Anthropic Posts $67B Run Rate as Distillation Fraud Threatens Revenue

Inflation Heats Up, Anthropic Investor Day, Tesla Roadster Mania| Diet TBPN — TBPN

TBPN

Anthropic faces billions in unauthorized token distillation extraction as third-party AI wrappers launder prompt traffic through automated proxy accounts.

AI-assisted summary of Inflation Heats Up, Anthropic Investor Day, Tesla Roadster Mania| Diet TBPN — original source. Check the original for full context.

Source published:

Key ideas

  • Anthropic reported Q3 revenue of $16.8 billion—annualizing to a $67.2 billion run rate—while disclosing that illicit token distillation through third-party SaaS wrappers represents 5% to 7% of its top-line revenue. (0:00)
  • Brad Gerstner announced Altimeter Capital fully exited its stake in CoreWeave after the stock plummeted roughly 50%, rotating approximately 10% of the fund directly into Meta and Tesla amidst broader tech sector rotation. (5:55)
  • Researchers at interpretability lab Goodfire revealed a breakthrough using agentic tools at scale, finding that the vast majority of neurons in trillion-parameter models are directly legible without requiring sparse autoencoders. (20:00)
  • Valar Atomics founder Isaiah Taylor sparked an ecosystem-wide venture debate after publicly clashing with Day One Ventures over contractual pro rata rights, highlighting founder-investor friction around defense sensitivities. (23:48)

Detailed notes

## Background Published on YouTube on October 16, 2026, by the tech talk show channel TBPN, this episode of Diet TBPN delivers a 30-minute digest hosted by John Coogan and Jordi Hays. According to the description, the program breaks down Anthropic's Investor Day disclosures, broader AI market turmoil, luxury inflation indicators, startup legal drama, and automotive anticipation around the Tesla Roadster and Porsche Mission S.

## TL;DR The hosts break down Anthropic's $67.2 billion annualized run rate alongside the technical crackdown on token distillation that accounts for 5% to 7% of its top-line revenue. They examine macroeconomic shifts, including Brad Gerstner exiting CoreWeave amid a broader AI selloff, surging high-end inflation tracked by the 'CPI Reserve' index, and an AI alignment breakthrough at Goodfire showing natural neuron legibility. The episode finishes with an analysis of Valar Atomics' public pro rata dispute with Day One Ventures, followed by predictions and deposits placed ahead of the imminent Tesla Roadster and Porsche Mission S unveilings.

## Atmosphere & style The show moves with a rapid-fire, conversational cadence that blends sharp venture capital deconstruction with irreverent internet-native banter. John Coogan and Jordi Hays maintain a playful dynamic, handing out literal physical 'Fell For It Again' awards for falling for online slop and staking on-air bets over whether Elon Musk's Roadster will require stunt ramps to fly. The tone oscillates effortlessly between serious frontier model analysis and comedic Silicon Valley cynicism, making complex technical topics like mechanistic interpretability and cap-table dilution feel like lively watercooler debates.

## Bulletin - [00:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=0s) Anthropic reported Q3 revenue of $16.8 billion—annualizing to a $67.2 billion run rate—while disclosing that illicit token distillation through third-party SaaS wrappers represents 5% to 7% of its top-line revenue. - [05:55](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=355s) Brad Gerstner announced Altimeter Capital fully exited its stake in CoreWeave after the stock plummeted roughly 50%, rotating approximately 10% of the fund directly into Meta and Tesla amidst broader tech sector rotation. - [20:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1200s) Researchers at interpretability lab Goodfire revealed a breakthrough using agentic tools at scale, finding that the vast majority of neurons in trillion-parameter models are directly legible without requiring sparse autoencoders. - [23:48](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1428s) Valar Atomics founder Isaiah Taylor sparked an ecosystem-wide venture debate after publicly clashing with Day One Ventures over contractual pro rata rights, highlighting founder-investor friction around defense sensitivities.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=0s) Anthropic posted a sequential 45% QoQ revenue increase to $16.8 billion in Q3 ($67.2 billion annualized), but markets reacted with disappointment against overextended expectations of a $90 billion upside print. - [04:47](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=287s) Illicit model distillation accounts for an estimated $3.5B to $5B in annualized revenue for Anthropic, with attackers routing automated bot queries through unwitting third-party SaaS wrapper apps to bypass security filters. - [05:55](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=355s) Recent market pullbacks reflect sector rotation out of high-flying AI plays into defensive equities and Treasuries rather than a systemic capital liquidation, despite hot Producer Price Index (PPI) prints. - [10:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=600s) The custom 'CPI Reserve' luxury index demonstrates that ultra-wealthy expenditures—including Chanel flap bags (+139%), Met Gala tickets (+233%), and Big Law partner rates—have dramatically outstripped standard inflation over the past decade. - [14:44](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=884s) Blake Byers's macroeconomic thesis posits that capital shifts toward high-yield AI infrastructure capex will trigger severe sovereign interest rate spikes before surging productivity growth outpaces debt costs. - [14:44](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=884s) Corporate hiring has devolved into an AI arms race, with candidate spam bots battling HR filtering bots and forcing applicants into AI avatar interviews observed by human partners. - [20:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1200s) Mechanistic interpretability is experiencing a major breakthrough as scaled agentic inspection demonstrates that trillion-parameter neural network circuits are far more legible than superposition theories previously suggested. - [23:48](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1428s) The standoff between Valar Atomics and Day One Ventures exposes how contractual pro rata rights are being weaponized as proxies in defense-adjacent startups managing sensitive government scrutiny. - [27:43](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1663s) Tesla's highly anticipated Roadster reveal features high-risk reservation mechanics, demanding a $5,000 card deposit followed by a $45,000 wire within ten days ahead of promised flight or jump demonstrations. - [31:07](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1867s) Leaked footage of the upcoming Porsche Mission S reveals aggressive hypercar styling that directly targets the $500,000 to $2,000,000 exotic vehicle segment dominated by McLaren and Ferrari.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=0s) Anthropic Investor Day, Distillation Fraud, and the Frontier

The broader tech market has been experiencing a month-long slump, but attention is laser-focused on Anthropic’s Investor Day disclosures. Anthropic reported Q3 revenue of $16.8 billion, which annualizes to a staggering $67.2 billion run rate—a massive leap from doing roughly a tenth of that figure just twelve months earlier. Even with sequential growth around 45% quarter-over-quarter, the market reacted with disappointment because investors had anchored their expectations to sustained 10x annual compounding, whispering about potential upside prints as high as $90 billion. Compounding their challenges, Anthropic is fighting battles across multiple fronts: Meta appears to have migrated business off Anthropic to use its own models, amid media leaks from Meta executives subtly taunting Anthropic's upcoming valuation.

Anthropic revealed that token distillation accounts for between 5% and 7% of its top-line revenue—a multibillion-dollar pool of illicit activity. While some critics had worried distillation was far larger, shutting it down means forfeiting billions of dollars in run-rate revenue in exchange for protecting their intellectual property from frontier model competitors. Rather than pursuing overly restrictive approaches like attempting to ban open source, David Sacks and Anthropic's team are prioritizing technical crackdowns. As John Coogan puts it: “Unless other people will spend a lot of money to try to emulate your model and recreate it, then you are not on the frontier.”

```mermaid flowchart TD accTitle: AI Distillation Scheme Flow accDescr: Flow diagram showing how unauthorized distillers extract frontier model weights through third-party wrapper SaaS products to avoid detection.

subgraph Distillers ["Distillation Operation"] D["Distiller / Distillery"] end

subgraph Wrappers ["Intermediary SaaS Layers"] W["AI Wrapper Apps<br>(Document Editors, Note Apps)"] Accounts["1,000+ Automated<br>Dummy Accounts"] end

subgraph Frontier ["Target Frontier Labs"] A["Anthropic API / Claude<br>(or GPT-6)"] Model["Distilled Open-Weight Model"] end

D -->|"spins up"| Accounts Accounts -->|"submits synthetic document queries"| W W -->|"aggregates & proxies API calls"| A A -->|"returns high-quality outputs"| W W -->|"funnels training data back"| D D -->|"trains clone weights"| Model ```

- **Reported Q3 Revenue:** $16.8 billion, annualizing to a run rate of $67.2 billion. - **Distillation Footprint:** 5% to 7% of gross revenue, representing an estimated ~$3.5B to ~$5B in annualized unauthorized distillation extraction. - **Open-Weights Ecosystem:** Meta released their open-source model *MuseFark Open*, illustrating their distinct ideological divergence from proprietary API-gated frontier labs.

---

[04:47](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=287s) Third-Party SaaS Wrappers and the "Distillery" Laundering Pipeline

Distillation actors exploit architectural loopholes by funneling prompt-generation traffic through innocent third-party SaaS wrappers rather than hitting Anthropic directly. Thousands of AI-native document editors, writing assistants, and specialized productivity software applications interface with Claude or GPT APIs under consumer or enterprise licenses. Distillers spin up hundreds or thousands of automated accounts across these smaller platforms.

Because wrapper companies aggregate customer queries across thousands of real users, Anthropic's security telemetry only sees legitimate enterprise SaaS API volume. In many cases, these wrapper startups report massive, artificial revenue growth driven purely by distillation bots—without knowing that almost no real human uses their tool. As John Coogan and Jordi Hays joke, this dynamic makes these wrapper startups nothing more than industrial "distilleries" exploiting privacy barriers to evade detection.

---

[05:55](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=355s) AI Selloff, Sector Rotation, and Brad Gerstner Exiting CoreWeave

The conversation shifts from Anthropic's internals to broader market turmoil. Tech commentators have descended into hyperbolic "blackpilling," with users on social media sarcastically advising people to sell all stocks, bonds, and fiat dollars to hoard land, food, medicine, and guns ahead of an imminent barter economy. In reality, capital is engaging in an aggressive sector rotation rather than an absolute liquidation. The NASDAQ is down roughly 0.3% to 0.4%, while the Dow Jones Industrial Average is up nearly a full percentage point, with defensive allocations and Treasuries soaking up liquidity draining from tech.

Frontier AI dynamics are also being dictated by "tall poppy syndrome": hyper-scalers like Microsoft and Amazon systematically fund competitors whenever any single model provider gets too dominant, forming collective alliances to avoid an intelligence monopoly. On CNBC, Altimeter Capital's Brad Gerstner announced a major portfolio shakeup, revealing that his fund had completely eliminated its CoreWeave stake after CoreWeave tumbled roughly 50% over the trailing month.

- Brad Gerstner explained the exit: “We have sold all of our CoreWeave. We don't own any of it today. We've taken exposure down over the last three months. We felt AI had overshot and would lead to a period of consolidation.” - Altimeter rotated approximately 10% of its fund book directly into Meta and Tesla. - Macroeconomic conditions worsened alongside the tech pullback, with Joe Weisenthal flagging the latest Producer Price Index (PPI) print as the hottest PPI reading in four years. - The hosts previewed *CPI Reserve* (`cpireserve.com`), a bespoke luxury inflation index tracking high-end expenses for the top 1%—such as Virgin Galactic suborbital spaceflights, Met Gala tickets, Chanel bags, and Porsche models.

[10:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=600s) Inflation for the 1%: CPI Reserve and the Cost of Luxury

John Coogan and Jordi Hays dive into the "CPI Reserve" basket (cpireserve.com), a tracking index monitoring the dramatic price surges of high-end luxury goods, elite experiences, and services over the past decade, playing a rapid-fire game of "overrated or underrated":

* **Classic Chanel Flap Bag:** Rose from $4,900 in 2016 to $11,700 in 2026. This represents a 139% jump over ten years—outpacing standard consumer price index growth threefold (which would have put the bag at roughly $6,800). Jordi dismisses it flatly: “You're a bozo if you're still buying this.” * **Met Gala Tickets:** Spiked from $30,000 in 2016 to $100,000, inflating at six times the national average rate. Jordi defends it as underrated: “This is charitable giving... And the Met Gala has grown so much. So many interesting folks there. So many interesting moments. Worth every penny.” * **Virgin Galactic Spaceflight Seat:** Rose from $250,000 to $750,000. Jordi passes, but John calls it the “steal of the century” and says he would go at any price. Jordi jokes that he would go to great lengths to block John from flying for safety reasons, imagining a disaster where John is left inside the cabin having to “roll a duct tape and I got to put it together and screw this” to improvise a parachute mid-air. * **Porsche 911 Carrera (Base):** Increased from $84,000 in 2016 to $135,000 (a 61% rise over 10 years, or 1.5 times baseline CPI). Jordi deems it still undervalued relative to the pure driving joy it produces, though he pines for an era where the entry-level 911 was locked at $99,000 forever. * **Big Law Partner Billing Rates:** Escalated from $1,500 an hour to nearly $3,000 per hour. John and Jordi joke that hiring a corporate partner is “no longer the cure for male loneliness,” riffing on dialing a lawyer just to shoot the breeze about how the family is doing while letting the massive 15-minute billing increments tick away. * **Manhattan Luxury Real Estate:** Noted as an outlier in the index because prices are actually down, driven by high local taxes prompting wealthy residents to relocate.

[14:44](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=884s) AI Interest Rate Spikes, Job Market Arms Races, and Bread vs. Circuses

The discussion shifts to macroeconomics via venture capitalist Blake Byers's modal thesis on artificial intelligence capital markets over the next decade. Byers argues that as capital shifts heavily out of safe Treasuries and into high-yield AI infrastructure capex, real interest rates will spike significantly:

> “My modal prediction for the next decade: interest rate spikes as capital shifts from saving treasuries to invest... due to higher returns promised by AI related capex.”

According to Byers, this will induce a multi-year debt-servicing panic for sovereign governments and borrowers before skyrocketing economic productivity and revenue growth ultimately dwarf borrowing costs, creating a new normal where the developed world adapts to 5% GDP growth and demands 10%.

The hosts then review Derek Thompson's analysis of a bizarre arms race in corporate hiring: AI application spammers are battling AI HR filtering bots, resulting in candidates being forced to interview with AI avatars. At McKinsey & Company, job candidates now complete 10-minute automated evaluations against an AI while human assessment partners stand in the room watching:

