# Model keynotes for Jacob's 45 minutes

Thirteen talks, transcribed and taken apart, to find the moves Jacob Steeves should borrow for Exploit Conference, Montreal, 28 September 2026. Full transcripts with timestamps are in `docs/sources/talks/`. All quotes below are verbatim from those transcripts; the timestamp is the time in the source recording named in each transcript's header.

## Executive summary: eight techniques that recur in the best talks

1. **Announce the shape in one breath, then keep the promise.** Jobs: "Just three stories." Victor: four ideas, mapped at minute five, recapped at minute 28. The room relaxes when it knows the map.
2. **Concede the weakness early and pair it with a date.** Kennedy: "we are behind... in this decade we shall make up." Jobs 1997: "drifted away from doing the basics." Ilya grades his own 2014 slides. Every strong talk did this; the weak one did not.
3. **Remove the option of opting out before you ask for anything.** Kennedy: "The exploration of space will go ahead whether we join in it or not." Ilya: "Pre-training will end." Naval's tweet is Jacob's version.
4. **Make the enemy a pattern or a past self, not a person in the room.** Antonopoulos quotes the bankers; Victor blames von Neumann; Jobs blames Apple's drift; Harari says "a small elite." Nobody names a rival by name.
5. **Teach the pattern three times before applying it to your thing.** Antonopoulos: cars, electricity, internet, then Bitcoin. Harari: land, machines, data. The audience predicts the fourth case, so it feels like their conclusion.
6. **One metaphor, carried to the end.** King's bad check. Harari's mirror. Antonopoulos's muddy roads. Not five metaphors for the same danger.
7. **Anchor every number in the room.** Kennedy measures rockets in football fields and the budget in cigarettes. Ilya reads the log axis aloud. Numbers without anchors are forgotten by the next slide.
8. **Split the ask by audience and land on a printable sentence.** Harari: "If you are an engineer..." Jobs: "Stay hungry, stay foolish," three times. Karpathy: "I can't wait to build it with all of you."

Two supporting findings. First, none of the talks that survive as text used slides; the two that survive as video (Victor, Ilya) used sparse ones. Second, the most quoted line in each talk is short, repeated, and followed by a reason within ten seconds.

## The thirteen talks

| # | Speaker, talk, year | Category | Length | Transcript |
|---|---|---|---|---|
| 1 | Steve Jobs, internal "Think Different" meeting, 1997 | (a) founder rallies the faithful | 15 min | `jobs-think-different-internal-1997.md` |
| 2 | Steve Jobs, Stanford commencement, 2005 | (d) classic | 15 min | `jobs-stanford-commencement-2005.md` |
| 3 | John F. Kennedy, Rice University moon speech, 1962 | (d) classic movement | 18 min | `kennedy-rice-moon-speech-1962.md` |
| 4 | Martin Luther King Jr., "I Have a Dream", 1963 | (d) classic movement | 16 min | `king-i-have-a-dream-1963.md` |
| 5 | Balaji Srinivasan, Network State Conference opening and closing, 2023 | (a) crypto founder | 12 + 7 min | `balaji-network-state-conference-opening-2023.md` |
| 6 | Gavin Wood, "Ethereum for Dummies", DEVCON1 2015 | (a) crypto founder to his movement | 18 min + Q&A | `wood-ethereum-for-dummies-devcon1-2015.md` |
| 7 | Andreas Antonopoulos, "Infrastructure Inversion", 2016 | (b) contrarian, crypto | 26 min + Q&A | `antonopoulos-infrastructure-inversion-2016.md` |
| 8 | Bret Victor, "The Future of Programming", 2013 | (b) contrarian reframe | 33 min | `victor-future-of-programming-2013.md` |
| 9 | Yuval Noah Harari, "Why fascism is so tempting", TED 2018 | (b) civic reframe | 14 min + Q&A | `harari-why-fascism-is-so-tempting-ted-2018.md` |
| 10 | Ilya Sutskever, NeurIPS Test of Time talk, 2024 | (c) AI-lab leader | 17 min + Q&A | `sutskever-neurips-test-of-time-2024.md` |
| 11 | Andrej Karpathy, "Software Is Changing (Again)", 2025 | (c) technical founder, one thesis | 39 min | `karpathy-software-is-changing-again-2025.md` |
| 12 | Demis Hassabis, "Artificial Intelligence and the Future", RSA 2016 | (c) AI-lab leader | 45 min | `hassabis-ai-and-the-future-rsa-2016.md` |
| 13 | Peter Thiel, "Competition is for Losers", Stanford 2014 | (a) founder, one contrarian thesis | 42 min + Q&A | `thiel-competition-is-for-losers-2014.md` |

Crypto founders: Balaji Srinivasan and Gavin Wood; Antonopoulos as the crypto movement's best speaker. AI-lab leaders: Sutskever and Hassabis, with Karpathy as the technical-founder voice. Not obtained: Vitalik Buterin's Devcon keynotes (Gavin Wood's DEVCON1 talk stands in), Sam Altman's DevDay keynote, and Chris Dixon (see the sourcing note at the end). Thirteen rather than twelve so that the crypto-founder and AI-lab requirements are both met with real keynotes.

## Talk by talk

Each section follows the same order: what it is, thesis, structure, enemy, turn, proof, ask, devices, slides, weakness, moves for Jacob, mapping to the Bittensor thesis.

### 1. Steve Jobs, internal "Think Different" meeting (Apple, 1997)

**What it is.** A 15-minute talk to Apple employees eight weeks after Jobs returned, with the company near bankruptcy. He explains what he has cut, then unveils the "Crazy Ones" ad. This is the purest founder-rallies-the-faithful talk on record: no press, no customers, just the people who have to do the work.

**One-sentence thesis.** Apple's core value, that people with passion can change the world, has not changed, and we will rebuild the company by saying who we are instead of what our products do.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:17 | Hook | "Good morning. We were up until 3 o'clock last night finishing this advertising. And I want to show it to you in a minute." A promise of a reveal, then he makes them wait nine minutes. |
| 00:32–01:11 | Frame and admission | "We're trying to get back to the basics of great products, great marketing, and great distribution... Apple has pockets of greatness, but in some ways has drifted away from doing the basics really well." Then the cut: "we actually got rid of 70% of the stuff on the product road map." |
| 01:56–03:15 | Proof the team is with him | "I came out of the meeting with people that had just gotten their projects canceled, and they were three feet off the ground with excitement because they finally understood where in the heck we were going." Then inventory: "we're not smart enough to do that. I don't think Einstein's smart enough to do that." |
| 03:15–04:29 | The pivot to values | "To me, marketing is about values. This is a very complicated world. It's a very noisy world... we have to be really clear on what we want them to know about us." Then what not to do: "not to talk about speeds and feeds... not to talk about why we're better than Windows." |
| 04:35–05:27 | Two analogies | Got Milk ("focuses on the absence of the product") and Nike: "Nike sells a commodity. They sell shoes... They honor great athletes, and they honor great athletics. That's who they are." |
| 05:27–06:19 | The question | "Apple spends a fortune on advertising. You'd never know it. You'd never know it." Then: "our customers want to know who is Apple and what is it that we stand for? Where do we fit in this world?" |
| 06:19–07:19 | The core value | "Apple at the core, its core value is that we believe that people with passion can change the world for the better." |
| 07:33–08:04 | What changed, what did not | "A lot of things have changed... But values and core values, those things shouldn't change." |
| 08:04–10:07 | The reveal | Sets it up, then plays the 60-second ad. |
| 10:07–11:00 | Pre-empt the critics | "I know that some people will criticize us for not talking about all those things about why we have better plug and play. But we've got to let people know who Apple is and why it's still relevant in this world." |
| 11:00–15:10 | Logistics as proof | Toy Story premiere, newspapers, billboards, Yoko Ono: "I don't think there is another company on earth that could have done this campaign." |
| 15:10–15:51 | Close | "This company is absolutely going to turn around. As a matter of fact, I think the question now is not, can we turn around Apple? I think that's the booby prize. I think it's, can we make Apple really great again?" |

**How the enemy is framed.** Not Microsoft. He names Windows only to say he will not talk about it (04:29). The real enemy is Apple's own drift: "drifted away from doing the basics" (00:51), "clearly suffered from neglect" (04:13), a product line so confused "I couldn't figure out the damn product line after a few weeks" (01:11). Blaming the past self, not the rival, keeps the room on his side.

**The turn.** 03:15. Everything before is operations. "To me, marketing is about values" changes the register from what the company does to what it believes. The ad at 09:14 is the second turn, from words to feeling.

**The proof.** Before the reveal: engineers "three feet off the ground" (02:03). After: the logistics, the permissions from estates, and Yoko Ono (12:29). The proof that the campaign matters is that people who never endorse anything said yes.

**The ask.** Implicit until the end: "I really deeply appreciate all of the commitment that's in this room... to turning this company around" (15:30). The ask is to keep believing while the products catch up.

**Rhetorical devices, with quotes.**

- Delayed reveal announced in the first sentence and delivered nine minutes later.
- Repetition for emphasis: "You'd never know it. You'd never know it." (05:36)
- Analogy to companies the room admires: Nike, Disney, Coke, Sony (03:48), Got Milk (04:42).
- Antithesis between changing and unchanging things: products, distribution, manufacturing "totally different... But values and core values, those things shouldn't change" (07:40–07:46).
- Reframing the goal upward at the close: "can we turn around Apple? I think that's the booby prize" (15:38–15:46).
- Pre-empting the critique in the critic's own words: "why we have better plug and play" (10:32).

**Slides.** None. Jobs stands and talks. The only visual is the ad itself and a tape of print work.

**Admission of weakness.** In the first 90 seconds and again at 04:13. "Apple has pockets of greatness, but in some ways has drifted." He also admits the limit of the plan: "advertising is not everything" (15:10). Each admission is paired with the fix already under way.

**Moves worth stealing for Jacob.**

1. "Apple at the core, its core value is..." One sentence, said once, slowly. Jacob's version: "Bittensor at the core believes that intelligence should be owned by the people who build it, not the people who own the building." Say it once. Do not explain it.
2. Blame the drift, not the rival. Jobs never attacks Microsoft in this room; he attacks Apple's own neglect. Jacob can spend his critical energy on Bittensor's own gaps (the key, three emission models, emissions versus revenue) and let the labs' concentration speak for itself through the numbers.
3. Reframe the goal upward at the end. Not "can we turn around" but "can we make Apple really great again." Jacob: "The question is not whether Bittensor survives the year. It is whether, in 2030, the most valuable thing on earth has an owner list that looks like Bitcoin's or like NVIDIA's."

**Mapping to the Bittensor thesis.** This is the talk for the holders and builders in the room, the internal audience. Jobs' structure, admit the drift, cut the noise, restate the value, show the proof, pre-empt the critics, reframe the goal, maps directly onto sections 11 to 13 of the thesis. The "speeds and feeds" warning is also a warning for the subnet section: do not list 128 subnets and their revenue; say what they are for.

### 2. Steve Jobs, Stanford commencement (2005)

**What it is.** A 15-minute commencement address with no slides, no product, no company pitch. It is the most-watched speech of its kind because the structure is so simple that anyone can copy it.

