 27 Sep] [chart]
- Chart: Where the vote went
- Open: [JACOB] the X (how many times exterior demand into Bittensor); three sentences on pillar 3 (alpha as a first-class primitive in the emission vector); whether to say revenue offsets 5–10% of emissions.



## Slide 21 — "People have critiques about me as a centralized figure, and I have not given up power yet because we still have time to build the foundations of this network in the most aligned way possible."

- Consequence for the sudo key: the moral claim is "rules nobody can change", so the one person who can must go and the date must be public. The ossification date is where the claim becomes true. Pivot of the talk. (also fits 22) [Jacob 14 Sep corrected core]
- "the number of people who have to say yes before you can be switched off; for us that number is going to zero" [Jacob 15 Sep]
- "There is one key. It is mine. On the day it is gone, this stops being a promise and becomes a property." (also fits 22) [Jacob 15 Sep]
- Approach it head on. "One of the first responses to all of this is: well, you're not decentralized. It costs you [on keys; meaning the sudo key sits with you]. And this is true." [dictation 21 Sep future]
- "We've gone from an organisation in Canada to effectively a small group of individuals, including myself, that controls the key and [the upgrades] of the chain." [dictation 21 Sep future]
- "We are built on a technology that can and will find decentralization. A chain that can flip a switch and become decentralized. A chain that can flip a switch and turn off my power, or anyone's power. There is no company on earth that has that ability. A stockpile of GPUs that provides the inference for [a company] can never be decentralized. A company like OpenAI can never be decentralized." [dictation 21 Sep future]
- "I'm asking this community for a little bit more time. We understand there is a ticking timeline. Behind us are the regulators, and they want to close our AI, tear it down, regulate it, turn it off. And there is a chance that one day I will need to go into hiding." [dictation 21 Sep future]
- Coordinator note: "asking for a little more time" needs the date or the list of what gets done; without the list it is an apology, with the list it is a plan. The "flip a switch" framing is strong and true: a protocol can remove its own ruler, a company cannot. (also fits 22) [coordinator note 21 Sep]
- Bittensor is not fully decentralized today. The foundation is a permissionless proof-of-authority blockchain; a sudo key can upgrade the chain and take executive action; that key runs through Jacob. [context: honest state]
- Bittensor can move to proof of stake or proof of work and become truly governed internationally, with no single group in control; this path is the difference between Bittensor and centralized AI companies. (also fits 22) [context: honest state]
- On the Sept 2024 promise of proof of stake "by end of next year": Jacob says there is still time. Do not hide it; own the timeline. (also fits 22) [Jacob 12 Sep]
- Softened form: "well, you're not decentralized. There is a key, and it's yours. And this is true." [outline v2]
- Jacob's own slide from 22 September 2024 on screen: "The chain operates on a Proof of Authority (PoA) mechanism, with plans to transition to Proof of Stake (PoS) by the end of next year." Stamped: Missed. Same evening: "The key is a multi sig but it does have root privs which is dangerous -- ill admit that." [outline v2]
- July 2024: the key halted the chain in about half an hour during a wallet drain; the same key could be used badly. [outline v2]
- drafted alt: Smaller on purpose, on the way to zero. [outline v2]
- Jacob: "even Bitcoin wasn't decentralized. It was one guy for the first 3 years." [outline v2]
- Jacob: "I consider a success that I'm not needed anymore." "Take the leader away and it still runs, a floating machine in space that prints cash." drafted alt: The key is the last place I am still necessary. [outline v2]
- Jacob: "The goal of power is to give it away. The art of power is its decentralization." (also fits 22) [outline v2]
- The honest condition: production and funding pass the "nobody can be sent the letter" test; control does not; in 2024 he promised proof of stake by the end of 2025 and missed; the new date is public; until the key is gone it is a promise with a deadline, said first, by him. [trajectory §15]
- drafted alt: I do not get to stand here and describe a world with no switch while holding one. So the key goes, and the date is public. [two worlds]
- drafted alt: That date is not a detail at the end of the talk. It is the day the claim I have been making for four years becomes true instead of intended. (also fits 22) [two worlds]
- Hostile push 3: "You just described a world with no switch while holding one. Your key is the switch." Answer: Yes, and that is why the date is public: December 2027 the key is gone and the rules freeze, and until that day I am asking you to hold me to it, not to believe me. (also fits 24) [two worlds App. C]
- Jacob, on the Altman firing: "Power came from above when firing Sam. / Where it was, and always will be until we draw it down into the hands of many. / Decentralization is just this: forcing the separation and distribution of control. / Pulling it down." (314 likes) [tweet 18 Nov 2023]
- Jacob: "we got to subnets on Bittensor because I lost my laptop and realized there was no entry for me in the network we created. / Decentralization is all about this: / we're less important than we think we are. So who's next?" [tweet 5 Oct 2023]
- Jacob: "Are we fully decentralized? No, it's going to take some time." [PBW ≈18:30]
- Jacob: "This is the role that we play, reluctantly, while it's still required." [TWIST 37:45]
- Jacob: "We could clutch our pearls and refuse to give up power with decentralization. I think that would be a risk." [TBR 69:20]
- Jacob: "You're not as good as your community, right?" [UNSH 32:05]
- Jacob, Paris 2025: "As someone that's been putting my lifeblood into making sure that it's well organized, I consider a success that I'm not needed anymore. And I had a bit of a hard time writing this presentation because I was like, why am I on stage?" [POT25 ≈14:48]
- Figure spec: Figure 11, "Nobody can be sent the letter": three rows, filled black square = decentralized, hollow = centralized. Production: decentralized, concentrated like Bitcoin's hashrate (three pools mine more than half; none can print). Funding: decentralized by construction, ~3,555 TAO/day. Control: centralized, one key, multisig through Jacob, date December 2027 [JACOB]. Annotation: "The third has a deadline." Plus the 2024 tweet as plain text in the deck typeface (not a screenshot), date line "Missed. New date: December 2027." until Jacob confirms. [chart]
- Chart: Nobody can be sent the letter
- Open: [JACOB] keep or cut "there is a chance that one day I will need to go into hiding" (it will be the headline); how to handle the 22 Sep 2024 PoS-by-end-of-next-year slide ("Missed"); the 2024 "you hold the keys" line: address or leave.