```mermaid flowchart LR accTitle: The AI Hiring Gladiatorial Arena accDescr: Visual flow showing candidates and AI sparring in front of corporate assessors.

subgraph Arena["The Interview Arena"] A["Job Candidate"] <-->|"Prompt injection & answers"| B["AI Avatar Evaluator"] end

subgraph Assessment["Observation Booth"] C["Corporate Partners & Assessors"] end

Arena -.->|"Evaluated as gladiatorial theater"| C ```

John quips that the hiring process has degenerated into a gladiatorial spectacle where applicants should simply focus on prompt-injecting the avatar: “I'm totally tempted to throw an AI avatar in front of that... just in the hope that someone convinces that AI avatar that it is in fact interviewing for the job.”

This sparks a broader tangent on the Roman concept of "bread and circuses." Jordi observes that modern elites are disproportionately pouring their wealth into entertainment assets: - Billionaires acquiring sports teams, UFC franchises, Formula 1, and viral exhibitions like the Savannah Bananas and Banana Ball. - Jordi is momentarily duped by viral social media rumors about an absurd Crumbl Cookies and Wingstop collaboration—earning himself John’s physical “Fell For It Again Award” prop for falling victim to timeline AI slop. - On the actual job front, Bank of America recently brought on 1,000 contractors to mitigate software vulnerabilities unearthed by automated AI tools. John analogizes the development to a self-perpetuating make-work economy: “Pandora's box opening creates thousands of jobs containing the stuff that is flowing out of Pandora's box.”

The segment wraps as the hosts tease fresh technical optimism around AI alignment and model interpretability, citing a post from Nick Cammarata and reactions from machine learning researcher Rune indicating that neural networks may prove far less opaque than once feared.

[20:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1200s) The Interpretability Breakthrough: Goodfire, Neurons, and Superposition

The conversation dives into a massive white pill for AI safety and mechanistic interpretability. A researcher at Goodfire, who co-authored early interpretability work at OpenAI, shared a breakthrough that challenges years of pessimism in the field. Previously, researchers feared that artificial neural networks were impossible to decode—growing like an unruly garden or "midwifing a deity instead of building a machine that is interpretable," where primitive tools only made roughly 2% of neurons legible. Now, utilizing modern agentic systems, the researcher found that the vast majority of neurons across models scaling up to trillions of parameters are completely legible, with many tangled circuits simply having been misunderstood. The breakthrough was framed as an application of the "bitter lesson" to interpretability: applying simple methods at absurd scale actually works. AI commentator Rune reacted on X by noting that this could be remembered as a landmark moment in scientific history, joking that the news is so promising that "even Eliezer Yudkowsky will have to lower his p(doom)."

```mermaid flowchart TD accTitle: Mechanistic Interpretability Breakthrough accDescr: Diagram contrasting previous superposition theory using sparse autoencoders with the modern direct neuron legibility breakthrough.

subgraph Previous["Previous Paradigm: Superposition"] A["Single Model Neuron"] -->|"Fires ambiguously"| B["Multiple Concepts (Cats, Dogs)"] C["Sparse Autoencoders (SAEs)"] -->|"Artificially expands model"| D["One Neuron per Feature"] end

subgraph Modern["New Breakthrough: Agentic Scale"] E["Modern Agentic Tools"] -->|"Direct Inspection"| F["Natural Legibility at Trillions of Parameters"] F --> G["Understand Existing Circuits"] G --> H["Precision Optimization for Safety & Capabilities"] end ```

The hosts break down what mechanistic interpretability means in practice for those unfamiliar with the underlying architecture. Historically, when text is passed through an LLM, individual neurons light up under seemingly unrelated circumstances—firing for cats as well as dogs. This led to the theory of "superposition," which posited that smaller models compress vast dimensional spaces into few neurons, requiring tools like sparse autoencoders (SAEs) to expand the model out so that every neuron maps to a single distinct feature. The new development suggests that researchers can finally bypass reliance on sparse autoencoders to look directly at the actual neurons and trace their circuits. Beyond AI alignment and safety, the hosts emphasize the massive capabilities upside: if developers can inspect the exact circuits running a model, they can surgically optimize performance on complex, niche topics like advanced mathematics.

- Co-author of OpenAI’s early interpretability work announced that modern agentic tools demonstrate the vast majority of circuits in trillion-parameter models are legible. - Early large-scale automated interpretability projects at OpenAI could only interpret roughly 2% of neurons. - Goodfire operates as a focused mechanistic interpretability lab, working similarly to independent third-party safety labs. - John Coogan previously built a project named "1776" using sparse autoencoders to expand compressed model representations into interpretable features.

---

[23:48](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1428s) Valar Atomics vs. Day One Ventures: The Pro Rata Proxy Fight

The timeline erupted into controversy over a fierce legal and contractual clash between Isaiah Taylor, founder and CEO of advanced nuclear startup Valar Atomics, and early backer Day One Ventures, led by Masha Bucher. On October 9th, Taylor published a detailed post revealing that all existing investors had agreed to waive their pro rata rights in a new funding round—except for Day One Ventures, who retained legal counsel to aggressively enforce their rights. Taylor brought the conflict into the public square on X, sparking a wider venture debate. Founders rallied behind Taylor's right to protect the company, while seasoned venture capitalists warned against discarding pro rata protections entirely. Founders Fund partner Delian Asparouhov cautioned the ecosystem with a "don't throw the baby out with the bathwater" stance, arguing that while Taylor may have justifiable reasons to dilute an investor with controversial foreign ties below major investor status, pro rata rights remain non-negotiable contractual standards that venture funds rely on to return capital.

```mermaid flowchart LR accTitle: The Valar Atomics and Day One Ventures Dispute accDescr: Flowchart illustrating the contractual standoff between Valar Atomics and Day One Ventures.

subgraph Conflict["The Impasse"] A["Isaiah Taylor (Valar Atomics)<br/>Expanding into sensitive defense/government work"] B["Day One Ventures (Masha Bucher)<br/>Enforcing binding contractual pro rata rights"] A <-->|"Dispute over dilution & cap table presence"| B end

subgraph Positions["VC Community Split"] C["Founders & Supporters<br/>Remove controversial international associations"] D["VCs (Delian Asparouhov)<br/>Protect pro rata precedent: 'Don't throw the baby out'"] end

Conflict -.-> Positions ```

The hosts point out that the public dispute over pro rata rights functions primarily as a convenient proxy for a deeper relationship breakdown. Valar Atomics is aggressively securing sensitive U.S. government and defense contracts, making Day One's controversial international ties a potential liability when competing for awards or undergoing scrutiny. However, legally undoing an investor's rights after papering them is nearly impossible. As one host summarizes the institutional VC perspective: “As long as I keep showing up with the money, me as an investor with a pro rata right, I can have 5% of this company forever.” Although Taylor clarified that Day One lacks technical information rights, the hosts observe that financial information rights alone can carry immense geopolitical value, referencing recent news around alleged corporate espionage at OpenAI.

> “My view is I don't even know if it's worth having the pro rata debate, because this isn't about pro rata at all.”

---

[27:43](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1663s) Tesla Roadster Reveal: Ramps, Thrusters, and the $50k Gamble

The episode shifts gears to automotive spectacle with the long-awaited Tesla Roadster reveal event scheduled for that evening (7:55 PM Central / 5:55 PM Pacific / 8:55 PM Eastern). Jordi Hays announces on air that he put down a deposit for the new Roadster the previous day, fully embracing the risk of looking foolish: “PUT DOWN a deposit yesterday for Roadster, and I'm fully ready to put that pin on the 'Fell for It Again' award.” Securing a place in line required placing $5,000 immediately on a credit card, with a required $45,000 wire transfer due within 10 days. Jordi explains his arbitrage strategy: if Elon Musk's promised "most amazing product demo ever" turns out to be incredible, he preserves an early queue position before millions rush the site; if the unveiling disappoints, he can cancel his initial deposit and jump out of line before sending the wire.

The hosts revisit an ongoing on-air bet regarding Elon Musk's hints that the Roadster will literally hover or leave the ground. John Coogan bets on physical jumps from flat ground via cold gas thrusters, rocket engine tech, or ducted fans. Jordi predicts theatrical driving maneuvers, arguing that the car will achieve flight using ramps:

> “So if there's ramps involved, Jordi wins. If it jumps off the ground, off a flat ground, I win. That's the bet.”