**One-sentence thesis.** Your time is short, so trust your own path and do the work you love.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:13 | Hook | Humility and a joke: "I never graduated from college and this is the closest I've ever gotten to a college graduation." |
| 00:45 | Contract | He tells the room the shape of the talk: "Today, I want to tell you three stories from my life. That's it. No big deal. Just three stories." |
| 00:55–05:31 | Story 1, connecting the dots | Adoption, dropping out, the calligraphy class, the Mac. The story ends in a principle: "you can't connect the dots looking forward. You can only connect them looking backwards." |
| 05:38–08:51 | Story 2, love and loss | Apple, getting fired at 30, NeXT, Pixar, the return. Ends in the ask: "If you haven't found it yet, keep looking and don't settle." |
| 09:04–12:55 | Story 3, death | The cancer diagnosis, told in clinical detail. Ends in the principle: "have the courage to follow your heart and intuition." |
| 13:09–14:31 | Close | The Whole Earth Catalog, the back-cover photo, "Stay hungry, stay foolish," repeated three times. |

**How the enemy is framed.** There is no external enemy. The enemy is internal: "dogma, which is living with the results of other people's thinking" (12:37) and "the noise of others' opinions" (12:42). This lets the talk feel generous rather than combative.

**The turn.** Each story has its own reversal. The strongest is at 07:14: "it turned out that getting fired from Apple was the best thing that could have ever happened to me. The heaviness of being successful was replaced by the lightness of being a beginner again."

**The proof.** Concrete, checkable detail instead of adjectives: "I returned Coke bottles for the five cent deposits to buy food with" (03:10); "a $2 billion company with over 4,000 employees" (05:56); "I had a scan at 7.30 in the morning" (10:13).

**The ask.** Small and personal, stated once per story: "keep looking and don't settle" (08:38); "have the courage to follow your heart" (12:46).

**Rhetorical devices, with quotes.**

- Rule of three, announced up front (00:45) and then delivered.
- Antithesis at the end of each story: "It was awful tasting medicine, but I guess the patient needed it" (08:07).
- Aphorism as landing pad: "death is very likely the single best invention of life. It's life's change agent. It clears out the old to make way for the new. Right now, the new is you." (12:05–12:19)
- Callback: the closing line is borrowed, credited, and repeated: "Stay hungry, stay foolish" (14:09, 14:16, 14:24).
- Self-deprecation to earn the right to be grave: "Sorry to be so dramatic, but it's quite true." (12:27)

**Slides.** None.

**Admission of weakness.** The whole second story is an admission: "I was a very public failure and I even thought about running away from the valley" (06:54). He names the people he apologised to (06:47). The weakness is told as the cause of the later strength, not as an aside.

**Moves worth stealing for Jacob.**

1. Announce the shape in one breath and keep the promise. "Three stories" is the whole map. Jacob's version: "Three things tonight: what is happening, what the real fear is, and what we built." The audience relaxes when it knows the shape.
2. Tell the failure as the origin of the strength. Jobs makes being fired the reason Pixar and NeXT exist. Jacob can make the sudo key and the Templar dispute the reason the dated decentralisation plan exists: "We chose speed. That was right. It is now wrong. Here are the dates."
3. Land every section on one sentence you could print on a wall. Jobs does it three times. Jacob has candidates already: "Do not fear the machine. Fear the machine owned by a few." Use one per section, not ten.

**Mapping to the Bittensor thesis.** The talk proves that a founder does not need slides or numbers to move a room, only true stories with a clear point. Jacob's origin material is off-limits (the jungle), but he has three true stories he can use: reading the Bitcoin paper and seeing a commodity market with no boss; the first miner who was not him; the day the chain was halted in 35 minutes.

### 3. John F. Kennedy, Rice University "We choose to go to the moon" (1962)

**What it is.** An 18-minute outdoor speech to 40,000 people in a football stadium in the Houston heat. It is the model for asking a large group to fund something expensive, uncertain, and slow, and to feel good about it.

**One-sentence thesis.** Space will be explored whether or not we join, so we will lead it, and we choose the hardest goal because hard goals organise a people.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:02 | Hook | Formal greeting, then a joke about the honorary professorship: "my first lecture will be very brief." |
| 00:36 | Frame | Triads: "a college noted for knowledge, in a city noted for progress, in a state noted for strength... an hour of change and challenge, in a decade of hope and fear, in an age of both knowledge and ignorance." |
| 01:49–03:16 | Compression device | All of human history compressed into 50 years: "Christianity began less than two years ago. The printing press came this year... Only last week did we develop penicillin and television and nuclear power." |
| 03:18–04:13 | Name the doubters | "some would have us stay where we are a little longer to rest, to wait. But this city of Houston... was not built by those who waited." |
| 04:46–06:15 | Inevitability | "The exploration of space will go ahead whether we join in it or not." Then the vow: "we intend to be first." |
| 06:50–08:21 | Values | "space science, like nuclear science and all technology, has no conscience of its own. Whether it will become a force for good or ill depends on man." |
| 08:23–09:13 | The turn and the line | "But why, some say, the moon?... We choose to go to the moon... not because they are easy, but because they are hard." |
| 09:19–12:00 | Proof | Rocket numbers, satellites, then the admission: "we are behind, and will be behind for some time in manned flight. But we do not intend to stay behind." |
| 12:09–14:49 | What is in it for you | Jobs, companies, the Houston centre, the budget. "$5,400 million a year, a staggering sum, though somewhat less than we pay for cigarettes and cigars every year." |
| 14:58–16:04 | The single impossible sentence | One 65-second sentence listing everything the rocket must do, ending: "and do it right, and do it first before this decade is out, then we must be bold." |
| 16:23–17:38 | Close | "it will be done before the end of this decade." Then Mallory: "because it is there. Well, space is there, and we're going to climb it." |

**How the enemy is framed.** Two enemies, one soft, one hard. The soft enemy is fatigue: those who would "rest" and "wait" (03:43). The hard enemy is never named as an enemy; it is a "hostile flag of conquest" (05:48) and "weapons of mass destruction" (06:00). He then generalises the stakes so that the mission becomes everyone's: "Its conquest deserves the best of all mankind" (08:13).

**The turn.** At 08:23 he voices the objection in the audience's words ("why, some say, the moon?"), answers with three more rhetorical questions, and then delivers the thesis line three times before completing it.

**The proof.** Numbers with a physical anchor in the room: a building "as long as two lengths of this field" (10:32); a Venus shot as accurate as "dropping it in this stadium between the 40-yard lines" (11:14); heat "almost as hot as it is here today" (15:56, which got the laugh).

**The ask.** Money and patience. He says the cost plainly (13:53–14:36), then reframes it per person per week, then admits "this is in some measure an act of faith and vision, for we do not now know what benefits await us" (14:43).

**Rhetorical devices, with quotes.**

- Anaphora: "We choose to go to the moon. We choose to go to the moon. We choose to go to the moon in this decade" (08:39–08:50).
- Compression of time to make the present feel like a hinge (01:49–03:12).
- Objection voiced then answered with questions: "why climb the highest mountain? Why, 35 years ago, fly the Atlantic? Why does Rice play Texas?" (08:29–08:37).
- Antithesis: "not because they are easy, but because they are hard" (08:55).
- Concession before claim: "We have had our failures, but so have others, even if they do not admit them" (11:35).

**Slides.** None.

**Admission of weakness.** Direct and early in the proof section: "To be sure, we are behind, and will be behind for some time in manned flight. But we do not intend to stay behind, and in this decade we shall make up and move ahead" (11:42–12:00). He also admits the benefits are unknown (14:49). Both admissions are followed within one breath by a dated promise.

**Moves worth stealing for Jacob.**

1. "Whether we join in it or not." Kennedy removes the option of opting out before he asks for anything. Jacob's equivalent is Naval's line: AI is not slowing down; the only decision is who owns it. Say it early, then everything after is a choice between owners, not a choice about AI.
2. Admit you are behind, then give the date. "We are behind... in this decade we shall make up." Jacob: "Our market cap is a rounding error against NVIDIA. The key is still in my hand. December 2027 it is gone."
3. Anchor the big number in the room. Kennedy measures the rocket in football fields and the budget in cigarettes. Jacob can measure Bitcoin's power draw in Polands and Anthropic's headcount in GM factories, but he should pick one anchor per number and say it in Montreal terms where he can.

**Mapping to the Bittensor thesis.** Kennedy is asking people to fund a decade-long bet with no guaranteed payoff, against a rival power. That is exactly the holder's ask. The "act of faith and vision" line gives Jacob permission to say the honest version: emissions are ~$300M a year against lab capex in the hundreds of billions, and we do not know yet what the network will produce, and we choose to do it anyway.

### 4. Martin Luther King Jr., "I Have a Dream" (1963)

**What it is.** A 16-minute speech to 250,000 people at the Lincoln Memorial. Studied here for one reason: it is the cleanest example of how a movement speech moves from grievance to vision to instruction without losing the crowd. The first half was scripted; the "I have a dream" section was improvised when a singer behind him called out "tell them about the dream."

**One-sentence thesis.** America wrote a promise it has not paid, we are here to collect it now, and we will collect it without hatred.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:00 | Hook | Places the moment in history at once: "what will go down in history as the greatest demonstration for freedom in the history of our nation." |
| 00:25–01:58 | The broken promise | "Five score years ago" (echoing Lincoln). Then the refrain "one hundred years later" four times, each with a new image: "manacles of segregation", "a lonely island of poverty in the midst of a vast ocean of material prosperity", "in exile in his own land." |
| 02:08–03:47 | The metaphor that carries the talk | The bad check: "America has given the Negro people a bad check, a check which has come back marked insufficient funds... So we've come to cash this check." |
| 03:49–05:59 | Urgency | "the fierce urgency of now"; "the tranquilizing drug of gradualism"; "Now is the time" four times. |
| 06:05–07:41 | Discipline: the ask turned inward | "we must not be guilty of wrongful deeds... We must forever conduct our struggle on the high plane of dignity and discipline." Then the bridge to allies: "We cannot walk alone." |
| 07:50–09:23 | The objection answered | "when will you be satisfied?" Answered with "We can never be satisfied" five times, ending in Amos: "until justice rolls down like waters and righteousness like a mighty stream." |
| 09:26–10:39 | Honour the audience's cost, then send them back | "Some of you have come fresh from narrow jail cells" (09:49)... "Go back to Mississippi. Go back to Alabama." |
| 10:42–13:30 | The vision | "I still have a dream." Eight "I have a dream" images, from the abstract (the creed) to the personal (his four children) to the specific (Alabama). Then Isaiah. |
| 13:30–14:08 | Faith as method | "With this faith, we will be able to..." four times: work, pray, struggle, go to jail together. |
| 14:08–end | Close | "Let freedom ring" from named mountains, north then south, then "free at last." |

**How the enemy is framed.** The enemy is a condition, not a person. He names states and a governor (12:31–12:52) but the villain of the speech is the "bad check", segregation, and "gradualism." This lets him make the ask to allies ("many of our white brothers as evidenced by their presence here today," 07:18, and "We cannot walk alone," 07:41) without contradiction.

**The turn.** At 10:42. Everything before is grievance and instruction. "even though we face the difficulties of today and tomorrow, I still have a dream" (10:42–11:02). The word "still" is the hinge.

**The proof.** None in the numeric sense. The proof is the crowd itself (00:00), the named suffering (09:49–10:07), and scripture, which the audience already accepts.

**The ask.** Three asks, in order: keep discipline (06:22–07:03); do not stop (07:50); go home and keep working (10:17–10:31).

**Rhetorical devices, with quotes.**

- Anaphora as the engine of the whole speech: "one hundred years later", "Now is the time", "We can never be satisfied", "I have a dream", "With this faith", "Let freedom ring."
- One extended metaphor carried to its end (the check, 02:08–03:47).
- Borrowed authority: Lincoln, the Declaration, Amos, Isaiah, *My Country, 'Tis of Thee*. The audience finishes the sentences with him.
- Geography as inclusion: naming states and mountains so every listener hears their own home.
- Antithesis: "from the dark and desolate valley of segregation to the sunlit path of racial justice" (04:32).