## Slide 22 — The timeline and what needs to happen before then. [JACOB: dates]

- "I believe we still have more time. And here are the things that we need to get done in that time." Jacob to list: roadmap items to decentralization; nominated proof of stake; governance pallet; sudo removal; date. [dictation 21 Sep future]
- The list, with dates [JACOB]: nominated proof of stake; governance pallet; sudo removal; ossification. Draft, his to confirm: nominated proof of stake within about a year; on-chain governance whose last phase is a vote to disable the key; full ossification, December 2027. Say it only if you will be held to it. [outline v2]
- Why the first date slipped, said unprompted: we rebuilt the economics first; I fixed what pays before I fixed who governs; right call then, wrong call now. [outline v2]
- drafted alt: As long as the key exists, the second supply has a switch in it, and the switch is in my hand. [outline v2]
- drafted alt: "Screenshot this slide. Hold me to it. Do not believe me." [outline v2]
- Draft sequence in one breath: "It goes: proof of stake within a year, the key disabled, the rules frozen in December 2027." [JACOB: dates] [mind doc]
- On screen: Text, not a figure: the list as four rows with a date lead: nominated proof of stake; governance pallet; sudo removal; ossification. Three leads read "date to confirm" until Jacob supplies them [JACOB]. [chart]
- Chart: none
- Open: [JACOB] the dates: nominated proof of stake; governance pallet; sudo removal; ossification (draft: PoS within a year; governance whose last phase disables the key; full ossification December 2027). Say only what he will be held to.