The banter touches on the original 2017 Roadster buyers, who paid the entire $50,000 to $250,000 deposit upfront and have waited nearly a decade with no delivered product. While skeptics call it a recurring promo scheme, the hosts note that Tesla eventually delivers its ambitious products—pointing out the production Cybertruck parked right outside the studio as proof that the company ultimately builds its wildest concepts.

[30:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1800s) The Tesla Roadster Delivery Dilemma

The conversation centers on the endless waiting game for the Tesla Roadster and what reasonable expectations remain for reservation holders. Pondering the absurdity of the decade-long wait, the hosts joke about an exit strategy where Elon Musk unveils a demo so wildly dangerous that authorities are forced to kill it: “And so I thought it'd be funny to demo the car in such an insane way that it just instantly, like... federal regulators come in and say, you cannot under any circumstances do this.” That way, Musk could throw up his hands, claim he built the vehicle but regulatory roadblocks make it impossible to ship, and announce that Tesla is retreating to simply producing only the Model Y.

When establishing the acceptable timeline for an actual rollout, the bar is set strictly at twelve months. As one host argues, “If it takes more than one year to ship this thing, then you fell for it again.” Having already waited a decade, buyers cannot be expected to endure another ten-year cycle after yet another launch event. The hosts try to recall the original deposit size, noting that reservation holders likely put down a quarter-million dollars up front for a car that still has not reached driveways.

- Expected shipping deadline: At most one year post-reveal to retain any credibility. - Original reservation cost remembered as roughly $250,000 for the full deposit. - Playful scenario envisioned: Regulatory ban acting as an easy scapegoat to drop the Roadster and revert entirely to Model Y production.

---

[31:07](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1867s) The Leaked Porsche Mission S Supercar

Shifting attention to traditional high-end performance, the hosts break down the late-night leaked footage showing a cover pulled off the Porsche Mission S. Reacting to the styling, one host confesses: “I feel like a kid who accidentally saw their Christmas present.” The aggressive aerodynamic silhouette makes an instant impression, leading to the bold prediction that it spells serious trouble for mid-range supercar mainstays like McLaren, Lamborghini, and Ferrari.

Between the functional roof scoop and the raw styling, the car immediately draws emotional comparisons to Porsche legends like the Carrera GT and the 918 Spyder. Even if Porsche claims higher production numbers or frames it differently in marketing, one host asserts, “I'm feeling Carrera GT, I'm feeling 918, and this just immediately becomes by far the most desirable car in the 500 and $2 million price range, in my opinion.” The design also carries an aggressively raw, antisocial element: the exposed rear tires look poised to fling massive rooster tails of gravel directly into anyone following behind on a spirited weekend drive, practically demanding aftermarket mud flaps.

The episode wraps up just 23 minutes before the official Mission S unveiling and hours before the hosts head out to watch the new Nathan Fielder film. They remind listeners that Christmas is a mere 70 days away, requesting five-star reviews on Apple Podcasts and Spotify, and pointing viewers to the daily newsletter at TBPN.com.

- Notable design features: Functional roof scoop, ultra-aggressive widebody styling, and exposed rear wheels that kick up rocks. - Market positioning: Emotional successor to the Porsche Carrera GT and 918 Spyder, targeting the $500,000 to $2,000,000 supercar tier. - Closing notes: Imminent official vehicle debut, plans to attend the new Nathan Fielder film, and subscription reminders for the TBPN ecosystem.

## Notable quotes > [00:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=0s) Unless other people will spend a lot of money to try to emulate your model and recreate it, then you are not on the frontier. — John Coogan

> [14:44](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=884s) Pandora's box opening creates thousands of jobs containing the stuff that is flowing out of Pandora's box. — John Coogan

## Mentioned Altimeter Capital, Anthropic, Banana Ball, Bank of America, Blake Byers, Brad Gerstner, Chanel, Cisco, Claude, Console, CoreWeave, CPI Reserve, CrowdStrike, Crumbl Cookies, David Sacks, Day One Ventures, Delian Asparouhov, Derek Thompson, Dow Jones Industrial Average, Eliezer Yudkowsky, Elon Musk, Ferrari, Figma, Formula 1, Founders Fund, Goodfire, Isaiah Taylor, Joe Weisenthal, John Coogan, Jordi Hays, Lamborghini, Mark Cuban, Mark Zuckerberg, Masha Bucher, McKinsey & Company, McLaren, Meta, Met Gala, MongoDB, MuseFark Open, NASDAQ, Nathan Fielder, Nick Cammarata, NYSE, OpenAI, Porsche, Porsche 911 Carrera, Porsche 918 Spyder, Porsche Carrera GT, Porsche Mission S, Public, Railway, Ramp, Rune, Sam Altman, Satya Nadella, Savannah Bananas, Shopify, TBPN, Tesla, Tesla Cybertruck, Tesla Model Y, Tesla Roadster, The New York Times, UFC, Valar Atomics, Virgin Galactic, Wingstop

## Video description Sign up for TBPN’s daily newsletter at TBPN.com

Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after.

Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella.

TBPN is made possible by: Ramp - https://ramp.com Public - https://public.com Cisco - https://www.cisco.com Console - https://www.console.com CrowdStrike - https://www.crowdstrike.com Figma - https://www.figma.com MongoDB - https://www.mongodb.com NYSE - https://www.nyse.com Railway - https://railway.com Shopify - https://www.shopify.com/

Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive

Chapters: 00:00 Anthropic’s $67B Run Rate + The AI Distillation Crackdown 05:55 AI Stock Selloff + CoreWeave’s 50% Drop 09:31 Inflation for the 1%: Chanel, Met Gala + Porsche 14:44 AI’s Interest Rate Shock + The Job Interview Arms Race 20:11 Can We Finally Understand AI? The Interpretability Breakthrough 23:48 Valor Atomics vs. Day One Ventures + The Pro Rata Fight 27:43 Tesla Roadster Reveal + The Leaked Mission S Supercar

[Source](https://www.youtube.com/watch?v=k8Ob9iXNrOU)

中文简报

Anthropic年化营收达670亿美元，蒸馏欺诈威胁其收入 第三方AI套壳应用通过自动化代理账户洗白提示词流量，Anthropic面临数十亿美元未授权Token蒸馏提取问题。 主持人深入剖析了Anthropic达672亿美元的年化营收，以及针对占其总营收5%至7%的Token蒸馏行为采取的技术打击。他们审视了宏观经济转变，包括Brad Gerstner在更广泛的AI抛售潮中清仓CoreWeave、“CPI Reserve”指数所追踪的高端通胀飙升，以及可解释性实验室Goodfire展示自然神经元可读性的AI对齐突破。本期节目最后分析了Valar Atomics与Day One Ventures公开的按比例跟投权争议，并分享了在特斯拉Roadster和保时捷Mission S即将亮相前所作的预测及定金支付情况。 Anthropic公布第三季度营收为168亿美元（年化营收折合672亿美元），同时披露第三方SaaS套壳应用的非法Token蒸馏占其总营收的5%至7%。 (0:00) Brad Gerstner宣布Altimeter Capital在CoreWeave股价大跌约50%后已全额清仓该公司股份，并在更广泛的科技板块轮动中将基金约10%的资金直接调仓至Meta和Tesla。 (5:55) 可解释性实验室Goodfire的研究人员披露了一项大规模使用智能体工具的突破，发现万亿参数模型中的绝大多数神经元直接可读，无需依赖稀疏自编码器。 (20:00) Valar Atomics创始人Isaiah Taylor在因合同约定的按比例跟投权与Day One Ventures公开起冲突后，引发了整个生态系统的创投争议，凸显了涉及国防敏感领域时创始人与投资者之间的摩擦。 (23:48)

Reading summary

TBPN explores shifting dynamics across the technology sector, highlighting the tension between rapid artificial intelligence scaling and revenue vulnerabilities alongside capital rotations into defense, luxury goods, and automotive developments. Frontier AI Scaling and Distillation Leakage (0:00) Anthropic reported Q3 revenue of $16.8 billion, representing an annualized run rate of $67.2 billion and roughly 45% sequential growth. Despite this expansion, public market sentiment softened because investors had anticipated run rates approaching $90 billion. Compounding these valuation expectations, Anthropic disclosed that token distillation accounts for an estimated 5% to 7% of its top-line revenue, translating to between $3.5 billion and $5 billion in annualized illicit model extraction. Unauthorized distillers bypass API detection by routing synthetic queries through third-party SaaS wrappers, such as document editors and writing assistants. By leveraging thousands of automated dummy accounts across these intermediary applications, distillation operations mask their activity within ordinary consumer enterprise traffic. While eliminating these vectors protects proprietary frontier intellectual property from competing open-weight efforts like Meta's MuseFark Open, shutting down the laundering pipeline directly forfeits billions in recurring software revenue. Capital Rotation and Emerging Macro Pressures (5:55) Broader financial markets are witnessing targeted sector rotation rather than an absolute liquidation of tech equities. While the NASDAQ dipped slightly and the Dow Jones Industrial Average gained nearly a full percentage point, investors have increasingly favored defensive assets and sovereign debt over volatile infrastructure plays. For instance, Brad Gerstner announced that Altimeter Capital liquidated its entire CoreWeave position following a 50% monthly slide, reallocating roughly 10% of the fund into Meta and Tesla due to concerns that AI infrastructure has temporarily outpaced near-term fundamentals. Macroeconomic headwinds further complicate the landscape, with recent Producer Price Index figures hitting four-year highs. Looking forward, venture capitalist Blake Byers projects that massive capital shifts toward high-yield AI capex could sharply drive up real interest rates over the next decade, triggering sovereign debt-servicing strain before potential productivity gains materialize. Mechanistic Interpretability and Model Legibility (20:00) AI safety research received a potential breakthrough from researchers at Goodfire, presenting evidence that challenges long-held pessimism regarding neural network opacity. Historical research had treated deep learning systems as uninterpretable black boxes, hypothesizing under the theory of superposition that individual neurons arbitrarily encode multiple unrelated concepts and requiring external sparse autoencoders to isolate features. Early attempts at OpenAI yielded legibility in only about 2% of examined neurons. According to Goodfire researchers, applying modern agentic tooling at scale demonstrates that the vast majority of individual neurons and underlying circuits in models scaling up to trillions of parameters are directly interpretable without artificial dimensional expansion. If validated, this direct circuit legibility could lower alignment uncertainty and allow engineers to surgically optimize model capabilities for complex, technical domains like advanced mathematics. Contractual Clashes in Frontier Venture Capital (23:48) A high-profile dispute between advanced nuclear startup Valar Atomics, led by founder Isaiah Taylor, and early backer Day One Ventures exposed growing friction between contractual venture norms and national security constraints. Taylor publicly revealed that Day One refused to waive pro rata rights in an upcoming funding round, seeking legal enforcement while all other existing investors consented to dilution. Venture capitalists like Founders Fund's Delian Asparouhov defended the sanctity of pro rata rights as an essential contractual mechanism for fund returns, warning against eroding baseline legal protections. However, commentators note that this dispute operates largely as a proxy fight over cap table control: as Valar Atomics pursues sensitive defense contracts, international ties tied to minority investors create perceived geopolitical exposure, highlighting how dual-use and defense-oriented startups face unique ownership frictions under standard venture contracts. Automotive Ambitions: Tesla and Porsche (27:43) Consumer automotive markets are balancing long-delayed experimental concepts against conventional supercar manufacturing. Tesla renewed public focus on its second-generation Roadster through an unveiling event, requiring prospective buyers to commit $50,000 reservation deposits—divided between an initial credit card hold and an upcoming wire transfer—with earlier buyers having waited nearly a decade since initial 2017 reservations. Market observers note that Tesla faces acute delivery accountability, suggesting the company has at most a one-year window to ship production vehicles before exhausting customer patience. Concurrently, leaked footage of Porsche's Mission S supercar showcased an aggressive design featuring a functional roof scoop and exposed rear tires. The vehicle is positioned as a direct spiritual successor to the Carrera GT and 918 Spyder, targeting the $500,000 to $2,000,000 performance tier against legacy mid-engine manufacturers. The source leaves unresolved whether frontier AI labs can systematically eradicate wrapper-based distillation without undermining third-party developer ecosystems, and whether hardware commitments like the Tesla Roadster will meet near-term commercial viability.