**Slides.** None.

**Admission of weakness.** Not of his own, but of the movement's risk: "we must not be guilty of wrongful deeds. Let us not seek to satisfy our thirst for freedom by drinking from the cup of bitterness and hatred" (06:22–06:38). He names the temptation his own side faces, which is a form of weakness-admission that makes the speaker trustworthy to outsiders.

**Moves worth stealing for Jacob.**

1. One metaphor, carried all the way. King does not offer five metaphors for injustice. He offers the bad check and cashes it. Jacob has "the machine that pays you back" and Elysium. Pick one and return to it three times; do not stack Elysium, Gaza, feudalism, and fascism in one talk.
2. Answer "when will you be satisfied?" for your own community. Holders ask "when is it decentralised?" Builders ask "when do the rules stop moving?" Voice the question in their words and answer with a refrain: "Not until the key is gone. Not until the rules are frozen. Not until..."
3. Make the ask to your own side first. King tells his own people to keep discipline before he asks anything of America. Jacob can tell the Bittensor community what it must not do (chase emissions, dump on newcomers, fight over root) before he asks newcomers to join.

**Mapping to the Bittensor thesis.** King's audience had already paid a price to be in the room. Jacob's has too: holders through drawdowns, builders through three emission models in sixteen months. Naming that cost (09:26–10:07) before the vision is what makes the vision land.

### 5. Balaji Srinivasan, Network State Conference opening and closing remarks (Amsterdam, 2023)

**What it is.** A 12-minute opening (Balaji speaks from 01:17) and a 7-minute closing to a room of founders, investors and crypto people at his own conference. Included because it is a crypto founder addressing his own movement, and because it shows both a strong framing device and several habits to avoid.

**One-sentence thesis.** If you can start new companies, communities and currencies, you can start new countries, and the way to do it is a thousand startups that each replace one legacy institution.

**Structure, minute by minute (opening).**

| Time | Move | What happens |
|---|---|---|
| 00:03–01:16 | Warm-up by an MC | Call-and-response, "give a thunderous round of applause." |
| 01:17–02:01 | Housekeeping | QR code for proof of attendance; "tell them what you're going to tell them, then tell them, then tell them what you told them." |
| 02:02–02:37 | The thesis in three lines | "We started new companies like Google. We started new communities like Facebook. We started new currencies like Bitcoin and Ethereum. Can we start new countries?" Then the book plug. |
| 02:43–04:02 | The analogy | "If Bitcoin's a decentralized currency, the network state is a decentralized country." The growth ladder from one person to "1,729,314 people, it's a cloud country somewhere between Latvia and Bahrain." |
| 04:12–05:42 | The new concept | "the parallel establishment. So imagine 1,000 different startups, each of them replacing a different legacy institution." Satoshi (exit) versus Satya Nadella (reform) versus Steve Ballmer (deny; the transcript mis-hears the surname). |
| 05:44–07:39 | Two halves | Parallel societies (physical) and parallel institutions (media, education, finance, law, science). |
| 07:39–11:45 | Agenda read-through | Every speaker of the day, in order, with one line each. |

**Closing remarks (separate video, timestamps restart).** Seven minutes: a head count ("we have more than 1,200 people," 00:27), "So the movement is absolutely growing" (00:36), an exit poll by QR code (01:09–03:46), a book plug (03:47–05:09), and a group selfie for "the network state movie" (05:19–07:01). The one idea in it is the reason for the poll: "what we're doing is inherently political" (03:17) and he wants to find "where the center of mass of the community is" (03:23).

**How the enemy is framed.** As legacy institutions to be replaced, not attacked ("each of them replacing a different legacy institution," 04:23): San Francisco, Harvard, the New York Times, Wall Street (04:26–07:07). The enemy is a list, which keeps the tone builder-like rather than angry. The one sharp moment is the Ballmer joke about denial (04:49).

**The turn.** There is not one. The talk moves from theory to practice at 04:05 ("But how does it work in practice?") but never changes emotional register.

**The proof.** Names and counts: a dozen physical startup societies, Replit's 20 million developers, the Thiel Fellowship. Proof by roster.

**The ask.** Implicit: stay for the day, meet the builders. No explicit call to act.

**Rhetorical devices, with quotes.**

- Triad that ends in a question: "new companies like Google... new communities like Facebook... new currencies like Bitcoin and Ethereum. Can we start new countries?" (02:19–02:25). This is the strongest thing in the talk.
- Analogy to a known success: "If Bitcoin's a decentralized currency, the network state is a decentralized country" (02:45–02:48).
- A coined term introduced with "I'm going to introduce a new concept today" (04:15).
- A three-way character test: Satoshi, Satya, Steve (04:41–04:50).
- Orders-of-magnitude ladder: 17, 172, 17,000, 1.7 million (03:13–03:38).

**Slides.** Dense. The ladder, the parallel-establishment grid, speaker headshots. The slides carry information the speech repeats.

**Admission of weakness.** None. The closest is "Nothing here violates any laws of physics" (03:52), which is a claim, not a concession.

**Moves worth stealing for Jacob.**

1. The triad-into-question. Jacob's version writes itself: "Bitcoin showed a network can make money without a bank. Ethereum showed a network can run contracts without a court. Can a network make intelligence without a lab?" That is the whole talk in three sentences.
2. Name the third path with a character test. Balaji's Satoshi-Satya-Steve is a way to make "exit" respectable without insulting reformers. Jacob's version: those who want the labs regulated (reform), those who want them broken up (fight), and those who build a network outside both (exit). Say all three are honest, then say which one he is.
3. Coin one term and define it in one sentence. "Parallel establishment" is defined in one line and then used as a lens for the whole event. Jacob has "permissionless basic capital" and "the machine that pays you back." Pick one, define it in one sentence, use it as the lens.

**What not to copy (see Pitfalls).** Opening with a QR code; reading the agenda for four minutes; plugging the book in the first two minutes; slides that duplicate the speech. The strongest three sentences in the talk are buried at minute two after housekeeping.

**Mapping to the Bittensor thesis.** Balaji's "parallel establishment" grid is a ready frame for the subnet section. Jacob's five layers (compute, inference, data, knowledge, free intelligence) are parallel institutions: a parallel cloud, a parallel API, a parallel dataset, a parallel model. The difference Jacob can claim is that Bittensor's parallel institutions pay the people who build them, in ownership, from day one.

### 6. Gavin Wood, "Ethereum for Dummies" (DEVCON1, London, 2015)

**What it is.** A 17-minute talk plus 6 minutes of questions by Ethereum's co-founder and first CTO, four months after the network launched, to the first Devcon. It is a founder explaining to his own movement what they have actually built, in plain words, after the whitepaper explained how. It stands in here for the Vitalik Devcon keynotes that could not be obtained.

**One-sentence thesis.** Ethereum is not a better Bitcoin; it is the world's first decentralised computer, one machine with no off switch, no owner, and no root, and everything that follows comes from taking that literally.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:19 | Hook | The title was a joke; the real question is harder: "Ethereum for dummies wasn't my original title." Then the confession that his own understanding "has changed somewhat." |
| 00:38–02:39 | The wrong first answer | "Ethereum was always kind of sold as this notion of Bitcoin thing, but Bitcoin with maybe some features removed." Then how, why and what: "we can answer the how... we got a pretty good idea of why... But the what, like what actually is it? That's the harder question." |
| 02:39–03:00 | The rule for the talk | "I'm going to try and avoid using these technical terms, because these technical terms help us describe how it works, but they don't really help us describe what it is." |
| 03:12–04:15 | The answer, then the weakness | "Ethereum's a computer. It's a computer a bit like this one." A 1950s picture. "Ethereum is a computer. It's not a very good computer. It's really slow. It's also really, really expensive." |
| 04:15–10:00 | Properties, one at a time | Global singleton; "there's no reset button, there's no power button, you can't turn it off"; ubiquitous and fair access; multi-user; object-oriented; auditable; atomicity; provenance ("we can check to see who sent the message"); permanence ("there's no system admin who can come along and actually sort of delete the object"); immutable code. |
| 10:06–12:35 | What the properties buy | "Ethereum is a commons for innovation." Servers as the enemy: "Servers provide barriers... these barriers basically make monopolies. And monopolies are generally bad for innovation." Then: "Ethereum is the world's first decentralized computer." |
| 12:44–15:41 | The turn: a law of history | Dispersed, then a strong player who "add[s] rules and they rule," then decentralisation once the rules hold. Coders, computing (mainframe, client-server, peer-to-peer), governments (anarchy, king, democracy), commerce. "We begin with nothing. We centralize in order to enforce these initial rules, and then we can decentralize once that structure is in place. We decentralize because it's more efficient, it's more scalable, and it's more resilient to attack." |
| 15:47–17:50 | Naming it | "Ethereum commoditizes trust." "the court of the internet." "a decentralized legal system native to the internet." Then "Web 3... the zero trust computing part." |
| 17:50–23:40 | Q&A | On people versus tools; on who resists ("people who at the moment have a relatively easy existence"); on private chains starving the public chain. |

**How the enemy is framed.** Structural, never personal: "Servers provide barriers. Servers are cumbersome... these barriers basically make monopolies" (11:08–11:17). Also the "false security of placing it on a third party server where only the elite hackers and those who want to pay them can get access to it" (11:44–11:51). In the Q&A the enemy becomes people, but generically: "there are people who stand to benefit a lot, and there are people who stand to benefit much, much less" (20:19–20:31).

**The turn.** 12:44–15:41. Until then the talk is a feature list. The history-of-centralisation pattern turns the feature list into an argument about where the world is going, and the argument is delivered as a general law with four examples before Ethereum is named.

**The proof.** Properties the audience can verify by using the thing, and one picture of a 1950s "home computer" for the laugh. No numbers.

**The ask.** In the Q&A, to the room: "and inclusive is probably a reasonably good way to set about it" (19:57). And on the public chain: "make sure that all of the technology that goes into private chains is free and open... make sure that resources are shared" (22:18–22:25).

**Rhetorical devices, with quotes.**

- The self-imposed language rule, announced (02:46–02:54) and kept.
- Deflation before inflation: name the object plainly, list its faults, then reveal its strange powers. "So, so far it's not seeming like such a good invention. Seems like we're going backwards a little bit. But actually it's got some really interesting properties." (04:01–04:15)
- Negatives as features: no reset button, no power button, no root, no off. "There is no off for this." (09:12)
- Pattern across four domains before the application (12:44–15:41).
- Coined labels, each one sentence: "a commons for innovation" (10:18), "commoditizes trust" (15:47), "the court of the internet" (16:21), "zero trust computing" (16:15).

**Slides.** A handful, sparse: the comedy still, the 1950s computer, a hand-drawn "artist's impression" of the Ethereum computer, and the history-pattern diagram.

**Admission of weakness.** Immediate and specific: "It's not a very good computer. It's really slow. It's also really, really expensive" (03:29–03:41). Also on his own understanding: "for a long period, we just didn't know" (03:02). In the Q&A, on outcomes: "it's not especially clear what that what that system is gonna be yet" (18:51–18:58).