## Slide 23 — Thank you. Questions.

- Purpose of the talk: rally the community and give direction; state the mission; make holders hold, builders build, newcomers excited. [context: purpose]
- Jacob: "Intelligence is cheap. / Spirit is rare. / Never give up." (277 likes) [tweet 23 Dec 2024]
- Jacob: "It's not can you out-compete, it is can you out-contribute." [DOC 61:33]
- Jacob: "Own Bittensor. Don't come work for Bittensor. Come own Bittensor. Come and be Bittensor." [CT25 ≈22:00]
- Jacob: "I don't actually really sell TAO the token. I sell that you can be part of this." "The reason why we have TAO in the first place is so that everyone can have some, and that everyone can join." [TWIST 60:00, 79:20]
- On screen: None. Black slide, "Thank you." or the two-line close held on screen. [chart]
- Chart: none
- Open: [JACOB] the printable last sentence; whether to end on the two black-slide lines or on "Thank you. Questions."



## Slide 24 — 15 minutes of questions.

- Q&A prep: "Does the operator mine its own subnet?" answer needed (Affine repo has a mining/ folder). [outline v2]
- Logistics: slot before or after Pierre Manseau at 14:00. [outline v2]
- Q&A on "how do you measure intelligence": "The original design was: we'll pay you if you're closer to AGI. That doesn't work. It's too ill-defined. So we constrain the degrees of freedom of the miners" (POT25 ≈8:48 and ≈17:24, lightly joined). [POT25]
- Q&A: "I can define my incentive mechanism as just: train a model by giving me the data and the Python script, and you can't do anything else." [POT25 ≈18:37]
- Q&A only: the Jensen Huang "mentioned Bittensor in a talk" passage; thin, do not raise. [CT26 ≈1:50]
- On screen: None. Keep the layers slide or the black close on screen during questions. [chart]
- Chart: none
- Open: [JACOB] Q&A prep list: operator-mining question; Bitcast influencer question; "you are describing an AI nobody can control"; collusion above 50%; "why fund it and not others"; slot before or after Pierre Manseau at 14:00.