精炼阅读摘要

本期节目梳理了科技与宏观市场的核心动态，重点探讨了Anthropic在快速增长下面临的未授权模型蒸馏风险与市场估值压力、AI基础设施扩张驱动的宏观资金轮动，以及机制可解释性与创投治理等前沿议题。 Anthropic业绩与蒸馏洗白管道 (0:00) 在科技市场持续月度低迷的背景下，Anthropic于投资者日公布第三季度营收达168亿美元，年化营收折合672亿美元，季度环比增幅约45%，但仍未能满足部分押注十倍年复合增长至900亿美元的激进投资者。同时，Meta据报道已将其业务迁移至内部模型，并发布开源模型MuseFark Open。Anthropic披露其总营收中有5%至7%来自未授权的模型“标记蒸馏”（token distillation），折合年化约35亿至50亿美元。管理层与David Sacks倾向于通过技术手段而非封杀开源来清理这一非法收入池，主持人John Coogan指出，若竞争对手不愿花费巨资仿制，模型便称不上处于前沿。蒸馏操作往往利用第三方SaaS套壳产品作为代理洗白渠道。攻击者在数以千计的文件编辑或笔记类SaaS工具中注册大量自动化傀儡账号，这些产品聚合大量普通请求并调用Anthropic或GPT接口。Anthropic的安全遥测系统因此只能监测到合法的企业级SaaS调用量，而部分套壳初创企业在未察觉真实用户缺失的情况下，因蒸馏流量获得了虚假的账面收入爆发，沦为事实上的工业级“蒸馏工坊”。 宏观再通胀与资金向AI基建轮动 (5:55) 二级市场上科技股出现明显回调，但资金呈现防御性行业轮动而非彻底出清。纳斯达克指数微跌0.3%至0.4%，而道琼斯工业指数上涨近1%，防御性资产与国债承接了流出资金。Altimeter Capital的Brad Gerstner公开透露其基金在过去三个月持续削减头寸，现已彻底清仓了近一月暴跌约50%的算力提供商CoreWeave，并将大约10%的账面资金重新配置于Meta与Tesla，理由是认为AI此前过度透支并将进入整合期。与此同时，美国最新生产者物价指数（PPI）创下四年来最高涨幅，富人定制通胀指数CPI Reserve更显示超高端消费品与服务在过去十年出现三至六倍于基准CPI的飙升，例如经典款Chanel手袋上涨139%至11,700美元、Met Gala门票跳涨至10万美元、大律所合伙人时薪逼近3,000美元。在更长期的宏观展望中，风险投资人Blake Byers提出核心预测：随着资本为追逐高收益而从安全国债大规模转移至AI基础设施资本支出，真实利率将在未来十年大幅飙升，这将在短期引发主权政府的债务偿还恐慌，直至AI带来的超额生产力与GDP增长最终压倒借贷成本。 可解释性突破与职场AI军备竞赛 (14:44) 在AI安全与机制可解释性方面，独立安全实验室Goodfire的前OpenAI研究员带来了乐观进展。此前学界受困于“叠加”（superposition）假说，认为单一神经元会同时响应猫和狗等不同概念，仅有约2%的神经元能被原始工具解析，往往依赖稀疏自编码器（SAE）展开维度。如今研究人员利用现代智能体工具进行直接检查，发现即使在数万亿参数规模的模型中，绝大部分回路同样具有天然可读性，印证了将简单方法拓展至极致规模的“苦涩教训”。这一突破不仅缓解了对未可知超级智能的安全担忧，研究员Rune更评论称这甚至会迫使Eliezer Yudkowsky下调其毁灭概率（p(doom)），同时这也让开发者能够对复杂学科的底层回路实施精准优化。但在就业与日常应用端，技术应用却演变为荒诞的消耗战。Derek Thompson指出企业招聘中出现了求职者AI投递机器人对抗HR筛选AI的军备竞赛，麦肯锡等机构甚至要求求职者与AI虚拟人进行10分钟面试，并由合伙人现场观摩。与此同时，美国银行雇佣了1,000名承包商，专门负责修复自动化AI审查工具挖掘出的海量代码漏洞，形成了自我循环的技术应对型劳动力需求。 Valar优先认购权争端与特斯拉交付困境 (23:48) 创投生态方面，先进核能初创企业Valar Atomics创始人Isaiah Taylor与早期投资方Day One Ventures爆发公开法律纠纷。由于涉及美国敏感的国防与政府合同，Valar试图清理具有争议性国际背景的投资者，除Day One外所有老股东均同意放弃新一轮的按比例优先认购权（pro rata rights）。虽然Founders Fund的Delian Asparouhov警告不能因个别案例彻底废弃风投赖以生存的优先认购契约基石，但评论指出，争端实质上并非关于投资比例，而是初创公司为了国家安全审查而剥离敏感股东的代理人战争，因为在当前地缘与企业情报风险下，即便仅保留财务知情权也可能构成实质安全隐患。在消费级硬件领域，特斯拉举办了拖延近十年的Roadster发布会，预订者面临首付5,000美元且十天内电汇45,000美元的条件。主持人分析了其营销与技术赌注，押注演示是否会包含冷气推进器带来的离地弹跳或特技飞跃，但同时强调若特斯拉在发布后一年内仍无法正式交付，其商业信誉将面临严峻考验。此外，泄露的保时捷Mission S超级跑车以激进的气动外形与车顶进气口亮相，被视为对50万至200万美元区间超跑市场的有力冲击。 本期内容反映出前沿AI行业正处于估值预期降温、技术黑盒初步被解构以及产业链灰色地带亟待清理的交汇期。而在硬件端与资本治理中，无论是特斯拉的十年期交付兑现，还是敏感技术初创企业的股权主权博弈，都展现出创新理想与现实商业契约之间的激烈碰撞。