**Moves worth stealing for Jacob.**

1. Say what it is, not how it works, and ban your own jargon. Wood refuses the conference's vocabulary for 17 minutes. Jacob's version: no "Yuma Consensus," "dTAO," "alpha," "root," or "emissions" until the thing has been named in plain words. "Bittensor is a market. It pays anyone who produces intelligence, and nobody runs it."
2. List the flaws before the powers. "It's really slow. It's also really, really expensive." Then the strange properties. Jacob: "It is small. It is not finished. I hold a key." Then: nobody can stop you mining, nobody was allocated anything, the schedule cannot be changed.
3. The centralise-then-decentralise law. Wood's pattern is exactly the frame Jacob needs for the sudo key: every system centralises to set its rules and decentralises once they hold, "because it's more efficient, it's more scalable, and it's more resilient to attack." That turns "not decentralised yet" from an apology into a stage in a known sequence, with a date.

**Mapping to the Bittensor thesis.** The DEVCON1 Q&A asked in 2015 the question Jacob's tweet asked in 2026: how do we make sure the public chain "doesn't get starved" (22:00) while private, funded alternatives flourish. Wood's answer was open code and shared resources. Bittensor's answer is a protocol that pays the public network directly. Jacob can quote the 2015 question, note that it is still open for open AI, and say that this is the thing his network is built to answer. His line that server barriers "basically make monopolies" (11:15) is also a one-line version of the firm-without-a-payroll argument: the lab is the server.

### 7. Andreas M. Antonopoulos, "Bitcoin and the Coming Infrastructure Inversion" (Zurich, 2016)

**What it is.** A 26-minute talk with no slides to a Bitcoin meetup in Zurich, followed by 36 minutes of questions. It is the best-known example of a crypto speaker turning the standard objection (it is slow, it is clunky, it does not work) into proof that the thing is early. It later became a chapter of *The Internet of Money*.

**One-sentence thesis.** Every disruptive technology is first forced to run on the infrastructure of the thing it replaces, looks bad doing it, and then flips, so Bitcoin's awkwardness today is the signature of a coming inversion in finance.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:00 | Hook | The concept, named and defined in 20 seconds: "how things change when infrastructure that is new is laid on top of infrastructure that is old and how that creates a conflict." |
| 00:24–01:37 | The objection, voiced in the critic's voice | "we're trying to somehow squeeze it on top of the existing banking system. And the result is messy... an opportunity for those who support the traditional banking system to go, See? See? It's not working. It's slow." |
| 01:42–03:30 | The pattern, stated | "When you read about it, 20, 30, 40 years in the future, it is all very smooth... hindsight provides clarity." Then the rule: "the disruptive technology has to live in a world created for the technology it is replacing." |
| 03:36–07:18 | Example 1, cars on horse roads | Mud, ruts, no gas stations, "You are in horse society, and you are the crazy one driving one of these vehicles." Then the flip: paved roads, and "Horses are very comfortable on asphalt." Ends: "That is an infrastructure inversion." |
| 07:56–12:04 | Example 2, electricity versus gas | "electricity is a fad"; newspapers "wrote about every house that burned down"; gas pipes cannot carry electricity; the chicken-and-egg objection. |
| 12:51–21:07 | Example 3, the internet over phone lines | The modem as a car on a dirt road; phone companies cutting off modem calls "because we're competing against the phone company. Kind of like banks shutting down..."; the reversal, all voice now runs over data; the comfort-noise story. |
| 21:12–22:30 | Bitcoin placed in the pattern | "we're riding the automobile on the muddy roads of banking. The Bitcoin supercar, the Formula One of finance, is riding along on the muddy roads of the 1970s, mainframe-based banking." The banks' objections, in their words. |
| 22:35–23:42 | The prediction | "over the next 15 to 20 years a great infrastructure inversion... the banks will resist, then the banks will adopt, then the banks will run their systems alongside... Finally, they will run all of traditional banking as an application on top of a decentralized, trusted ledger." |
| 23:42–25:31 | The joke that proves the point | "I can create an application that takes your Bitcoin transaction and makes it clear in three to five business days for a cost of $5. And I've implemented traditional banking. It's kind of like the comfort-noise generation." |
| 25:42–26:21 | Close | "Enabling the future on your legacy system is very difficult... until you flip the infrastructure, and then simulating the past on the network of the future becomes extremely easy. What we are part of now is the very early stages... the first stages of the greatest infrastructure inversion the world has ever seen." |

**How the enemy is framed.** The enemy is a recurring type, "those invested in the status quo" (10:29), shown four times in history so that the banks are just the fourth instance. He quotes them rather than describing them: "That's crazy. Those noisy, disgusting machines that are probably going to kill us all will never work" (02:17–02:20). The audience laughs at the horse people and then recognises the bankers.

**The turn.** There are four small turns, one per example, each marked with the same phrase: "That is an infrastructure inversion" (07:18) and its variants. The big turn is at 21:12, "Then we get to Bitcoin," when the pattern the audience has just learned three times is applied to the thing they came to hear about.

**The proof.** History told as story, with sensory detail: mud and "horse poo" (04:44), the modem handshake sung out loud (15:36–16:23), the click of a cut-off long-distance call (17:09). No numbers at all in the talk itself.

**The ask.** None stated. The implied ask is to stop apologising and keep building: "The ones who succeed are the ones who continue, even though the rest of society tells them they are crazy" (02:53).

**Rhetorical devices, with quotes.**

- Voicing the critic: "See? See? It's not working. It's slow. It doesn't work so well." (01:10)
- The same sentence pattern applied to four technologies so the audience predicts the fourth: "crazy pioneers who were made fun of by everybody else in society persisted until everybody could see what they were doing was correct" (03:10).
- Generic versus specialised as the underlying law: "one is extremely specialized. It has already chosen the application for you... The other one is very generic. Data means anything" (19:13–19:21).
- Reductio by demonstration: implementing a bank in one sentence (23:50–24:02).
- Escalating callback in the close: horse carriage, gas lighting, comfort noise, all three examples return in three sentences (24:59–25:31).
- Local colour: "If you want to do a nice tour of Zurich on horseback, I'm sure. Perfectly comfortable." (06:30)

**Slides.** None. He walks and talks for 26 minutes. This is a deliberate part of his style and it is why the talk survives as a book chapter.

**Admission of weakness.** He concedes the objection fully before answering it: Bitcoin on today's rails is "messy" (00:58), "a bumpy road" (21:58), and the chicken-and-egg problem is real (22:19–22:25). He never says the critics are wrong about the present, only about what the present means.

**Moves worth stealing for Jacob.**

1. Reframe every objection as a symptom of being early. Antonopoulos never denies Bitcoin is slow. He explains why the slowness is the expected shape of the moment. Jacob's objections are known: small market cap, emissions larger than revenue, a founder key, three emission models. Each one can be placed in a pattern: "This is what a market for intelligence looks like while it still runs on the labs' roads."
2. Teach the pattern three times before you apply it. Three historical examples, then Bitcoin. Jacob has Bitcoin itself (a joke in 2011, Poland's power draw now), plus open source software (a hobby, then the entire server market), plus perhaps the internet versus telcos. Then Bittensor.
3. Predict the sequence, not the date. "Resist, adopt, run alongside, run on top of." Jacob's version for the labs: ignore, dismiss, buy compute from it, publish weights to it. He can point out that step three has already begun (Chutes on OpenRouter, ETF filings) without promising when step four arrives.

**Mapping to the Bittensor thesis.** This talk is the model for the "honest state" and "what has been built" sections. It also fixes a tone problem: Jacob's notes are angry at the labs, and anger reads as insecurity. Antonopoulos is amused by the bankers. His confidence comes from the pattern, not from attacking anyone. The "comfort noise" story is a gift for the subnet section: a centralised AI API is comfort noise, a familiar interface laid over a network that can already do more.

### 8. Bret Victor, "The Future of Programming" (DBX, 2013)

**What it is.** A 33-minute conference talk performed as if it were 1973. Victor wears a pocket protector, uses an overhead projector and hand-placed transparencies, and describes the "recent" research of the 1960s as the obvious future. The audience knows the 40-year punchline: almost none of it was adopted. It is the best example of a contrarian talk that never says "you are wrong" out loud.

**One-sentence thesis.** The most dangerous thought a creative person can have is that they know what they are doing, because it makes them blind to better ways.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:06 | Hook | Straight-faced anachronism: "I'm really excited to be talking to so many programmers of automatic computing machines... There's now literally thousands of computers around the world." The room laughs; the frame is set in 30 seconds. |
| 01:33–05:20 | The real thesis, planted early | "technology changes quickly, people's minds change slowly." The binary coders who rejected assembly; von Neumann "furious" at a student who wrote an assembler. |
| 05:26–06:37 | The map | Four ideas, one sentence each: direct manipulation, goals not procedures, spatial representation, parallel models. |
| 06:40–09:38 | Idea 1 | Sketchpad (1962). Prediction: "There won't be any markup languages or style sheet languages, right? That would make no sense." |
| 09:41–16:20 | Idea 2 | Planner, Prolog, regular expressions, Licklider's "communicating with aliens." Prediction: "we're not going to have APIs in the future." |
| 16:29–21:52 | Idea 3 | Engelbart's NLS, GRAIL, Smalltalk. Prediction: "in 40 years we won't be writing code in text files." |
| 21:57–27:39 | Idea 4 | The actor model. "if in 40 years we're still using threads and locks, we should just like pack up and go home." |
| 27:48–28:39 | Recap | The four ideas restated in one breath each. |
| 28:44–32:47 | The turn and the close | Drops the act by degrees. "The real tragedy would be if people forgot that you could have new ideas about programming models in the first place." Then the message: "I don't know what I'm doing." |

**How the enemy is framed.** Never as a person in the room. The enemy is dogma, illustrated by people from history who could not see the next thing: "Assembly was shown to these guys, the guys coding in binary, and they just weren't interested at all. They just didn't get it" (03:29–03:35). Because the villains are historical, nobody in the audience is accused, yet everyone recognises themselves.

**The turn.** At 29:22 the joke stops being a joke. "But even more of a tragedy than these ideas not being used would be if these ideas were forgotten... But even that's not the biggest tragedy. That's not the real tragedy. The real tragedy would be if people forgot that you could have new ideas about programming models in the first place." Three escalating "tragedy" clauses.

**The proof.** Demonstrations, not claims. Each idea is shown as a working system with a date: Sketchpad 1962, NLS 1968, GRAIL 1968, Smalltalk, the Intel microprocessor. The proof is that the future already existed.

**The ask.** One sentence, to yourself: "I think the first step is you have to say to yourself, I don't know what I'm doing. We as a field don't know what we're doing" (32:22–32:27).

**Rhetorical devices, with quotes.**

- Sustained dramatic irony. Every prediction is stated as obvious, and every one is known by the room to have failed: "I'm pretty confident this is never going to happen, we're not going to have APIs in the future" (16:08–16:10).
- Rule of four, mapped at 05:26, delivered, recapped at 27:48.
- Historical villains as mirrors: von Neumann (03:51–04:23), the binary programmers (03:29), the assembly programmers (04:42–04:48).
- Escalation by negation: "But even that's not the biggest tragedy. That's not the real tragedy" (29:36–29:38).
- Freedom as the payoff: "once you truly believe that, then you're free. And you can think anything" (32:37–32:44).

**Slides.** About 40 overhead transparencies, one image or one line each, placed by hand. The physical projector is part of the argument: the medium is 1973, so the ideas feel old, which is the point.

**Admission of weakness.** The whole close is one: the speaker says the field, including himself, does not know what programming is. "We don't know what programming is. We don't know what computing is. We don't even know what a computer is" (32:31–32:35). He also concedes the predictions are uncertain: "you can't really predict the future, right? So these are some good ideas. I don't know what's going to happen to them" (28:47–28:53).