## Parking lot

- Event: Exploit Conference, Montreal (confirm exact name), 28 September 2026; topic: the state of Bittensor. Jacob's list says 30 min talk + 15 min questions, not 45. [context: event]
- Thesis direction: trace where the Bittensor ideas came from and why; build one cohesive thesis of what Bittensor solves and why the world will want to know. [context: direction]
- The thesis must be backed by modern economic theory on the changing nature of work. [context: economics]
- Order for the talk: make the room feel the wall, then open the door. [Jacob 14 Sep two-worlds session]
- Attack and answer: "unaligned is dangerous" — aligned to whom, and to what (superseded by the fear-of-blame argument on slide 7). [Jacob 14 Sep]
- Story with a face: find one real person by the 28th. [Jacob 14 Sep fiat AI]
- The "there's Bittensor" line in Ilya's NeurIPS Q&A was the audience questioner, not Ilya. [fact-check 16 Sep]
- Five-part outline point 5: other parts; Jacob asks what is missing. [Jacob outline 21 Sep]
- Reconcile the dictated opening with the script's Musk/Const opener when building later versions. [coordinator note 21 Sep]
- How Jacob wants to work: most effort into the thesis, not the slides; the thesis must be gripping, exciting, timely, novel, unique to the mission; prepare him, do not script him; the talk is not slide heavy. [context: how to work]
- Slide design: Bittensor theming, very white and black; reference the theme on bittensor.com. [context: slide design]
- Open questions: exact conference name and venue; slide tool or format (Keynote, Google Slides, PDF, web deck). [context: open questions]
- Jacob's X account: @const_reborn; tweet archive in docs/sources/tweets/. [context: reference voices]
- Order: world first, us second. Sections 2 and 3 are the world; Section 4 is the hinge; Sections 5 and 6 are us. [outline v2]
- Timing from outline v2 (45-min version): hello 0:30; current affairs to 6:30; real risk to 18:30; exploit to 26:30; where we have come to 36:30; where we go to 45:00. Rescale for 30 + 15. [outline v2]
- Three strongest segues: Naval's question (2→3); "You have one choice." then "Now back to that word from the top. Exploit." (3→4); "I believe we still have more time..." (5→6). [outline v2]
- CUT-1 (drop first, ~2:25): Microsoft example; Thiel and separation line; fiat-AI lineage 1695/1999/2009; three eras; GM and Hippius detail; "rights are grants" beat; engineer ask; OTC sentence if not confirmed. [outline v2]
- CUT-2 (drop second, ~1:55): Tilly sentence; Engels' pause; East India Company; "Nobody comes for you... API key"; three-pools sentence; e-gold and Liberty Reserve; break-it-fix-it quote; Bitcoin one-percent sentences; dTAO numbers (last to go); "even Bitcoin wasn't decentralized". [outline v2]
- Exact cut words and positions are marked inline in docs/thesis-v3-script.md. [outline v2]
- The argument order (Weber to Bittensor), one step per paragraph: each step must be true before the next can be said. See docs/trajectory.md. [trajectory]
- Pace check: Jacob's clean rate is about 150 words a minute, above the script's 135; the real risk is going too fast in the memorised close and the two landing lines ("You have one choice." / "Now you have two."). Rehearse those slower. [POT25 lessons]
- Paris 2025 faults to avoid: the close trailed off twice; the title was never fully spoken; unverified superlatives ("a trillion tokens", "growing faster than Google", "cheapest B200s", "better than Google AutoML"); no moral frame. [POT25 lessons]
- Metaphor noted, not for Montreal: "We are like the bosom, we're like the womb for AI that all of humanity can own." [CoinFoin ≈1:20]
- Chart rules: black on white or white on black; one grey (#8e8e8e); no gridlines, axes or legend unless the axis is the point; at most two data labels plus one annotation sentence; one figure per slide; every number sourced and dated in content/data.json. [chart]
- Charts rejected: value-per-employee dot plot (GM 1979 value unverified; mixed metrics); four checks as a table (collapsed to a 2×2); three pillars as a drawing; the 2024 tweet as a screenshot; a Bitcoin mining-pool chart; a revenue-vs-emissions bar (say it aloud on the honest-count slide instead). [chart]
- 27 Sep checklist: re-pull every series in content/data.json; npm run slides, build, deploy; confirm with Jacob the two Opentensor headcounts, the key date, and Cacheon. [chart]
- Reminder: announce the shape in one breath, then keep the promise (Jobs "Just three stories"; Victor's map at minute five). [technique]
- Reminder: concede the weakness early and pair it with a date (Kennedy "we are behind... in this decade we shall make up"; Ilya grading his own 2014 slides). [technique]
- Reminder: remove the option of opting out before you ask for anything (Kennedy "whether we join in it or not"; Ilya "Pre-training will end"; Naval's tweet is Jacob's version). [technique]
- Reminder: make the enemy a pattern or a past self, not a person in the room; nobody names a rival by name (Antonopoulos, Victor, Jobs, Harari "a small elite"). [technique]
- Reminder: teach the pattern three times before applying it to your thing (cars, electricity, internet, then Bitcoin; print, code, money, then mind). [technique]
- Reminder: one metaphor, carried to the end (King's bad check; Harari's mirror); not five metaphors for the same danger. Candidate here: the switch. [technique]
- Reminder: anchor every number in the room (Kennedy's football fields and cigarettes; Ilya reading the log axis aloud). [technique]
- Reminder: split the ask by audience and land on a printable sentence (Harari "If you are an engineer..."; Jobs three times "Stay hungry, stay foolish"). [technique]
- Supporting finding: the most quoted line in each great talk is short, repeated, and followed by a reason within ten seconds; the talks that survive as text used no slides, the two on video used sparse ones. [technique]