Detailed notes — zh

## Background 该内容由科技播客频道 TBPN 于 2026 年 10 月 16 日发布在 YouTube 平台上。本期为《Diet TBPN》的 30 分钟精简版节目，由 John Coogan 和 Jordi Hays 主持，聚焦当天科技界的重磅头条与宏观动态，涵盖通胀与 AI 板块波动、Anthropic 投资者日数据、特斯拉 Roadster 与保时捷超跑等话题。 ## TL;DR 节目首先分析了 Anthropic 投资者日披露的 672 亿美元年化营收以及围绕模型蒸馏形成的灰色产业链，并探讨了 AI 概念股抛售潮与宏观利率和通胀重燃对市场的冲击。随后，主持人剖析了求职面试中 AI 工具对抗的荒诞生态，以及机械可解释性研究在神经元回路微观层面上取得的重大突破。最后，节目聚焦了核能初创企业 Valar Atomics 与 Day One Ventures 的按比例优先认购权纠纷，并盘点了今晚特斯拉 Roadster 的揭幕预期与意外泄露的保时捷 Mission S 超跑谍照。 ## Atmosphere & style 整期节目节奏紧凑、风趣幽默，两位主持人以科技圈“局内人”的视角展开高信息密度的调侃与犀利吐槽。对话充满了硅谷黑话与网络迷因，既有对万亿级 AI 叙事和宏观异动的一针见血剖析，又有下注特斯拉跑车是否会用跳台飞天的无厘头赌局。收听体验就像在旁听一场兼具深度商业洞察与解构狂欢的科技沙龙，既硬核刺激又极具娱乐性。 ## Bulletin - [00:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=0s) Anthropic 投资者日公布 Q3 营收达 168 亿美元（年化运行率 672 亿美元），环比增长 45%，但未达极端乐观预期；同时蒸馏流量占其营收 5% 至 7%，催生出数十亿美元的欺诈黑产。 - [05:55](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=355s) 市场遭遇 AI 板块抛售与资产轮动，CoreWeave 近一个月暴跌约 50%，Altimeter Capital 已清仓 CoreWeave 并转向 Meta 和特斯拉，资金正从科技股流入传统蓝筹股。 - [20:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1200s) 机械可解释性（Mechanistic Interpretability）迎来突破，研究证实万亿参数模型的神经元回路高度可读，反驳了神经网络无法解析的悲观假说。 - [23:48](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1428s) Valar Atomics 创始人公开指控 Day One Ventures 强推按比例优先认购权（Pro Rata），引发关于防务敏感核能初创企业受早期资方地缘背景制约的代理人战争。 - [27:43](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1663s) 特斯拉即将揭晓新一代 Roadster 悬念，市场对平地火箭悬停或道具跳台充满对赌，买家设定 1 年交付窗口期以防重蹈跳票覆辙；同日保时捷纯电超跑 Mission S 谍照曝光引发震动。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=0s) Anthropic 披露 Q3 营收 168 亿美元、年化运行率 672 亿美元，但因市场此前定价了 900 亿美元等极端预期且 Meta 转向自研模型，未能提振市场情绪。 - [00:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=0s) “模型蒸馏”占 Anthropic 营收的 5% 至 7%，黑产通过批量注册 SaaS 套壳产品掩护抓取，主持人指出被竞争对手高成本蒸馏已成为评判前沿模型的真实硬指标。 - [05:55](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=355s) Brad Gerstner 透露 Altimeter Capital 已全数清仓 CoreWeave 并削减 AI 敞口，将 10% 仓位轮动至 Meta 和特斯拉，反映高估值科技股向传统防守资产的轮动。 - [09:31](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=571s) 美国 PPI 录得四年来最火爆数据，通胀重燃背景下，“CPI Reserve”指数显示面向富豪的奢侈品与体验通胀大幅跑赢常规 CPI，如 Chanel 经典包十年暴涨 139%。 - [14:44](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=884s) 投资者 Blake Byers 预测，AI 带来前所未有的资本支出将直接推高实际利率，并在经历三年短期恐慌后将发达国家常态 GDP 增速推高至 5%。 - [15:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=900s) 招聘生态陷入 AI 军备竞赛，AI 批量投递与 AI 面试官对抗衍生出“角斗场”式的提示词注入与越狱博弈，加剧了职场筛选流程的异化。 - [20:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1200s) 机械可解释性研究取得突破，万亿参数模型的神经元大多展现出清晰回路美感，业界或不再必须依赖复杂的稀疏自编码器（SAE）即可直接调优微观回路。 - [23:48](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1428s) Valar Atomics 与 Day One Ventures 的争议表面是 Pro Rata 优先认购权，实质是涉及美国防务与核能合同的企业在早期资方复杂地缘背景下的合规危机。 - [27:43](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1663s) 特斯拉新一代 Roadster 开启 5,000 美元定金预订，市场就其飞天噱头究竟是靠冷气推进器平地悬停还是道具跳台飞跃展开对赌。 - [30:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1800s) 买家对特斯拉 Roadster 设定了 1 年交付的心理底线，若再度陷入漫长跳票，其信誉将面临严峻考验。 - [31:18](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1878s) 提前泄露的保时捷纯电超跑 Mission S 搭载车顶进气道与激进空力套件，被视为 Carrera GT 与 918 的精神续作，或将重塑 50 万至 200 万美元区间的超跑竞争格局。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=0s) Anthropic 投资者日与“模型蒸馏”黑色产业链 市场近期的震荡主要源于对 Anthropic 投资者日（Investor Day）披露数据的反应，但整个 AI 交易板块实际上已经连续下挫近一个月。例如算力基础设施公司 CoreWeave 在过去一个月左右暴跌了约 50%。尽管 Anthropic 的第三季度营收环比增长达到 45%，但在投资者早已习惯“每年营收暴增 10 倍”的超高预期下，依然未能满足市场胃口。Anthropic 在投资者日上宣布其第三季度营收为 168 亿美元，折合年化运行率（annualized run rate）达到 672 亿美元。尽管主持人调侃这个数字仿佛在搞网络迷因（类似 Zoom 的谐音），但鉴于一年前其营收仅为目前的十分之一，这一规模本已相当惊人。然而市场此前普遍预期会有向上惊喜（如达到 900 亿美元），因此情绪依然受挫。 与此同时，Anthropic 还在多条战线上腹背受敌。首先是 Meta 的业务流失——Meta 正在转向自研模型，甚至有高管在媒体上隐晦放话称，如果 Meta 停止向 Anthropic 采购，可能会严重扰乱后者的 IPO 估值。其次是严峻的“模型蒸馏”（Distillation）问题。Anthropic 指出，蒸馏行为占据了其营收的 5% 至 7%。虽然 David Sacks 等人对此持乐观态度，认为无需通过封杀开源等极端手段就能彻底打击蒸馏，但按年化计算，这意味着有数十亿美元的 token 价值是在被非法用于蒸馏模型。主持人一针见血地指出：“Unless other people will spend a lot of money to try to emulate your model and recreate it, then you are not [frontier]”，即当前评判一个模型是否真正处于“前沿”（Frontier）的最真实标准，就是看有没有人肯砸重金去蒸馏它。 蒸馏黑产往往通过第三方的 SaaS“套壳”（wrapper）公司隐秘进行。市场上存在大量打着“AI 原生文档编辑器”旗号的产品，用户开通账户后，后台调用 Claude 或 GPT 进行推理。蒸馏者（被戏称为“蒸馏厂/distilleries”）为了规避 Anthropic 严格的监控，会在这些小包装公司上批量注册数千个账号进行大规模抓取。这导致许多包装公司看似营收暴涨，实则根本没有真实用户，完全沦为蒸馏黑产的跳板。 ```mermaid flowchart LR accTitle: AI模型蒸馏黑产链路 accDescr: 展示前沿模型通过套壳SaaS被黑产蒸馏的调用与数据回流流程 subgraph S1["前沿实验室"] A["Anthropic (Claude API)"] end subgraph S2["中间层"] B["SaaS 套壳产品<br>(AI 文档编辑器等)"] end subgraph S3["黑产端"] C["蒸馏者 / 黑产农场<br>(批量注册账号)"] D["衍生开源/专有模型<br>(低成本复制能力)"] end A -->|"API 推理输出"| B B -->|"混淆流量 / 导出文档"| C C -->|"训练数据回流"| D ``` - Anthropic Q3 营收为 168 亿美元，折合年化营收运行率（ARR）达 672 亿美元，季度环比增长 45%。 - 蒸馏流量占 Anthropic 整体营收的 5% 至 7%，对应数亿美元的直接投入与高达 50 亿美元规模的蒸馏欺诈运行率。 - 市场存在“高罂粟综合征”（Tall Poppy Syndrome）：一旦某家前沿实验室跑得太快，微软、亚马逊等大客户和竞争对手就会结成反抗联盟（如早期的反 Netflix、反 OpenAI 联盟），通过扶持其他实验室来防止智力垄断。 --- ### [05:55](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=355s) AI 概念股抛售狂潮与核心资产轮动 在 AI 板块剧烈回调的背景下，市场上充斥着悲观主义（“blackpilling”）的极端言论，甚至有人半开玩笑地喊出“卖掉美元、卖掉股票、囤积枪支和药物、准备重返以物易物时代”的末日论调。然而真实的市场行情并非全面崩盘，而是典型的板块轮动。知名投资人 Brad Gerstner（Altimeter Capital）当天上午在 CNBC 节目中公开透露，他们已经清仓了所有 CoreWeave 股票，过去三个月持续降低 AI 敞口，因为他们认为 AI 交易此前已经严重超买并即将迎来盘整期，同时他们将 10% 的仓位轮动调配到了 Meta 和特斯拉。 盘面数据直观反映了这一趋势：纳斯达克指数仅微跌 0.3% 至 0.4%，而道琼斯工业指数却上涨了近 1%，资金正从科技股明显分流至传统蓝筹股和美国国债。虽然这在短期内对 AI 交易构成了沉重打击，但主持人强调，Anthropic 的投资者日并非正式 IPO，更像是在上市前主动“晒出脏衣服”排雷，为未来的反弹奠定基础。 - Brad Gerstner 在 CNBC 上公开表示：“We have sold all of our CoreWeave. We don't own any of it today. We've taken exposure down over the last three months. We felt AI had overshot and would lead to a period of consolidation.” - Altimeter Capital 将其投资组合中 10% 的资金轮动买入 Meta 和 Tesla。 - 宏观盘面上，纳斯达克仅下跌 0.3%-0.4%，而道琼斯指数上涨近 1%，显示市场正在从高估值科技股向传统防守型资产轮动。 --- ### [09:31](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=571s) 通胀重燃与面向 1% 超级富豪的特权通胀指数 宏观经济数据再度亮起红灯。最新的 PPI（生产者价格指数）数据表现极其火爆，彭博知名记者 Joe Weisenthal 评价称“那是四年来最火爆的 PPI 数据”（that was the hottest PPI number in four years）。对于做空股票、做多通胀或持有硬资产的投资者而言，这一天可谓大获全胜。 与此同时，节目探讨了一个专门衡量超级富豪生活成本的趣味网站——“CPI Reserve”（cpireserve.com）。与普通人日常消费的一篮子 CPI 构成截然不同，该指数专门追踪全球前 1% 精英人群的专属通胀率，监测项目包括搭乘维珍银河（Virgin Galactic）进入太空的票价、顶级名牌与奢华体验的花销等，生动揭示了富豪阶层与普通大众在通胀体感上的巨大鸿沟。 ### [10:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=600s) 富豪级通胀指标：CPI Reserve 与奢侈品评测 主持人 John Coogan 和 Jordi Hays 探讨了一个名为 CPI Reserve 的趣味项目（网址为 cpireserve.com），该项目追踪了面向前 1% 富豪人群的物价通胀，并展开了一场“高估（Overrated）还是低估（Underrated）”的快速点评： * **Chanel 经典翻盖包（Classic Flap Bag）**：2016 年售价为 4,900 美元，到 2026 年飙升至 11,700 美元，10 年内涨幅超过 139%，通胀速度是美国常规 CPI 的 3 倍以上。如果严格按照 CPI 调整，它现在的售价应为 6,800 美元左右。Jordi Hays 毫不客气地评价道：“You're a bozo if you're still buying this.”（如果你现在还买这个，你就是个大傻瓜），认为它绝对是被严重高估，而且“一直以来都如此”。 * **Met Gala 门票**：2016 年一张门票需花费 30,000 美元，如今已涨至 100,000 美元，是全美平均通胀率的 6 倍。然而 Jordi Hays 却认为它依然被低估（Underrated），理由是 Met Gala 本质上属于慈善捐赠，门票更贵意味着捐款更多，再加上该活动的影响力持续壮大、汇聚了极多有趣的人和名场面，因此“物有所值”。 * **维珍银河（Virgin Galactic）太空游座位**：过去单张票价为 250,000 美元，如今已涨至 750,000 美元。Jordi Hays 惊呼这是“世纪大便宜”（Steal of the century），无论什么价格他都愿意去。John Coogan 则调侃称自己会竭尽全力阻止 Jordi 登机；面对安全隐患的质疑，Jordi 搞笑地表示自己无论发生什么都肯定能活下来，两人打趣脑补了飞船半空散架时 Jordi 现场掏出一卷强力胶布（duct tape）一边抢修、一边现场自制降落伞跳伞生还的画面。 * **保时捷 911 Carrera 基础款**：2016 年售价为 84,000 美元，如今涨到了 135,000 美元，10 年涨幅达 61%，约为同期 CPI 的 1.5 倍。Jordi Hays 认为考虑到驾驶带来的纯粹快乐，它依然处于被低估状态，不过他也感叹要是保时捷能永远将入门款 911 维持在 99,000 美元该有多好。 * **顶级律所（Big Law）合伙人时薪**：大型律所合伙人的收费涨幅同样远超 CPI。以前每小时约 1,500 美元，如今已逼近每小时 3,000 美元。两人开玩笑说，去律所咨询再也无法成为“治愈男性孤独症的良药”了，过去 1,500 美元一小时还可以随便打电话找律师闲聊家常，现在按 15 分钟甚至 10 分钟计费都贵得惊人，只能直接找律师包下一整个周末或者买下 100 小时的打包服务。 * **曼哈顿豪宅市场**：在这一片物价飞涨中，唯独曼哈顿的豪宅价格出现了下跌，他们将其归因于针对富人的“Pied-à-Terre 税”等高额税收促使富豪离开纽约。 --- ### [14:44](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=884s) 资本市场与 AI 带来的利率震荡 讨论转向了投资者 Blake Byers 针对 AGI 与资本市场所作出的十年核心预测模型。Blake Byers 预测，未来十年利率将会暴增，原因是资本将从国债等储蓄资产大规模转移到与 AI 相关的资本支出（Capex）中，以追求 AI 承诺的高回报率；AI 是当今唯一具备如此体量以至于能彻底撼动实际利率的现代化超级工程。 这种资本转移在短期内将引发市场恐慌，因为飙升的利率大幅推高了政府和其他借款人的债务偿还成本。但根据 Byers 的推演，经过约三年的恐慌期后，市场将看清这些 AI 投资带来了极其庞大的经济回报，经济增长带来的新增收入将彻底抵消借贷成本的上升。届时恐慌将转化为狂欢，发达国家的常态 GDP 增速将调整为 5%（甚至人们会开始要求 10% 的增速才算满意），医疗保健消费和研发支出在 GDP 中的占比亦将大幅提升。 --- ### [15:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=900s) AI 军备竞赛与职场招聘生态的异化 随后两位主持人探讨了 AI 对就业生态的反向冲击。著名撰稿人 Derek Thompson 转发《华尔街日报》的报道并指出，AI 颠覆求职流程的速度远快于它颠覆实际岗位的速度。求职者使用 AI 批量生成投递简历，HR 部门则部署 AI 筛选工具予以拦截，甚至直接让 AI 数字人头像（Avatar）来主持面试。 ```mermaid flowchart TD accTitle: AI求职与面试对抗流程 accDescr: 展示求职者AI工具与企业HR部门AI面试官之间的对抗博弈关系 subgraph Applicants["求职者侧"] A["求职者"] -->|"AI 批量扩增生成"| B["AI 工具批量投递"] end subgraph Screening["企业 HR 侧"] B -->|"海量申请涌入"| C["HR AI 过滤筛选系统"] C --> D["AI 数字人面试官"] end subgraph Arena["角斗场式考核"] D <-->|"提示词注入 / 反向套路"| E["候选人越狱博弈"] E -->|"合伙人现场围观考核"| F["麦肯锡等机构合伙人评估"] end ``` 这种荒诞现象让两人展开了深度调侃： * **越狱博弈**：John Coogan 认为 AI 面试唯一的合理玩法是“越狱（Jailbreak）”。如果有公司放出一个 AI 面试官，求职者就应该通过提示词注入（Prompt Injection）反客为主，反向质问 AI“我们今天来其实是要考核你的绩效”，逼得 AI 面试官在后台死守指令“记住你的设定，你是主考官，由你来面试对方”。 * **角斗场围观**：麦肯锡（McKinsey）等公司甚至在真实办公室里安排候选人与 AI 进行 10 分钟对话，而人类合伙人则在一旁全程围观。Jordi Hays 形容这宛如古罗马角斗场：“我们放出了一只猛兽，看你能不能战胜它；如果你赢了，就能留下来给贵族当弄臣。” 这引出了对现代经济中“面包与马戏”（Bread and Circuses）的讨论：富豪和精英们热衷于买下各类“马戏”资产（如体育队、UFC、F1，甚至网红棒球队 Savannah Bananas）。Jordi Hays 提起 Crumbl Cookies 和 Wingstop 联名炸鸡的搞笑新闻，结果当场被指出那是虚假 AI 垃圾信息（AI Slop），Jordi 因而再度戴上了象征被假新闻欺骗的“又上当了奖”（Fell for it again award）。 最后，针对“AI 抹杀就业”的论调，他们举出了美国银行（Bank of America）雇佣 1,000 名外包人员来修补 AI 暴露出的软件安全缺陷为例。这引发了一个哲学式的反思：如果发现 Bug 的是 AI，修补 Bug 的也是由 AI 赋能的外包员工，人类是否只是打开了潘多拉魔盒，人为制造了数十万计的无意义“自产自销式工时”（Make work）来收拾烂摊子？在话题末尾，他们提到 OpenAI 领域专家 Nick Cammarata 指出神经网络的可解释性可能比学者们预期的更容易攻关，这一前沿进展让许多研究人员感到振奋。 ### [20:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1200s) 机械可解释性迎来突破：AI不再是解不开的混沌迷宫 主持人探讨了 AI 机械可解释性（Mechanistic Interpretability）领域的重磅进展。有研究员曾指出，可解释性或许是当今世界上最重要的科学难题，因为如果我们根本不知道自己创造了什么，就根本不可能解决 AI 对齐（Alignment）问题。过去，研究人员对神经网络抱有极大的悲观态度，认为它们就像杂乱纠缠的死结，人类仿佛不是在制造可理解的机器，而是在“培育花园”或“为神明接生”。早年在 OpenAI 启动的首个大规模自动化可解释性项目中，使用原始工具只能勉强看懂约 2% 的神经元。但随着现代 Agent 工具与更细致的研究手段介入，局面发生了翻天覆地的转变：高达数万亿参数的模型中，绝大多数神经元其实都是清晰可读的，其内部回路甚至展现出令人惊叹的美感；过去许多被认为“乱成一团”的神经元案例，事后证明纯粹是人类当初理解有误。这一进展让乐观情绪迅速蔓延，研究者预言“野心勃勃的可解释性问题将在我们最需要它的时候被彻底攻克，甚至可能很快就会实现”，这也再度验证了苦涩的教训（The Bitter Lesson）——最简单的方法配合荒谬的庞大算力尺度往往最管用。正如 Rune 在社交网络上转发评价的那样：“News so good, even Eliezer Yudkowsky will have to lower his P-do.”（消息好到连埃利泽·尤德科夫斯基都不得不下调他对人类灭亡概率的预估）。 主播们进一步解释了机械可解释性背后的技术原理与演进。这家名为 Goodfire 的独立安全实验室正全力攻坚此方向。过去研究人员观察单个神经元时深感困惑：给模型输入一段关于猫的文本，神经元 A 被激活；再输入一段关于狗的文本，同一个神经元居然又亮了。这种一个神经元对应多个主题或特征的现象被称为“叠加态”（Superposition）。过去业界的解决方案是训练稀疏自编码器（Sparse Autoencoders, SAE），假设大模型本质上是一个超大特征空间的高维压缩版，试图通过将其解压展开，让每个神经元回归到单一特征（比如单独设立一个专属于“猫”的神经元）。然而最新的突破表明，人类或许不再需要依赖复杂的稀疏自编码器，而是可以直接洞察神经元本身的真实运作机制。这一飞跃不仅为 AI 安全筑牢了防线，也为模型能力优化打开了直通车——工程师可以精准定位模型解某类特定数学题时的回路，并直接针对性调优。 ```mermaid flowchart TD accTitle: 机械可解释性研究路线演进对比 accDescr: 对比传统叠加态假设配合稀疏自编码器与现代自动化 Agent 工具直接解析模型回路的新范式 subgraph Past["早期悲观假说与方案"] A["神经元呈现叠加态 (Superposition)<br/>(同个神经元同时响应猫和狗)"] --> B["稀疏自编码器 (SAE)<br/>强行展开为单特征虚拟神经元"] B --> C["仅约 2% 神经元可被原始工具理解"] end subgraph Present["现代全新突破"] D["现代自动化 Agent 工具深入探测"] --> E["万亿级参数模型回路具备高度可读性与结构美"] E --> F["直接解析与干预物理神经元"] end Past -.->|"技术范式迭代"| Present ``` - 该领域代表性实验室 Goodfire 专注于模型最微观层面的神经元回路机制剖析。 - 过去的认知障碍源自工具简陋，导致大量纠缠特征被误读；全新自动化工具证实了“简单方法乘以巨大规模”同样适用于可解释性科学。 - 这一突破使模型能力的精细化调优成为可能，工程师可针对小众垂直领域的数学与推理回路展开微观手术。 --- ### [23:48](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1428s) Valar Atomics 围剿 Day One：按比例优先认购权背后的代理人战争 节目将焦点转向了过去这个周末在科技创投圈引发震荡的公开争端：核能初创企业 Valar Atomics 创始人兼 CEO 艾赛亚·泰勒（Isaiah Taylor）与早期投资机构 Day One Ventures 之间的正面对决。事情起因于 10 月 9 日，泰勒发布长文公开声讨 Day One，透露公司在最新一轮融资中已成功说服所有其他投资方放弃按比例优先认购权（Pro Rata Rights），唯独 Day One 拒绝让步，甚至聘请顶级律所发起诉讼施压、死咬不放。眼看私下协商破裂，泰勒选择将纠纷彻底引爆至 X 平台公屏，掀起了一场风险投资与创业者立场的大辩论。 这绝非一场单纯关于优先认购权的法务摩擦。Valar Atomics 正在承接越来越多的美国政府防务及高敏感度核能合同，而 Day One 及其主理人玛莎（Masha）在国际组织与海外背景上的复杂关联，让其持有核心大股东地位变得极为微妙与棘手。虽然德利安（Delian Asparouhov）等风险投资人发声提醒创业圈“不要倒洗水连孩子一起倒掉”，警示创始人不能将合法谈判写入合同的 Pro Rata 当作可以随意撕毁的软性条款；但主持人一针见血地指出，所谓优先认购权根本只是掩护，实质是一场代理人战争：“My view is I don't even know if it's worth having the pro rata debate, because this isn't about pro rata at all.”（我的看法是，这场辩论甚至不值得围绕 Pro Rata 展开，因为这根本就不是优先认购权的问题）。 主播分析，接受风险投资就如同步入一场漫长的法律婚姻，一旦白纸黑字写入合同，只要资方持续开出支票行使 Pro Rata，理论上就能永久锁定 5% 的股权。尽管泰勒坚称 Day One 并不拥有核心技术信息权，且公司大可将其置于臂长之外（Arm's length），但财务信息权在国际地缘博弈背景下同样价值连城——正如近期 OpenAI 爆发的间谍风波一样。真正的问题在于，伴随 Day One 的深度持股，Valar Atomics 是否正在丢掉关键政府大单，这是否已经成为竞对攻击其安全资质的软肋。 - 创始人艾赛亚·泰勒已令其余全体投资机构签署 Pro Rata 弃权协议，Day One 成为唯一的对簿公堂者。 - 投资人阵营力挺 Pro Rata 的契约神圣性，防止未来创始人群起效仿将不喜欢的资方恶意稀释下桌。 - 争议核心在于涉足政府与国防级业务时，早期资方的地缘政治敏感背景对企业产生的连带合规杀伤力。 --- ### [27:43](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1663s) 特斯拉 Roadster 今晚揭盅：火箭推进器还是飞跃跳台？ 节目的最后部分转向了当天科技界最受瞩目的重头戏——特斯拉全新跑车 Roadster 的发布会（定于美国中部时间今晚 7:55 / 太平洋时间 5:55 开幕）。其中一位主持人甚至在昨晚直接刷信用卡支付了 5,000 美元定金锁单，准备在今晚揭晓答案后决定是补齐余款还是当机立断止损，并戏称自己随时准备好认领“又被忽悠了奖”（Fell For It Again Award）：“So it's five grand on a credit card. And you have to wire 45,000 within 10 days, but I'm going to know whether or not it's a good car by tonight.”（信用卡刷了五千刀定金，十天内得电汇四万五，但我今晚就能知道这车到底行不行了）。如果马斯克口中“有史以来最震撼的产品演示”真能惊艳全场，提前下定就能在潮水般的预订大军涌入前死死守住靠前的排队顺位。 两位主持人还就马斯克反复暗示的“飞天跑车”展开了下注对赌。一派观点极度实用主义，认为车要上天最省事的方式就是借助道具，断言现场一定会搭建各种形式的跳台（Ramps），由特技飞车冲坡腾空；另一派则坚信马斯克会兑现冷气推进器技术，押注跑车能在纯平地面上依靠火箭喷气发动机或涵道风扇实现悬停离地。双方敲定赌局：只要使用了起跳坡道，押注跳台的一方获胜；如果是平地旱地拔葱，押注火箭动力的一方胜出。 回顾历史，主持人们感叹道，最早在 2017 年掏出全额巨款订购 Roadster 的车主才是真正的苦主，足足等了近十年却连车影都没见到。但这也算不上特斯拉的诈骗，毕竟 Cybertruck 虽然屡屡跳票且充满争议，但最终还是扎扎实实地下线量产、跑在了大街上。今晚的这场大秀，既是马斯克工程狂想的试金石，也将终结车迷长达数年的漫长等待。 - 订车资金门槛：信用卡预刷 5,000 美元，随后 10 天内需通过银行电汇补齐剩余 45,000 美元定金。 - 核心悬念对赌：特斯拉究竟是依靠 SpaceX 合作的推进器在平地离地腾空，还是在舞台表演中依靠物理跳台飞跃。 - 交付信誉背景：2017 年初版 Roadster 全款预定用户经历近十年空窗期，而 Cybertruck 的最终量产交付给今晚的技术落地提供了一定支撑。 ### [30:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1800s) 特斯拉 Roadster 的交付魔咒与监管困局假想 主持人继续深入探讨特斯拉新一代 Roadster 的漫长交付周期与消费者的心理预期底线。大家普遍感到厌倦：如果特斯拉要求客户付下巨额定金，却让大家再等上整整十年，这是绝对无法接受的。因此核心问题在于，究竟怎样的交付时间表和功能缩水会让预订者彻底失望。主持人戏谑地构想了一种极具戏剧性的演示场景：马斯克把这款跑车设计并演示得过于疯狂逆天，以至于联邦监管机构当场介入叫停，直接下达禁令声明绝不允许在任何情况下将这种车开上路；随后马斯克便能顺水推舟地摊手甩锅：“产品我已经造出来了，只是我发不了货，那我们还是回头老老实实造 Model Y 吧。” 针对究竟还要等多久的问题，主持人斩钉截铁地设立了严格的窗口期，直言特斯拉已经拖延了十年，绝不能再消耗客户下一个十年。针对这一交付预期，一位主持人立场鲜明地断言：“It's got to be one year. If it takes more than one year to ship this thing, then you fell for it again.” 随后两人聊起当年第一批 Roadster 的预订定金，回忆起当时似乎需要交出全额车款，金额大概高达 25 万美元（quarter mil）。在他们看来，与一再画饼的特斯拉 Roadster 相比，即将面世的保时捷 Mission S 无疑是一个更让人安心、大概率会准时交付的务实选择。 - 特斯拉 Roadster 的交付底线被设定为最多 1 年，否则市场和买家就会彻底沦为又一次营销狂欢的受害者。 - 主持人调侃当年 Roadster 的创始版（Founder Series）定金几乎是全款，约为 25 万美元。 --- ### [31:18](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1878s) 保时捷 Mission S 谍照曝光与超跑市场冲击 话题随即转到前一天深夜意外泄露的保时捷 Mission S 超跑实车揭幕视频。画面中工作人员正缓缓揭开车辆遮罩，展露出这台纯电猛兽的全貌。看到谍照的主持人难掩兴奋，直呼自己就像是“一个不小心提前撞见圣诞礼物的孩子”，整台车的视觉冲击力和设计侵略性远远超出了原本的温和预估。这种极具压迫感的外观造型，被认为会对迈凯伦（McLaren）、兰博基尼（Lamborghini）和法拉利（Ferrari）等传统中置超跑制造商带来沉重打击。 针对保时捷在设计上的突破与超跑格局的震动，主持人给出了极高的评价：“Emotional feeling it's giving me 918 successor, and I know that they might say oh, it's not it's a different thing it's gonna be a high production or whatever. It doesn't matter emotionally. I'm feeling Carrera GT, I'm feeling 918 and this just immediately becomes by far the most desirable car in the $500k to $2 million price range, in my opinion.” 两人随后细致品味了整车的硬核机械细节，车顶赫然配备了招牌式的车顶进气道（roof scoop），两人笑称这简直是“顶级媒体人（S tier journalist）”在旧金山街头炸街兜风的梦中情车。不过，激进的空气动力学设计也带来了一些极端的视觉特征，比如外露后轮卷起石子的角度显得格外狂暴，甚至被调侃为具有“反社会属性（antisocial）”——如果谁在周日开着它跟车队巡游，后方车辆必定会被后轮卷起的石子砸得千疮百孔，除非车主老老实实装上挡泥板（mud flaps）。 ```mermaid flowchart TD accTitle: 汽车定位与市场竞争关系 accDescr: 对比保时捷 Mission S 继承的传奇跑车血统及其对中置超跑市场的冲击 subgraph Heritage ["传承基因"] CGT["Carrera GT (自然吸气巅峰)"] P918["918 Spyder (混动三剑客)"] end subgraph MissionS ["保时捷 Mission S 超跑"] Design["激进造型 + 车顶进气道 (Roof Scoop)"] Market["50 万至 200 万美元区间绝对标杆"] end subgraph Rivals ["传统中置超跑竞品"] FER["法拉利 (Ferrari)"] LAM["兰博基尼 (Lamborghini)"] MCL["迈凯伦 (McLaren)"] end Heritage -->|"设计与情绪价值延续"| MissionS MissionS -->|"构成降维打击"| Rivals ``` --- ### [33:04](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=1984s) 节目收尾与未来展望 节目的尾声洋溢着轻松欢快的倒计时氛围。主持人向观众致谢，并提醒大家关注即将开始的官方发布会：距离保时捷 Mission S 的正式亮相仅剩 23 分钟，而距离他们前往观看 Nathan Fielder 的最新电影也仅剩几个小时。 在互动的最后，两位主持人呼吁听众在 Apple Podcasts 和 Spotify 平台上给 TBPN 打出五星好评，同时半开玩笑地提醒大家距离圣诞节其实只剩下 70 天，别忘了前往 TBPN.com 订阅每日新闻通讯（newsletter）。关于本期节目的最后一段呼吁，主持人热情总结道：“Leave us five stars on Apple Podcasts and Spotify. Get ready for Christmas. It's only 70 days away. And sign up for a newsletter at tbpn.com.” ## Notable quotes > [00:00](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=0s) Unless other people will spend a lot of money to try to emulate your model and recreate it, then you are not [frontier] — John Coogan / Jordi Hays > [05:55](https://www.youtube.com/watch?v=k8Ob9iXNrOU&t=355s) We have sold all of our CoreWeave. We don't own any of it today. We've taken exposure down over the last three months. We felt AI had overshot and would lead to a period of consolidation. — Brad Gerstner ## Mentioned Anthropic, CoreWeave, Meta, David Sacks, Brad Gerstner, Altimeter Capital, Tesla, Joe Weisenthal, CPI Reserve, Chanel, Met Gala, Virgin Galactic, Porsche, Blake Byers, Derek Thompson, The Wall Street Journal, McKinsey, Bank of America, Nick Cammarata, Goodfire, Eliezer Yudkowsky, Valar Atomics, Isaiah Taylor, Day One Ventures, Delian Asparouhov, OpenAI, SpaceX, Cybertruck, Tesla Roadster, Porsche Mission S, McLaren, Lamborghini, Ferrari, Nathan Fielder, TBPN, John Coogan, Jordi Hays ## Video description Sign up for TBPN’s daily newsletter at TBPN.com Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after. Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella. TBPN is made possible by: Ramp - https://ramp.com Public - https://public.com Cisco - https://www.cisco.com Console - https://www.console.com CrowdStrike - https://www.crowdstrike.com Figma - https://www.figma.com MongoDB - https://www.mongodb.com NYSE - https://www.nyse.com Railway - https://railway.com Shopify - https://www.shopify.com/ Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive Chapters: 00:00 Anthropic’s $67B Run Rate + The AI Distillation Crackdown 05:55 AI Stock Selloff + CoreWeave’s 50% Drop 09:31 Inflation for the 1%: Chanel, Met Gala + Porsche 14:44 AI’s Interest Rate Shock + The Job Interview Arms Race 20:11 Can We Finally Understand AI? The Interpretability Breakthrough 23:48 Valor Atomics vs. Day One Ventures + The Pro Rata Fight 27:43 Tesla Roadster Reveal + The Leaked Mission S Supercar [Source](https://www.youtube.com/watch?v=k8Ob9iXNrOU)