**Moves worth stealing for Jacob.**

1. Let the audience discover the point before you say it. Victor never says "the industry ignored these ideas." He shows 1962 and lets the room compute 2013. Jacob can show Bitcoin in 2011 (a joke, a toy, "for criminals") next to Bitcoin in 2026 (Poland's power draw, ETFs) and let the room compute where Bittensor is on that curve. Do not say "we are early." Show it.
2. Make the enemy historical so nobody in the room is accused. Jacob's room contains people who work at labs, people who hold TAO, and press. "The people who could not see assembly" is a safer villain than "OpenAI." Use the horse-and-carriage crowd, the bankers of 2011, the von Neumann who was furious at the assembler.
3. End with the humble sentence, not the proud one. "I don't know what I'm doing" is why the talk is trusted. Jacob's version: "Bittensor is not finished. It is not decentralised yet. We do not know what the network will produce. We know the mechanism works because Bitcoin proved it."

**Mapping to the Bittensor thesis.** Victor's thesis is Jacob's thesis one level down: the danger is a generation that grows up thinking the current model is the only model. Victor says it about programming; Jacob says it about how intelligence gets owned. "They grow up with dogma. And once you grow up with dogma, it's really hard to break out of it" (30:29–30:33) is a line Jacob can quote about a world where AI is assumed to belong to companies.

### 9. Yuval Noah Harari, "Why fascism is so tempting, and how your data could power it" (TED 2018)

**What it is.** A 14-minute TED talk delivered as a hologram-style avatar from Tel Aviv, followed by a 4-minute conversation with Chris Anderson. It is the closest published talk to Jacob's core argument: the danger is not the technology, it is who ends up holding it.

**One-sentence thesis.** Dictatorship has always meant concentration of the era's most important asset, and in this era the asset is data, so the fight is over who controls it.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:01 | Hook | Self-referential joke about being digital, then a provocation: "How many fascists are there in the audience today?" |
| 00:39–04:55 | Define the term | "we've forgotten what fascism is." Nationalism versus fascism, each defined in one sentence: "nationalism tells me that my nation is unique... Fascism, in contrast, tells me that my nation is supreme." |
| 05:00–07:12 | Reframe the villain | "we tend to depict fascism as a hideous monster without really explaining what was so seductive about it." The Voldemort point, then the "fascist mirror." |
| 07:18–08:34 | The turn | "Fascism and dictatorships might come back, but they will come back in a new form." Land, then machines, then data: "dictatorship now means that too much data is being concentrated in the hands of the government or of a small elite." |
| 08:40–10:04 | Why now | Centralisation used to be inefficient; AI removes that handicap: "the main handicap of authoritarian regimes in the 20th century... will become their greatest advantage." |
| 10:10–11:24 | Second danger | Algorithms that know you better than you know yourself: "A dictator may not be able to provide me with good healthcare, but he will be able to make me love him." |
| 11:29–12:19 | The ask, split by audience | "The number one question that we face is who controls the data? If you are an engineer, then find ways to prevent too much data from being concentrated in too few hands." Then the rest of us: do not be manipulated. |
| 12:23–13:23 | The enemy's method | "They hack our feelings, not our emails... This is actually a method that Silicon Valley pioneered in order to sell us products." |
| 13:27–14:13 | Close | Return to the mirror: "just break that mirror." |
| 14:22–18:19 | Q&A | Anderson asks about corporations. Harari: "there isn't such a big difference between the corporations and the governments, because... the question is who controls the data. This is the real government." |

**How the enemy is framed.** Not as a person or a party. First he redefines the word so the audience cannot dismiss it as an insult (00:45–00:51). Then he makes the enemy structural: "too much data... concentrated in the hands of the government or of a small elite" (08:29–08:34). He is careful to say it is not a rerun of the 1930s (07:18), which lets a mainstream audience stay with him.

**The turn.** 07:38–08:34, the land-machines-data progression. Three sentences, same grammar, each one a historical era. The audience sees the pattern before he states the conclusion.

**The proof.** Historical analogy, not data. "In the 20th century, democracy and capitalism defeated fascism and communism because democracy was better at processing data and making decisions" (08:53–09:00). He then removes the assumption: "it is not a law of nature that centralized data processing is always less efficient than distributed data processing" (09:18–09:28).

**The ask.** Precise and addressed to engineers: "find ways to make sure the distributed data processing is at least as efficient as centralized data processing. This will be the best safeguard for democracy" (11:53–12:04).

**Rhetorical devices, with quotes.**

- Opening question that implicates the room (00:34).
- Definition before argument, so the argument cannot be waved away (02:21–02:37).
- Concrete images for abstractions: Satan "as a gorgeous hunk" (06:08); "the fascist mirror" (06:45).
- Parallel construction across eras: land, machines, data (07:38–08:24).
- Bookend: the mirror opens the seduction section and closes the talk (06:45, 14:09).
- Rule of three in the ask: "fear and hate and vanity" (12:32, 12:54).

**Slides.** None. The avatar is the visual.

**Admission of weakness.** In the Q&A: "we are not there yet" (15:37) and "in theory, yes, in theory, you can rise against the corporation... But in practice, it is extremely difficult" (15:54–16:02). He does not pretend the fix is easy. On his own record: "you should never underestimate human stupidity. It is one of the most powerful forces that shape history" (18:00–18:04).

**Moves worth stealing for Jacob.**

1. Define the scary word before you use it. Jacob's notes ask whether state-lab capture is "fascism." Harari shows the way: spend 90 seconds defining the term as an economic structure, say explicitly it is not the 1930s, then use it. Or use Harari's own frame instead: "dictatorship now means too much [intelligence] concentrated in too few hands." Substitute "intelligence" or "compute" for "data" and the sentence is Jacob's thesis.
2. Split the ask by audience, engineers first. Harari's "If you are an engineer" is the template for Jacob's three-way close (holders, builders, newcomers). Bittensor is a literal answer to Harari's engineering ask: distributed processing that is "at least as efficient as centralized."
3. Say why the old safeguard stopped working. Harari's best paragraph is the one where he explains that centralisation used to fail because it was inefficient, and AI removes the failure. Jacob's equivalent: labour used to be leverage; AI removes the leverage. Same structure, one paragraph, no numbers needed.

**Mapping to the Bittensor thesis.** This talk is the intellectual permission slip for Jacob's argument. A mainstream historian, on a TED stage, said the danger is concentration in "a small elite," that corporations and governments are the same for this purpose, and that engineers must build distributed alternatives that compete on efficiency. Jacob can quote him and then say: "Here is one."

### 10. Ilya Sutskever, NeurIPS 2024 Test of Time talk (2024)

**What it is.** A 17-minute award talk plus 7 minutes of questions, given by the co-founder of OpenAI and founder of Safe Superintelligence, to several thousand researchers in Vancouver. He was asked to look back at a 2014 paper. He used the slot to declare an era over and to name what comes next. Timestamps below are from the full Test of Time session audio (the saved transcript); the host hands over at 22:14 and Ilya's first transcribed line is at 22:34, so subtract about 22:20 to get the time in the standalone YouTube upload.

**One-sentence thesis.** Pre-training as we know it will end because there is only one internet, and what comes after will be agentic, reasoning, unpredictable systems, so the field should start thinking about that now.

**Structure, minute by minute (session time).**

| Time | Move | What happens |
|---|---|---|
| 22:34 | Hook | Thanks, then a photo: "a screenshot from a similar talk 10 years ago at NeurIPS in 2014 in Montreal. And it was a much more innocent time... This is the before. Here's the after, by the way." Laugh on his grey hair. |
| 23:26–23:53 | Contract | "a 10-year retrospective... a lot of the things in this work were correct, but some not so much." |
| 23:56–24:17 | Compression | The whole paper in three bullets: "It's an autoregressive model trained on text. It's a large neural network, and it's a large data set, and that's it." |
| 24:27–28:31 | What we got right and wrong | Reads his own 2014 slides. Right: the deep learning hypothesis, autoregression, scaling. Wrong: "an LSTM is the things that poor deep learning researchers did before Transformers"; "As we now know, pipelining is not wise. But we were not as wise back then." |
| 28:33–30:23 | The idea that lasted | Connectionism. Then credit to Radford, Kaplan, Amodei for "the age of pre-training." |
| 30:28–31:22 | The turn | "pre-training as we know it will unquestionably end. Pre-training will end. Why will it end?... The data is not growing because we have but one internet... data is the fossil fuel of AI... we've achieved peak data, and there'll be no more." |
| 31:29–32:21 | What comes next, others' answers | Agents, synthetic data, inference-time compute. He credits the field rather than claiming the answers. |
| 32:25–34:56 | Proof from nature | The brain-to-body mass chart: hominids have "a different slope on their brain-to-body scaling exponent... there is a precedent. There is an example of biology figuring out some kind of different scaling." |
| 35:03–38:58 | The long term | Superintelligence: agentic, reasoning, "the more it reasons, the more unpredictable it becomes," self-aware "because why not?" Then: "I'm not saying how, by the way, and I'm not saying when. I'm saying that it will." |
| 38:54 | Close | "on this uplifting note, I will conclude." Deliberate irony. |
| 39:08–46:16 | Q&A | Biology, hallucinations, incentives (see below), generalisation. |

**How the enemy is framed.** There is no enemy. The constraint is physical: one internet, finite data. "The data is not growing because we have but one internet" (30:50). This makes the turn feel like a law of nature rather than an opinion.

**The turn.** 30:28. The most-quoted line of the year in AI was a five-word sentence repeated: "Pre-training will end." He follows immediately with a why (30:34) and a metaphor (30:59, fossil fuel). Statement, reason, image, in 40 seconds.

**The proof.** One chart, explained slowly, with the axes read aloud: "this x-axis is log scale. You see this is 100. This is 1,000, 10,000, 100,000" (34:31–34:35). He teaches the room to read the chart before he draws the lesson from it.

**The ask.** Indirect. "without doubt, the field, everyone who's working here will figure out what to do" (34:56–34:59). He asks the room to think, not to act.

**Rhetorical devices, with quotes.**

- Self-deprecation about his own past work: "Was it wise to pipeline? As we now know, pipelining is not wise. But we were not as wise back then" (27:49–27:54).
- Repetition of the key sentence: "Pre-training will end" (30:32), "We have but one internet" (30:50, 30:57), "The more it reasons, the more unpredictable it becomes" (37:05–37:11).
- Metaphor that reframes a technical fact as an economic one: "data is the fossil fuel of AI" (30:59); "peak data" (31:06).
- Refusal to over-claim as a way of being believed: "I'm not saying how, by the way, and I'm not saying when. I'm saying that it will" (38:03–38:06).
- Naming collaborators by name in the middle of a victory lap (30:01–30:09), which reads as confidence.

**Slides.** About 20, sparse. Several are literally his 2014 slides, reused. One chart. One photo. Text slides carry a single sentence.

**Admission of weakness.** Woven through the first half. The LSTM was the wrong architecture, pipelining was wrong, and the 2014 optimism was "if one is charitable" correct (28:24). On the present: today's models are "kind of strangely unreliable and they get confused while also having dramatically superhuman performance on evals, so it's really unclear how to reconcile this" (36:24–36:36).