## Do not say

- Do not mention the jungle, Peru or ayahuasca origin; Jacob considers it not relevant. [context: origin]
- Do not say Bitcoin's compute out-competes data centers in FLOPs without a defensible framing (hashrate is ASIC hashing, not FLOPs). [context: Bitcoin]
- Do not say "fascism" unless defined on the spot as the economic form (state–corporate fusion); prefer corporatism, state–corporate capture, techno-feudalism, "the merger of state and lab". [context: 13 Sep]
- "Exterminated" is the harsher word; Jacob's own "deprecated" is colder and safer; decide. [Jacob 14 Sep]
- Do not say "the only hope"; say the only exit with no one standing in the doorway. [Jacob 14 Sep]
- Do not say "fully decentralized"; use "decentralized" sparingly. [Jacob 15 Sep]
- Do not say "has trained" or "has built"; say "could be trained", "proves there can be". [Jacob 14 Sep]
- Do not claim Bittensor grants more power to more humans. [Jacob 14 Sep]
- Do not say "Affine on par with Opus 4.6"; base matches Opus 4 / Sonnet 4.5. [fact-check 16 Sep]
- Do not say "best model ever trained outside a lab". [fact-check 16 Sep]
- Do not say Jacob "presented results" at NeurIPS; he gave a sponsor Expo talk. [fact-check 16 Sep]
- Drop "We are the masters of alignment" and "God made the lock, nature made the key." [exploit doc, 20 Sep]
- Do not attribute the "there's Bittensor" line to Ilya. [fact-check 16 Sep]
- Orwell / 1984 line dropped; superseded by the fear-of-blame argument. [Jacob 14 Sep]
- "bioterrorism" needs a source or cut. [coordinator note 21 Sep]
- "masters of incentive" flagged as hubris; "masters of alignment" dropped. [coordinator note 21 Sep]
- "Fedora-hat people" is a jab at rationalists; consider whether to keep. [coordinator note 21 Sep]
- Do not attribute the capital-and-labor check to Weber alone; Weber gave the monopoly-on-violence definition. [coordinator note 21 Sep]
- Do not say "no venture round bought a slice before the public could" (600k TAO OTC to Firstmark, DCG, Polychain at ~$18, 2021–2023). [Jacob 12 Sep]
- Never attribute to Thiel "the state is the only thing that can protect us from the state." [Thiel adapted 15 Sep]
- Take Thiel's mechanism, not the Antichrist theology or the Greta name; take Praxis's three-options structure, not the brand. [context: reference voices]
- Thiel sizes AI as late-90s internet and preaches monopoly; source scale claims to market data; answer monopoly with Bitcoin hashrate. [Thiel adapted 15 Sep]
- Do not say the conference was named for the benchmark; say the coincidence is the right one. [outline v2]
- Never say Microsoft "49%" or cite the board firing "from a tweet"; say thirteen billion dollars and a right to half the profits [VERIFY]. [outline v2]
- Never "trained on Bittensor" for Affine; say post-trained, made better. [outline v2]
- Never "this could never happen here". [outline v2]
- Never "five pools"; say "Three pools mine more than half of Bitcoin's blocks" [VERIFY weekly]. [outline v2]
- Gaza line: draft cuts it [JACOB]. [outline v2]
- Leave out "God made the lock, nature made the key"; Jacob walked back religious framing in Dec 2024 ("Bittensor is powerful tech, but it does NOT make us Gods... Yikes"). [exploit doc]
- Do not say "intelligence maximises energy dissipation" as a law. [exploit doc]
- Do not repeat from Paris 2025: "you can just invest right now", "I'm pitching you", "all the way up to God", "satanic". The Montreal room is political and press will be present. [POT25 lessons]
- Do not quote the "nation state" lines (CoinFoin 2024) on stage; only "power was in check". [CoinFoin note]
- Do not lead with the 450x number; if used, keep "measured in hashes". [POT25 lessons]
- Thiel items unverified before any slide: Nashville dates and venue; whether the 14 Sep clip is from Nashville; Thiel's funding of Yudkowsky's institute; "They promised us flying cars" as his own wording. [Thiel doc]