**The Bittensor moment (Q&A, 42:37–43:33).** A questioner asks: "How do you create the right incentive mechanisms for humanity to actually create it in a way that gives it the freedoms that we have as homo sapiens." Ilya: "I don't feel confident answering questions like this because it's like you're talking about creating some kind of a top-down structure government thing. I don't know. It could be a cryptocurrency, too. I mean, there's Bittensor, you know, there's things. I don't feel like I am the right person to comment on cryptocurrency." (43:14–43:35). CORRECTION (16 Sep 2026, see docs/research/affine-neurips-factcheck.md): a speaker-fingerprint check of the SlidesLive audio shows "It could be a cryptocurrency, too" and "there's Bittensor, you know, there's things" were said by the audience questioner, not Ilya. Ilya's words were "Yeah, I mean..." and "I don't feel like I am the right person to comment on cryptocurrency." Do not attribute the Bittensor line to Ilya. Safe framing: Bittensor was named from the floor at Ilya's NeurIPS Q&A, in the one question that asked about incentives, and he did not dismiss it.

**Moves worth stealing for Jacob.**

1. Declare the era over in five words, then give the physical reason. "Pre-training will end. Why will it end? Because... we have but one internet." Jacob's equivalent: "Payroll is over. Why? Because the firm no longer needs the people." One sentence, one reason, one image (labour share at 52.9%).
2. Use your own old slides as evidence. Ilya reads his 2014 slides and grades them. Jacob can put up the 2021 whitepaper's Yuma Consensus diagram and the Sept 2024 "proof of stake by end of next year" promise and grade them out loud. Honesty about the misses buys belief in the hits.
3. "I'm not saying how, by the way, and I'm not saying when. I'm saying that it will." This is how to make a strong prediction without a hostage date, and it works because he gave hard dates elsewhere. Jacob should give hard dates on what he controls (the key, ossification) and use Ilya's line on what he does not (when open networks out-compete labs).

**Mapping to the Bittensor thesis.** Ilya's talk is the AI-lab leader's own admission that the current scaling recipe is running out, that the next step is unknown, and that "different scaling" is possible. Jacob can quote it and then make the claim Ilya would not: a different scaling exponent needs a different substrate, and a permissionless market for compute, data and models is one. The Q&A gives him the line: at NeurIPS, asked about incentives, the room's answer was "a cryptocurrency... Bittensor."

### 11. Andrej Karpathy, "Software Is Changing (Again)" (Y Combinator AI Startup School, 2025)

**What it is.** A 39-minute keynote to about 2,500 students and new founders, by the former OpenAI founding member and Tesla AI director. It is the current best example of a dense technical talk that a general audience can follow, because every idea is an analogy the room already owns.

**One-sentence thesis.** Neural networks became programmable in English, so we are at the 1960s of a new kind of computer, and there is a decade of software to rewrite.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:13 | Hook | "Wow, a lot of people here. Hello." Then straight to the audience's situation: "you're about to enter the industry... it's actually like an extremely unique and very interesting time." |
| 00:44–01:09 | Thesis in one breath | "software is changing again... software has not changed much on such a fundamental level for 70 years. And then it's changed, I think, about twice, quite rapidly in the last few years." |
| 01:12–05:33 | Vocabulary | Software 1.0 (code), 2.0 (weights), 3.0 (prompts). "remarkably these prompts are written in English." |
| 06:40–09:02 | Three analogies for what an LLM is | Utility ("CAPEX to train... OPEX to serve... metered access"), fab, and operating system. |
| 11:00–12:05 | The 1960s frame | "LLM compute is still very expensive... that forces the LLMs to be centralized in the cloud. And we're all just sort of thin clients... the personal computing revolution hasn't happened yet." |
| 12:49–13:30 | The turn | "LLMs... flip the direction of technology diffusion... typically it is the government and corporations that are the first users... But I feel like LLMs are kind of like flipped around." |
| 14:42–18:03 | Psychology | "they're kind of like people spirits. They are stochastic simulations of people." Superpowers and "cognitive deficits": hallucination, "jagged intelligence." |
| 18:26–29:00 | Opportunities | Partial autonomy apps, the "autonomy slider", "keep the AI on the leash", the Iron Man suit. |
| 29:14–33:33 | Programming in English | Vibe coding; his own MenuGen story: "the vibe coding part, the code, was actually the easy part." |
| 33:35–38:11 | Build for agents | robots.txt to llms.txt; "meeting LLMs halfway." |
| 38:14–39:22 | Close | Recap of the three analogies in one sentence, then the ask: "we're going to take the slider from left to right... and I can't wait to build it with all of you." |

**How the enemy is framed.** No enemy. The nearest thing is over-excitement: "I think a lot of people are getting way over-excited with AI agents" (22:58) and "when I see things like, oh, 2025 is the year of agents, I get very concerned... this is the decade of agents" (27:33–27:42). He positions himself as the calm adult in a hype cycle, which reads as credibility.

**The turn.** 12:49–13:30. The observation that LLMs went to consumers before governments and corporations, the reverse of every previous technology. He does not draw the political conclusion; he leaves it.

**The proof.** Personal use and personal projects. "This is certainly like a lot of my use" (13:30). MenuGen, built in a day, then a week of DNS and payments (29:36–33:33). The proof is "I did this," not a chart.

**The ask.** "We need to rewrite a ton of code" (38:18) and "I can't wait to build it with all of you" (39:19). Invitation, not instruction.

**Rhetorical devices, with quotes.**

- Analogy stacking with explicit labels: "These LLMs are kind of like utilities, kind of like fabs, but they're kind of especially like operating systems" (38:24–38:27).
- Coined phrases that stick: "people spirits" (14:50), "jagged intelligence" (16:20), "autonomy slider" (20:24), "keep the AI on the leash" (22:56).
- Concrete absurd examples: "They will insist that 9.11 is greater than 9.9, or that there are two Rs in strawberry" (16:29–16:32).
- Historical rhyme: "It's like 1960s of operating systems" (38:30); "time-sharing" (11:22).
- Self-deprecation about the deck: "So this slide is not very good, I'm sorry" (23:46).

**Slides.** About 60, hand-drawn style, one idea per slide, dense with screenshots. He talks over each for 20 to 60 seconds. The slides are notes, not art; the analogies do the work.

**Admission of weakness.** About the technology, constantly: hallucinations, "cognitive deficits" (16:06), "I'm still the bottleneck" (23:06). About the future: "It's not clear what this looks like. Maybe some of you get to invent what this is" (12:02–12:05). About his own deck (23:46). He does not admit weakness about himself or his thesis.

**Moves worth stealing for Jacob.**

1. Give the room a vocabulary in the first five minutes, then use only that vocabulary. Karpathy's 1.0/2.0/3.0 is the spine; everything after is placed on it. Jacob's spine can be Bitcoin's two inventions: money nobody can print, and a market nobody runs. Everything after is "the second invention, pointed at intelligence."
2. Steal the 1960s frame directly. Karpathy says LLMs are mainframes, we are thin clients, "the personal computing revolution hasn't happened yet." That is a lab insider describing centralised AI as a temporary phase. Jacob can quote it and say Bittensor is the personal-computing move: many small owners instead of one time-shared machine.
3. Coin phrases, but make them describe things, not slogans. "Jagged intelligence" and "autonomy slider" work because they name a real observation. Jacob's "the machine that pays you back" and "permissionless basic capital" are fine, but he should test each: does it name something the audience can see?

**Mapping to the Bittensor thesis.** Karpathy's diffusion flip (13:20) is the deepest gift in this talk: this technology went to ordinary people first. Jacob's argument is that ownership has not followed use. People got the chatbot; a few companies got the equity. The 1960s frame supplies the arc: mainframe (labs), then personal computer (open weights, local inference), then, Jacob adds, the network that pays the owners of the machines. Also note the framing risk: Karpathy never names a villain and still moves a room, which is evidence that Jacob does not have to say "OpenAI" to make his point.

### 12. Demis Hassabis, "Artificial Intelligence and the Future" (RSA, London, 2016)

**What it is.** A 45-minute lab-leader keynote to a general public audience six months after AlphaGo beat Lee Sedol, by the co-founder and CEO of DeepMind. It is the standard form of the AI-lab mission talk: mission statement, taxonomy, results, demos, applications, benevolent close. Included as the model of what a lab leader sounds like when he is winning, and as a contrast to the founder talks above. Timestamps are from the archive.org audio; the recording has 12 minutes of silent pre-roll, the host begins at 12:04 and Hassabis at 14:20.

**One-sentence thesis.** Solve intelligence, then use it to solve everything else.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 12:04–14:09 | Host | Housekeeping, hashtag, bio. The host, not Hassabis, supplies the moral frame: "for the benefit of all humanity, not just an elite." |
| 14:20–15:11 | Hook | Courtesy, then the frame: "one way I'd describe it, is as an Apollo program effort for AI... bring together the world's best scientists and best engineers and put them in a perfect environment." Headcount: "around 350 people now." |
| 16:06–16:34 | Mission in two steps | "step one, fundamentally solve intelligence. And then step two... help us as a society solve everything else. Now, that might seem a little bit fanciful and far-fetched, But I hope by the end of this talk, I'll have convinced you that at least it's plausible that this could happen." |
| 16:39–19:56 | Taxonomy | Four dimensions: learning versus handcrafted, general versus narrow, grounded versus logic-based. "one system that out of the box can do a wide range of tasks." |
| 20:08–21:31 | Foil | Deep Blue: "Where does the intelligence of that software reside? It's not really in the machine or the algorithm. It's actually in the mind of the programmers." |
| 21:37–27:31 | Method and first proof | Deep reinforcement learning; why games ("when you're on a very long-term mission, like we are, a multi-decade mission, then it's even more important to know in the short term that you're heading in the right direction"); Atari Breakout. |
| 27:50–48:52 | The big proof | AlphaGo, told at length: the search space, the two networks, the match ("280 million viewers watched the five games overall," "35,000 press articles"), move 37 and Lee Sedol's reaction. Twenty minutes on one result. |
| 48:52–51:05 | Neuroscience | Systems neuroscience, the hippocampus, transfer to the real world. |
| 51:16–57:55 | Applications | NHS, data-centre cooling ("maybe we could save 10% of the energy consumption"), WaveNet speech and music demos. |
| 58:09–59:03 | Close | "We really think about AI as a kind of meta-solution to all these other issues. And if we can sort of fundamentally solve AI and make intelligence abundant... The two I'm most excited about are science and healthcare... Thanks for listening." |

**How the enemy is framed.** There is none. The only foil is narrow AI, Deep Blue, and it is treated with respect ("a huge technical feat," 20:19). The talk assumes goodwill; the word "elite" is used only by the host.

**The turn.** There is no turn in the founder sense. The emotional peak is move 37 (39:56–45:00), a demonstration rather than an argument: the machine did something "humans have been undervaluing" (42:00).

**The proof.** Overwhelming and single-sourced: one match, twenty minutes, with video of the opponent's face. The lesson for anyone building a proof section is that one deep result beats ten shallow ones.

**The ask.** None. The audience is asked to be convinced that the mission is "plausible" (16:34).

**Rhetorical devices, with quotes.**

- The Apollo analogy, repeated by the host and the speaker (13:30, 14:50), which frames a private company as a public mission.
- Two-step mission statement that everyone can repeat: "solve intelligence... solve everything else" (16:12–16:23).
- Foil-then-contrast: Deep Blue's intelligence was "in the mind of the programmers" (21:22); AlphaGo's was learned.
- Scale markers used as credibility: "around 350 people now and over 250 research scientists" (15:09–15:11), "280 million viewers" and "35,000 press articles" (38:15–38:33).
- Long-mission humility as a research virtue: the "multi-decade mission" line (23:29).

**Slides.** About 40, a conventional lab deck: mission slide, the four-dimension taxonomy, Atari and AlphaGo screenshots, the match video, application logos, audio demos. Dense, and the talk depends on them.

**Admission of weakness.** Small and technical: the music model "doesn't understand a thing about long-term melody yet" (57:07); "it's a bit like a drunken pianist at the moment" (57:39). Nothing about the mission, the ownership, or the risks. Compared with Ilya's 2024 talk, this is the difference between a lab leader at the start of a run and one at the end of it.

**Moves worth stealing for Jacob.**

1. A two-step mission everyone can repeat. "Solve intelligence, then solve everything else" has survived ten years because it has two verbs and no adjectives. Jacob's equivalent should be as short: "Pay for intelligence directly, and let anyone earn it."
2. Spend twenty minutes on one proof if you have one. Hassabis does not list DeepMind's papers; he tells one match, one move. Jacob's subnet section should pick one subnet (the strongest customer story, likely Chutes on OpenRouter or Targon with Intel) and tell it as a story with a customer's face, then mention the other layers in one sentence each.
3. Use the long-mission line to buy patience. "when you're on a very long-term mission, like we are, a multi-decade mission, then it's even more important to know in the short term that you're heading in the right direction" is the right way to introduce near-term proof points in a talk about a 2027 or 2030 destination.

**What not to copy.** The absence of an enemy, an ask, or a weakness. This is a talk for a room that already trusts the speaker and owns nothing. Jacob's room owns the asset and has been told the founder holds a key; the Hassabis form would sound evasive from him.

**Mapping to the Bittensor thesis.** This talk is the shape of the thing Jacob is arguing against, and that is useful. The Apollo frame ("bring together the world's best scientists... in a perfect environment") is the centralised model stated proudly: a few hundred people in one building. Jacob can quote it and say the opposite: the machine that produced us was not an Apollo program; it was a forest. The line "Where does the intelligence of that software reside?" (21:18) can be turned into Jacob's question: where does the ownership of that intelligence reside?

### 13. Peter Thiel, "Competition is for Losers" (Stanford CS183B, 2014)

**What it is.** A 42-minute lecture plus questions, to Stanford students in Sam Altman's "How to Start a Startup" course, with no slides and a whiteboard. It is the model for a founder who takes one contrarian idea and drives it through economics, history and psychology without ever leaving it.

**One-sentence thesis.** Value created and value captured are independent variables, so the only durable business is a monopoly, and the whole history of innovation is a history of who captured the value and who did not.

**Structure, minute by minute.**

| Time | Move | What happens |
|---|---|---|
| 00:25 | Hook | "I have a single idée fixe that I'm completely obsessed with on the business side... you always want to aim for monopoly... you want to always avoid competition. And so hence competition is for losers." |
| 01:06–01:54 | The formula | "you have a valuable company if two things are true. Number one, that it creates x dollars of value for the world. And number two, that you capture Y% of X... X and Y are completely independent variables." |
| 02:04–03:25 | Proof by contrast | Airlines versus Google: airlines matter more and make almost nothing; "the entire 100 year history of the airline industry... something like a quarter that of Google." |
| 03:40–05:30 | Two kinds of business | Perfect competition versus monopoly: "you often don't make money" in the competitive case (04:25). Later, at 08:09: "World of perfect competition is a world where all the capital gets competed away." |
| 05:34–13:13 | The lies | "the people who have monopolies pretend not to... the non-monopolists pretend to have monopolies." The British-food-restaurant-in-Palo-Alto joke; Google "almost never describes itself as a search engine these days." |
| 13:25–18:43 | How to build one | Start small, expand: "you want to go after small markets. If you're a startup, you want to get to a monopoly" (13:39–13:46). |
| 18:43–27:27 | Characteristics and durability | Proprietary technology, network effects, economies of scale, branding ("I never quite understand how branding works. So I never invest in companies where it's just about branding"). The last-mover idea: "Google is the last search engine" (23:18). "we undervalue durability." |
| 27:42–37:00 | The turn: history of innovation | "The scientists never make any money. They're always deluded into thinking that they live in a just universe that will reward them for their work and for their inventions." Einstein: "you don't get to be a billionaire, you don't even get to be a millionaire." |
| 37:02–42:10 | The psychology of competition | "we always think of losers as people who can't compete. And I want us to really rethink and, and revalue this and consider whether it's possible that competition itself is off." The law-firm story: "from the outside everybody wanted to get in, on the inside everybody wanted to leave." |
| 42:00 | Close | "don't always go through the tiny little door that everyone's trying to rush through. Maybe go around the corner and go through the vast gate that no one's taking." |

**How the enemy is framed.** Not a company. The enemy is a habit of mind: competition as validation. "so much of people's identities got wrapped up in winning these competitions that they somehow lost sight of what was important, what was valuable" (41:22–41:32). This lets him criticise the room's own path (Stanford, law school) without insulting anyone present.

**The turn.** 27:42–30:32. The lecture has been about companies. He then applies X-and-Y to all of science and technology and arrives at a moral claim: the people who created the most value captured none of it. The formula from minute one becomes a lens on history.

**The proof.** Pairs of companies and one formula applied to each. No charts. The airline number, the Google number, "Twitter is worth far more than anything Einstein did" as a reductio (36:11–36:16).

**The ask.** Personal and stated at the end: go through the gate nobody is taking (42:00–42:06).

**Rhetorical devices, with quotes.**

- One formula, introduced in the first two minutes and returned to at least four times: "X and Y are independent variables" (01:34, 29:05, 36:23).
- Provocation in the title, then a redefinition: "competition is for losers... which is always this provocative way to, to title things" (37:02–37:11).
- The "lies people tell" frame, which makes the audience distrust every company pitch they have heard (05:34–07:47).
- Personal story as the emotional proof: leaving the law firm "after seven months and three days" (41:07); "it's really reassuring to see you leave, Peter. I had no idea that it was possible to escape from Alcatraz" (41:12–41:17).
- Closing image that inverts the opening: the small door versus the vast gate.

**Slides.** None. Whiteboard only.

**Admission of weakness.** About his own knowledge: "I never quite understand how branding works" (21:54); "there's probably no sort of single formula to it" (18:52). About his own life: the law-firm mistake. He does not admit weakness in the thesis itself.

**Moves worth stealing for Jacob.**

1. X and Y. Thiel's formula is Jacob's economics section in one line: AI creates X, a huge number, and Y, the share that reaches ordinary people, is heading toward zero because payroll was the capture mechanism and it is gone. Then: Bitcoin was the first system where Y was distributed by protocol to whoever contributed. Bittensor applies that to X.
2. "The scientists never make any money." This is the strongest available framing for why open AI keeps losing to closed labs, and it comes from a Silicon Valley insider, not a critic. Open weights create X and capture no Y. Jacob's "they can defund it" tweet is the same point. A funding mechanism that no one can switch off is how Y stops being zero for open AI.
3. One idea, obsessively. Thiel announces "a single idée fixe" and never leaves it for 42 minutes. Jacob's notes have six ideas competing. The talk should have one: who owns the machine.

**Mapping to the Bittensor thesis.** Thiel's lecture is the mirror image of Jacob's. Thiel tells founders to build monopolies and capture Y. Jacob's claim is that AI monopolies are about to capture nearly all of Y for the first time in history without a payroll to share it, and that the answer is not to regulate the monopolies but to build a network where Y is issued to contributors by rule. Quoting Thiel on how monopolies "pretend not to" have monopolies (05:50) is also a clean way to describe how labs talk about competition and safety.

## Comparison table

| Talk | Length | Slides | Opening type | Enemy framing | Close type |
|---|---|---|---|---|---|
| Jobs, internal 1997 | 15 min | None (one ad, one tape of print work) | Promise of a reveal ("finishing this advertising... show it to you in a minute") | Own drift and neglect; rival named only to be dismissed | Goal reframed upward ("can we make Apple really great again?") |
| Jobs, Stanford 2005 | 15 min | None | Self-deprecating joke, then the contract ("Just three stories") | Internal: dogma, others' opinions | Borrowed line repeated three times ("Stay hungry, stay foolish") |
| Kennedy, Rice 1962 | 18 min | None | Formal greeting, joke, then triads on the moment | The tired ("rest... wait") and an unnamed hostile power | Quotation (Mallory) and a blessing on the adventure |
| King, 1963 | 16 min | None | Places the moment in history | A condition (the bad check, gradualism), not a person | Anaphora crescendo ("Let freedom ring") into song |
| Balaji, Network State 2023 | 12 + 7 min | Dense, duplicating the speech | Housekeeping and QR code, thesis at minute two | Legacy institutions as a list | Logistics ("let's get started"; exit poll, selfie) |
| Wood, DEVCON1 2015 | 18 min + Q&A | A handful, sparse | Confession that his own understanding changed; a rule to avoid jargon | Structural: "Servers provide barriers... make monopolies" | Coined names for the thing, then "Any questions?" |
| Antonopoulos, Zurich 2016 | 26 min + Q&A | None | Concept named and defined in 20 seconds | A recurring type, voiced in its own words ("See? See? It's not working") | Callback to all three examples, then the largest claim |
| Victor, DBX 2013 | 33 min | ~40 transparencies, one image each | Straight-faced anachronism, laugh in 30 seconds | Historical figures who could not see the next idea | Act dropped; humble sentence ("I don't know what I'm doing"); "then you're free" |
| Harari, TED 2018 | 14 min + Q&A | None | Question that implicates the room | A structure: concentration in "a small elite"; explicitly not the 1930s | Bookend image ("just break that mirror") |
| Sutskever, NeurIPS 2024 | 17 min + Q&A | ~20, sparse, several reused from 2014 | Photo of himself ten years ago; laugh | A physical constraint (one internet) | Ironic ("on this uplifting note, I will conclude") |
| Karpathy, YC 2025 | 39 min | ~60, hand-drawn, one idea each | Direct address to the audience's situation | None; over-excitement is the nearest thing | Recap of the three analogies, invitation ("build it with all of you") |
| Hassabis, RSA 2016 | 45 min | ~40, conventional lab deck | Courtesy, then the Apollo frame | None; narrow AI as a respectful foil | Mission restated, benevolent list, "Thanks for listening" |
| Thiel, Stanford 2014 | 42 min + Q&A | None (whiteboard) | The obsession stated in one sentence | A habit of mind: competition as validation | Inverted image ("the vast gate that no one's taking") |

Reading across the rows: the talks that are still quoted a decade or more later (Jobs twice, Kennedy, King, Antonopoulos, Victor, Wood, Thiel) used no slides or slides that were themselves the joke. Every one of them named the shape of the talk in the first minute and made the enemy either a past self, a condition, or a historical type. The two talks with the most slides (Balaji, Hassabis) are the two with no admission of weakness and no explicit ask.