- Do not say "Chutes serves 120–160 billion tokens a day" (that was mid-2025); do not give any revenue total without attribution. [ecosystem]
- "Two frontier models were switched off for every customer on earth" without adding that the order targeted foreign nationals and the lab chose the all-customer shut-off because it could not verify who was who. [fact-check]
- "She got it back eighteen days later." (Day nineteen.) [fact-check]
- "About three million requests a day." (Two to three million; 2.4 million this week.) [fact-check]
- "Sixty-three percent of renters were agents." (Rentals, not renters.) [fact-check]
- "At about half the maker's price" for Engy today. (Half at launch; about a third less now.) [fact-check]
- "A million dollars burned" as an exact figure. (About a million; 3,738 TAO; block 9032919.) [fact-check]
- "Eighty-six percent of voters like the bill." (The poll came before the bill and asked about an off switch.) [fact-check]
- "Sixty-eight percent support a pause" without "a bill to pause and ban superintelligence" and the pollster's name. [fact-check]
- "Tilly, 1990" for "War made the state." (1975.) [fact-check]
- "Wages doubled after the Black Death." (Rose by half or more.) [fact-check]
- "The largest permissionless training run anyone has attempted" without "tried to train all the way"; never "being trained right now" unless the dashboard shows it live that day. [fact-check]
- "Ten to twenty times larger" for the frontier. (Ten to twenty-five, by open models; closed sizes unpublished.) [fact-check]
- "A few hundred gigabytes" for Bitcoin. (Under a terabyte; 769 GB.) [fact-check]
- "Nine months ago I said no subnet had revenue." (Ten.) [fact-check]
- "About a tenth of what the protocol pays out." (Five to ten percent.) [fact-check]
- "My April tweet... my June tweet." (Essays.) [fact-check]
- "Teutonic-I is the world's most performant decentralized LLM." [fact-check]
- "The Democrats are anti-AI" or any party name. (Say Washington.) [fact-check]
- "$43 million in quarterly revenue" or "$172 million run-rate." [fact-check]
- "Thiel said 'you can still get out of North Korea.'" (He said "could".) [fact-check]
- "Forged Through Fire says workers..." (It says soldiers. Jacob extends it.) [fact-check]
- "A great coup of authority" (Montesquieu). Say "great strokes of authority" and name the translation. [fact-check]
- "Opentensor has four engineers" without naming which entity and when. [fact-check]
- "TAO is a slice of every unit of intelligence" as a property right. (A share of what the protocol pays out.) [fact-check]
- Do not say "27 to 35 billion active" for Affine; do not put an Affine internal score next to an Anthropic score. [Affine fact-check]
- Do not say "Ilya Sutskever mentioned Bittensor" or put the quote on a slide with Ilya's name under it; do not say the NeurIPS talk showed a trained model. [Affine fact-check]
- Do not say: "The Antichrist comes to power by talking about Armageddon nonstop, by talking about existential risk nonstop." [Thiel, Nashville 14 Sep]
- Do not say: "what do you do? You are supposed to surrender to the one-world state." [Thiel on Yudkowsky, Nashville 14 Sep]
- Do not say "It's someone like Greta or Eliezer." Do not say "The threat is much more someone like Greta or her fellow travelers." [thiel-ideas]
- Do not say "I no longer believe that freedom and democracy are compatible." [thiel-ideas]
- Do not say "At its core, artificial intelligence is a military technology." [thiel-ideas]
- Do not say anything with "Antichrist", "Armageddon", "legionnaires", 1 Thessalonians, or katechon. [thiel-ideas]
- Do not say "woke companies", "ESG is a hate factory", "sociopathic grandpa", "finance gerontocracy", or "the hippies took over the country". [thiel-ideas]