## A proposed shape for Jacob's talk (45 minutes)

This borrows Jobs' announced three-part contract, Wood's ban on jargon until the thing is named, Kennedy's "whether we join or not" and "we are behind, but", Antonopoulos's pattern-taught-three-times, Harari's definition-before-argument and split ask, Ilya's five-word era-ending sentence, and King's one metaphor carried to the end. Timings assume about 130 words a minute and no Q&A inside the 45.

| Min | Section | What happens | Borrowed from |
|---|---|---|---|
| 0:00–2:00 | Open in the room | No slide. Name the feeling everyone in the room has had this year. "The fear is correct. Most people are pointing it at the wrong thing." Then the contract: "Three things tonight: what is actually happening, what the real fear is, and what we built." | Jobs 2005 (three stories), Harari (the room-implicating question) |
| 2:00–4:00 | Remove the exit | Naval's line, in Jacob's words: AI is not slowing down; the only decision is who owns it. Quote Ilya: "the personal computing revolution hasn't happened yet" is Karpathy's; "pre-training will end" is Ilya's. One slide, one quote. | Kennedy ("whether we join in it or not") |
| 4:00–10:00 | What is happening: the firm without a payroll | Four numbers, each anchored in the room: labour share 52.9%; Anthropic ~$1T with a few thousand people; GM 853,000 people in 1979; four firms $720–745B capex. Say why payroll was the transmission mechanism and why it is gone. One paragraph on why labour was leverage (Thelwall, the tax base, Engels' pause). End on the question: "What does organising look like when the machines no longer need the people?" | Kennedy (numbers with physical anchors), Harari (why the old safeguard fails) |
| 10:00–14:00 | The real fear, defined | Define the word before using it: not the machine, the owners. Harari's land-machines-data progression, with "intelligence" as the fourth line. Elysium, once, as the single metaphor of the night. Then: the people building this agree (Altman's 2.5% fund, Amodei's 3% tax) and every fix they propose runs through a treasury. | Harari (define the term), King (one metaphor) |
| 14:00–18:00 | Bitcoin: the proof | Teach the pattern: 2011 joke, 2016 "for criminals", 2026 Poland's power draw and ETFs. Bitcoin's second invention: a commodity market nobody runs. "Hashes are not FLOPs" said out loud before anyone else can. The problem: the product is a lock. | Antonopoulos (pattern taught before applied), Victor (show the date, let the room compute) |
| 18:00–21:00 | Nature and energy | Short. Forests, reefs, the body: no CEO. Memristors: current flows, paths strengthen, structure grows. "Money is energy that flows through a network, and where it flows, structure grows." | Jobs 1997 ("Apple at the core") for the one-sentence value |
| 21:00–25:00 | What Bittensor is | One breath: Bitcoin's mining machine pointed at intelligence, with the market deciding what counts. Subnets as markets. Yuma Consensus in one sentence. dTAO in one sentence. 21 million, halving done. Fair launch, no pre-mine, fairly mined then sold. | Karpathy (vocabulary first, then only that vocabulary) |
| 25:00–29:00 | The third path | Every lab that matters answers to Washington or Beijing. Bittensor answers to neither by construction. Montreal knows what downstream feels like. "Everyone else funds open AI until someone tells them to stop. We built the one that nobody can tell." | Balaji (Satoshi/Satya/Steve, the exit made respectable), Harari (corporation and state are the same for this purpose) |
| 29:00–34:00 | What has been built | Five layers in Jacob's own tweet order: compute, inference, data, knowledge, free intelligence, one live subnet each. Then the three facts that outsiders will believe: Chutes on OpenRouter, Grayscale and Bitwise filings, the first halving held. Then the scale admission: $2–3B against $5T, "a proof of mechanism, not yet a counterweight." | Jobs 1997 (logistics as proof; "not speeds and feeds"), Kennedy (admit you are behind) |
| 34:00–39:00 | I hold a key | Plainest words in the talk. Proof of authority today. The 35-minute halt in 2024. Templar and the apology. Wood's law: every system centralises to set its rules and decentralises once they hold. Then the dated path: NPoS, governance replacing the key, ossification around December 2027. Then the turn: "A centralized AI lab can never make this promise. Our key is a bug we are removing. Theirs is the product." | Jobs 2005 (the failure as origin of strength), Ilya (grade your own old slides), Wood (centralise, then decentralise) |
| 39:00–43:00 | The ask, three audiences | Holders: you own the only AI capital never allocated to anyone; stake well; hold through the year the key comes off. Builders: 128 markets, room for yours; build for a customer, not for emissions. Newcomers: you just heard the shape of the next twenty years; there is no gate. | Harari ("If you are an engineer..."), King (ask your own side for discipline first) |
| 43:00–45:00 | Close | Return to the first beat: the fear is correct, aim it correctly. "Bitcoin proved that a network with no owner can grow larger than any company. Bittensor is that network, pointed at intelligence, paying ownership to anyone who helps build it." The fork: a station in orbit, or a forest. "Build it with me." | Kennedy (the single long sentence, then Mallory), Jobs 2005 (borrowed closing line repeated) |

**Slide budget.** Twelve to fifteen slides, white on black. One quote, one number, or one image each. No slide for the opening two minutes or the key confession. Sections that carry the emotional weight (open, real fear, the key, the close) get no slide or a black slide.

**Three sentences to print on the wall and return to.**
1. "Do not fear the machine. Fear the machine owned by a few."
2. "Everyone else funds open AI until someone tells them to stop. We built the one that nobody can tell."
3. "Our key is a bug we are removing. Theirs is the product."

**What to cut if the talk runs long.** The neuromorphic section (18:00–21:00) can become one sentence inside the Bittensor section. The subnet layer list can drop to three layers. Do not cut the key confession or the scale admission; every strong talk here put the weakness in and the weak ones left it out.

## Pitfalls seen in weaker talks of the same kind

Drawn from the weaker moments inside this set (Balaji's opening and closing, the Q&A stretches, Hassabis's 2016 deck) and from the failure modes the strong talks visibly avoid.

1. **Housekeeping before thesis.** Balaji's strongest three sentences arrive at 02:19, after an MC warm-up, a QR code and a book plug. Jobs 1997 opens with "We were up until 3 o'clock last night finishing this advertising." Say the thing first.

2. **Reading the agenda.** Four of Balaji's twelve minutes are a speaker list. A conference audience already has the programme. The same failure in a founder talk is listing subnets, partners, or roadmap items one by one. Jobs 1997 names the warning: "The way to do that is not to talk about speeds and feeds."

3. **Attacking the rival by name.** None of the strong talks do it. Jobs 1997 mentions Windows only to refuse to talk about it. Antonopoulos quotes the bankers instead of insulting them. Victor makes von Neumann the villain so no one in the room is. Naming OpenAI or Anthropic as enemies from a stage in a recorded talk gives them the headline and gives the room permission to leave.

4. **Slides that repeat the speech.** Balaji's grid and speaker headshots say what he is saying. Hassabis's 2016 deck is a lab update: charts, results, logos. The talks that survive as text (Jobs, Kennedy, King, Antonopoulos, Harari) have no slides at all; the ones that survive as video (Victor, Ilya) have sparse ones. A keynote for holders is not a research update.

5. **No admission of weakness.** Balaji's opening has none, and it is the least trusted talk in the set. Every strong talk here concedes something real early: Kennedy "we are behind"; Jobs "drifted away from doing the basics"; Ilya "pipelining is not wise"; Antonopoulos "the result is messy." Leaving out the sudo key, the emissions-versus-revenue gap, or the scale gap does not hide them; it hands them to the first questioner.

6. **Stacking metaphors.** Jacob's notes currently hold Elysium, Gaza, feudalism, fascism, corporatism and a forest. King used one metaphor (the check) and cashed it. Harari used one (the mirror) and broke it. Pick one for the danger and one for the alternative.

7. **Using a loaded word without defining it.** Harari spent 90 seconds defining fascism before using it, and said explicitly it was not the 1930s. Using "fascism" for state-lab capture without that work will be the only line anyone quotes.

8. **Hedging the prediction into nothing.** The Q&A stretches in Ilya's and Harari's sessions are weaker than the talks because both speakers retreat to "I don't know" and "it's impossible to predict." Ilya's talk works because he found the form: "I'm not saying how, and I'm not saying when. I'm saying that it will." Give hard dates on what you control and use that form for the rest.

9. **Over-claiming the proof.** The Bitcoin FLOPs line ("out-computes Google") is the kind of claim that a technical audience will check during the talk. Kennedy's numbers were exact and anchored. Say hashes, say Poland, say "more machinery than any company has assembled for one purpose," and stop.

10. **Ending on logistics.** Balaji's opening ends "So in the beginning, let's get started." Karpathy ends "I can't wait to build it with all of you." Kennedy ends on Mallory. End on the sentence you want quoted, then stop talking.

11. **Speaking to one audience.** Hassabis's RSA talk is for a general public and loses the researchers; a lab all-hands loses the public. Harari split the ask ("If you are an engineer... As for the rest of us"). Jacob has three audiences in one room and must address each by name.

12. **Apologising for the medium.** Karpathy's "this slide is not very good, I'm sorry" is charming once. Balaji's "It's going to be almost like TikTok style, but in person" tells the room to expect less. Do not lower the bar out loud.

## Sourcing note

**How the transcripts were made.** YouTube blocks caption and video requests from this machine (cloud IP; "Sign in to confirm you're not a bot" even in a real browser), and the public Invidious mirrors return empty caption files. So every recording was taken from a non-YouTube host: archive.org mirrors (Jobs 1997 and 2005, Kennedy, King, Balaji, Wood, Antonopoulos, Victor, Harari, Karpathy, Hassabis, Thiel) and SlidesLive's official NeurIPS session audio (Sutskever). Each was transcribed locally with faster-whisper and cross-checked against a published text where one could be fetched: the archive.org human captions for Jobs 1997, the NASA/JFK Library text for Kennedy (archive.org copy), WeUseCoins for Antonopoulos, the glamour-and-discourse timestamped transcript for Victor, the Y Combinator and Singju Post transcripts for Karpathy, and the shun-liang GitHub transcript for Sutskever. Jobs 2005 and King were checked only by comparing the two Whisper passes against the widely published wording of those speeches (the Stanford News and TED pages timed out or refused the request from this machine). Balaji, Wood, Hassabis and Thiel have no published transcript; two Whisper passes were compared. Timestamps are from the recording named in each transcript's header and match the public video except where the header says otherwise (Sutskever: session audio, Ilya starts at about 22:20; Hassabis: 12 minutes of silent pre-roll).

**Machine-transcript caveats.** Each recording was transcribed twice, first with Whisper `small` to read from and then with Whisper `medium` for the saved files (except Antonopoulos, where `small` matched the published transcript better and was kept; each header says which). Expect small word errors, especially on the 1963 King recording and in fast Q&A. Where `medium` still mis-heard a word that a published text confirms (for example Kennedy's "$5,400 million", Victor's "if in 40 years", Sutskever's "one internet", Thiel's "idée fixe", King's "tied up with our destiny"), the correction was applied to the saved transcript and is listed in that transcript's header under "Corrections". Every quote in this document was then checked by script against the saved transcripts: all transcript quotes match verbatim; the only quoted strings that do not are Jacob's proposed lines and talk titles. Speaker changes in Q&A are not marked by the transcriber; the Sutskever Bittensor exchange should be checked against the video before it is quoted as his words.

**Not obtained.**
- Vitalik Buterin, Devcon 7 opening keynote "Ethereum in 30 minutes" (Bangkok, 12 November 2024). The Devcon archive lists only a YouTube source and a Swarm hash; the Swarm gateways return 404 and no transcript is published. Devcon 6 (Bogota, 2022) has the same problem. The only Vitalik talk on archive.org is a two-minute Zuzalu segment from the Network State Conference. Gavin Wood's DEVCON1 talk (Ethereum co-founder, first Devcon, four months after launch) was used instead as the crypto-founder-to-movement Devcon example.
- Sam Altman, OpenAI DevDay 2023 keynote. YouTube only; no mirror found.
- Chris Dixon, "Read Write Own." The only mirror is a 73-minute podcast conversation with Balaji, not a talk; the C-SPAN book talk on archive.org requires a login to download. Dropped.
- Jensen Huang GTC keynotes, Dario Amodei, Demis Hassabis Nobel lecture 2024: no non-YouTube recording found. Hassabis's 2016 RSA talk was used instead.

**Working files.** Recordings, Whisper outputs and scripts are under `/tmp/talks/` on the machine that produced this and are not part of the store.