Source quotes

[0:00] John Coogan / Jordi Hays: Unless other people will spend a lot of money to try to emulate your model and recreate it, then you are not [frontier] [5:55] Brad Gerstner: We have sold all of our CoreWeave. We don't own any of it today. We've taken exposure down over the last three months. We felt AI had overshot and would lead to a period of consolidation.

Source quotes

[0:00] John Coogan: Unless other people will spend a lot of money to try to emulate your model and recreate it, then you are not on the frontier. [14:44] John Coogan: Pandora's box opening creates thousands of jobs containing the stuff that is flowing out of Pandora's box.

Original transcript

[0:00] So the market is mostly reacting to the Anthropic Investor Day numbers, but it's been in a decline for a long time for the last month. Core weaves down something like 50 percent over the last month or so. It's been rough in the AI trade broadly. Way off. Only 16 percent over the last month. 45% quarter on quarter and still not meet expectations, but when people have had in their minds 10x every year, 10x revenue growth every year, and Anthropic goes to their investor day and says that Q3 revenue was $16.8 billion, which annualizes to $67 billion, $67 billion. I don't know what they were doing there. That seems like a little bit of like... Okay, I get it. It's like the Zoom reference, but this is serious financial. There's a lot of money on the line. Can we not do the meme thing right now? Anyway, they are, in fact, annualizing their quarterly run rate to an annual run rate of $67.2 billion, which is remarkable for a company that was doing... you know, a tenth of that a year ago. But it's – seemingly not enough because everyone in their minds was like, they're going to surprise to the upside. They're going to throw up 90. But they really have been fighting. fires on multiple fronts. So there's like the meta business, which has pivoted over. They're probably using their own model now. And maybe they, and there was that like leak of like an executive at meta saying like, oh yeah, if we stop buying Anthropic, like it could really mess up their IPO. That would be a shame. I would hate for a company to IPO at the lower valuation than us, since we're almost $2 trillion. Like that would be terrible. So it seemed like sort of a veiled threat in the media. And then distillation. So – Distillation is between 5% to 7% of revenue, which I think is a little bit lower than people were expecting, but it feels like the Anthropic team, David Sachs, was celebrating this that, hey, there was this big problem, and one of the pro solutions was like, bad open source, do all this crazy stuff, slow down, and David Sachs didn't want to do that. and we believe that we can shut down distillation. Pretty staggering that five to seven percent, you know, that that's that's billions of dollars spent stealing from because that's the value of the of the of the distilled tokens, which is pretty crazy that that you can be at that scale of essentially fraud. Like you're not supposed to be distilling. It's against the terms of service. like reselling the data or whatever. It's like, yeah, they might have like, A COUPLE MILLION DOLLARS DEPLOYED INTO THIS SCHEME. BUT YOU'RE LOOKING AT, YOU'RE LOOKING AT FIVE BILLION DOLLARS OF FRAUD, RUN RATE FRAUD, YOU KNOW, SOMETHING LIKE THAT. BUT AS THEY, AS THEY FIGHT THAT FIRE, Obviously, that's going to be a drag on revenue growth. If they open the floodgates, they're like, yeah, go distill away. You know, at least in the short term, revenue would be growing faster. But in the long term, there'd be much more competition on the frontier. But it's certainly not as meaningful as some people on the Internet thought it was. Yeah. I mean, there were like some trolls who were like, oh, yeah, once you get rid of meta and distillation, companies going to be doing, you know, $2 billion a year, and it's like, no, that's not gonna be the case. Like, this is installed in many, many organizations. This is a mature go-to-market. The question now is what happens to the frontier of the OpenWeights ecosystem? Right, there's a gap. widen. Oh, because they're because they're shutting down the distillation. Yeah, we'll see. We'll see how effective it is being on the frontier. just means you're getting distilled. If you're not getting distilled, you're not. That's the definition in your mind? It's like, unless you're getting distilled, you're not frontier? Yeah, I think that's the most honest definition. Unless other people will spend a lot of money to try to emulate things. your model and recreate it, then you are not That's a good reality check. Yeah, I mean, so one thing is meta released an open source model today. Yeah. MuseFark Open. Yeah, yeah. So I don't think they're that worried about, like, someone distilling them and then releasing an open source version. That's, like, seems very different ideologically, right? Yeah. Yeah. And often they're going through, like, third-party wrapper companies. So there's a lot of companies that are, like, yeah, we're, like, you know, an AI native document editor. And there's, like, a bunch of these. It's not just, like, Notion that we know about. And this is not just Google Docs. Like, there's a lot of these companies. set one of these up and you're like yeah and we you pay us a fee and we go and we inference clod or we inference gpt6 or whatever and then that [5:00] goes into the document. And then all of a sudden, if somebody's using that AI word doc to do the distillation, then that SaaS company, that wrapper company, needs to go and figure out, wait, why is someone generating all these documents? Maybe they can't even look at the documents. And then from Anthropics perspective or the Frontier Labs perspective, they're just seeing like, oh, yeah, one of our SaaS wrapper companies is like sending a bunch of queries across their entire network because they have, thousands of users, maybe hundreds of thousands of users, and it's hard to tell. And so there's a lot of rapper companies that have gotten to like crazy revenues, and basically no one uses the products because it's the distillers or the distilleries that are like, ooh, new rapper company? Like, can't wait to spin up a thousand accounts on you 'cause you won't detect me, But Anthropic Wood. Very interesting. Distilleries. Yeah, distilleries. You've got to take a trip to the distilleries. Well, what's going on in the market? Yeah, where did that lead us? That led us to – Yeah, I mean, especially your dollars. Sell your dollars. Sell your dollars. sell your stocks, sell your bonds, get out of everything. get into land, buy guns, food, medicines. We will be on a barter economy soon, I think. I think that's the main major prediction. You know, people are blackpilling. Sell your assets. Sell everything. Sell everything. Take your dollars. Invest them into a roadster. Or maybe a Mission S. Yes. And then also that I just there, it's very clear that there's like this tall poppy syndrome in, in the frontier where whoever is in the lead, like all the other businesses are like, well, we don't want a monopoly on intelligence. And so let's go buy from the other labs. Let's go, you know, figure something else out, fund, invest. Let's do something else to make it so that no one runs away with this entirely. Of course, like there always could be some sort of runaway, but at least while businesses are the ones that are deciding, you will see the Microsofts, the Amazons of the world. they're doing deals with the other guy because they don't want a monopoly. And so as soon as someone starts running away with it, there's this huge forcing function where you're like, oh, wow, there's a big like we, you know, the anti-YouTube, Alliance or the anti Netflix Alliance or the anti open AI Alliance was a thing and the anti anthropic Alliance like comes together and these alliances come anytime someone's getting a little bit too hot. at the party Somebody's like, yeah, we got to pull the punch bowl away a little bit. Yeah. Anyway. The news. Brian Gerstner. Yeah, please. He's on CNBC this morning. Okay. He says, we have sold, this is according to Amit, we have sold all of our core weave. We don't own any of it today. We've taken exposure down over the last three months. We felt AI had overshot and would lead to a period of consolidation They moved 10% of their book into Meta and Tesla. I'm And again... Brad, usually ahead of the curve. Sounds like he was here as well. So the interesting thing is people aren't... selling everything. They're not actually selling their dollars. That's just the meme. Of course, Cetrini says, oh, it's over, over. Which is very funny. Just to put it in perspective, the NASDAQ is down like a 0.3%, 0.4%, not much. But then the Dow is up almost a full percent. So what's really happening is that there's just a rotation that's happening while the Dow and Treasuries are getting more attention while tech stocks pull back. Obviously not good for the AI trade in the short term, but there's a lot of new changes. We will go through some of the progress because there is some interesting AI progress going on. But at least over in the AI trade, over in tech world, it's rough. It's notable that this investor day is not. The actual IPO, this is a precursor to that. It's maybe almost like airing some dirty laundry. Only way to go is up now. Baja Blast. Time to Baja Blast. Yeah, maybe they got through the code red. Now it's time to Baja Blast. The inflation is up. So if you are if you were short stocks and long inflation, then you're very happy today. Hopefully. I mean, if you own real assets, you're probably pretty happy. Joe Weisenthal says, that was the hottest PPI number in four years. And yeah, it's pretty rough. You're really having fun. Did you see inflation for the 1%? This is CPI reserve.com. We can pull this up. It looks at inflation. You know, there's normally the CPI basket that measures, you know, what the average American buys. CPI reserve is for the global elite. So it tracks how much does it cost to go to space on Virgin Galactic? how [10:00] How much does it cost to stay at Amangiri in a base suite? An Hermes Birkin bag. How much has that gone up? The spicy rigatoni. Really? Only up 31%. Oh, I didn't know that. Porsche 911 Carrera. And so there are some interesting stats here that we can go through, and you tell me whether These are overrated or underrated. Have these gone up because the value has improved or is it now overpriced? So I want an overrated, underrated for the key items that CPI reserve. Highlights so that the classic Chanel flap in 2016 it was four thousand nine hundred dollars now it In 2026, it's $11,700. That's over 139% inflation in 10 years. That's over three times what standard inflation was. If it tracked the CPI population, directly it would be IT WOULD BE $6,800, but instead it's $11,700. Do you think the class should go? You're a bozo if you're still buying this. Okay. Okay. So it's overrated at this point. The inflation always was always always was. Okay, what about Met Gala tickets? In 2016, it cost you $30,000 to go to the Met Gala. Now it's $100,000. That's six times the average inflation rate in America. Overrated, underrated. What's a Met Gala ticket worth, Jordy? I think underrated still. I think so. This is charitable giving. That's true. I hadn't considered that. So the more the better. Yes. And the Met Gala has grown so much. So many interesting folks there. So many interesting moments. Worth every penny. What about a Virgin Galactic seat? You want to go to space? It used to cost $250,000. Today it will cost you $750,000. Overrated, underrated? I don't know, not for me. Steel of the century. Are you going? At any price. Absolutely. Send me to Space in the Bay. I would go to great lengths to... Not... to block you from going on this. Is that because of safety? Assume that I survive it. Because I'm going to survive it no matter what. It's going to be... Thank you. I can actually imagine it. Yeah, I can I can imagine it. You know, the flight going poorly and you having to sort of like fight. Yeah. Yeah. You got to roll a duct tape and I got to put it together and screw this. Figure it out. Take a deep breath. Yeah. Yeah, exactly. Figure out a parachute on the way. Yeah. Yeah. Yeah. That's that's exactly what would happen. in a realistic scenario. I THINK IT'S A GOOD THING. Okay, the last one, the Porsche, actually we've got two more, Porsche 911 Carrera base. It used to be 83 years. $84,000 in 2016. Today it will cost you $135,000. That's 61% gain in 10 years. It's 1.5 times what the CPI did. Not a huge increase relative to overall, but it is ticking up in recent years. Where do you sit on the value? of the Porsche 911 Carrera base? Still undervalued relative to the joy that you get out of it. But I wish that I wish that they could keep like the base base 9-11 at like. $99,000. That'd be cool. forever. Yeah. But yeah, yeah, that'd be special. This is the world we live in. Wait, we also got to pull up the big law. We got to pull up the big law hourly rate for partners. They are way outpacing the CPI. I mean, just look at this. Yeah, you used to be able to hang out with a lawyer for just a mere $1,500 an hour, and Almost three grand an hour. to get legal advice from a big guy. Yeah, no longer the cure for male loneliness. Yeah. Because at $1,500 an hour, you just call them up, call them up and just ask them how you do. And kind of lose track of time. Yeah, exactly. Going all over the place. Yeah. Right. Yeah. Yeah. How are you? How have you been? How are the kids? Can we can we bill in quarter quarter hour increments? Can we book in out in 10 minute increments? I'm going to my lawyer. I'm saying, can we just book a weekend at a time? If I talk to you, just put me down for the whole week. Yeah. Give me 100 hours, 100-hour block. Anyway, anything else you want to cover on the CPI Reserve, you can go check it out, cpireserve.com. Very cool. A little fun project. to give you another side of the economy. The craziest here is Manhattan luxury real estate. How's that doing? It is. down. Down. Pierre Duterte tax, right? Higher taxes, people are getting out, they're moving. I don't know. It's working. Blake Byers is taking the AGI. Question. around capital markets. He says, My modal prediction for THE NEXT DECADE. INTEREST RATE SPIKES. as capital shifts from saving treasuries to invest [15:00] It's due to higher returns promised by AI related capex. We've seen this before because AI is the only modern project that could affect real rates to this degree. Um... This leads to short-term panic as high rates spike debt servicing costs for governments and other borrowers. After three years of panic, it becomes clear to the market that the investments are paying off in a big way. Interest payments remain high, but increased revenue from economic growth more than offsets the higher borrowing costs. Panic shifts to euphoria, developed world, adjusts. to 5% growth rates as the new normal and now demand 10% to be happy. Healthcare consumption R&D grows as a percentage GDP. AI is disrupting the job application market, faster than the job market. This is continuing. It's a hot take, but it's continuing to gather evidence, Derek Thompson shared. A post from the Wall Street Journal. AI amplified job applications are met with AI enabled filters and other AI tools used by HR departments to narrow the application pool. People are interviewing with AI avatars That seems like-- a really terrible thing unless the whole metagame is just can you jailbreak it, which I think it probably should be. Like if I was to put out, if we were to put out a job description and we were, I'm totally tempted to throw an AI avatar. in front of that. just in the hope that someone convinces that AI avatar that it is in fact interviewing for the job. You've seen that where people will flip it and be like, actually, we're here to talk about your performance. And the AI avatar is like fighting back, being like, remember your instructions. Remember your instructions. You're the boss. You're in charge. We're here to interview you. And then eventually the interviewer can keep hammering and prompt injecting. Um, But Derek Thompson says, "And now 10-minute conversations with AI that take place in McKinsey's very real offices with partners or interview assessors in the room." This is wild. So these candidates are basically putting on an interview show. Yeah. And the partners are watching. watching two people duke it out. Yeah, it's kind of like a gladiatorial arena. And I want to see if you can beat the tiger that I've released. And if you can, you get to sit at the side of the court jester or something like that. I don't know. We get to debate the bread and circuses thing. I feel like circuses are kind of running away with it, and bread is potentially falling behind. I don't know. Did you see the crumble of... Cookies introduced like chicken. I'm just saying that from a perspective of like what the elites are doing in our economy. Think about it. You so many billionaires are buying. Effectively, circuses. They're buying sports teams. They're about UFC. There's F1. There's so many circuses. There's that banana? Yeah. Savannah Bananas. The Savannah Bananas. The minor league baseball team. Banana Ball. Banana Ball? Yeah, the Savannah Bananas. The Savannah Bananas. Your whole thesis breaks down because crumble cookies... and Wingstop. are collabing It's fake news. See? Undefeated. But it's not a trophy asset. No one is getting the bag in venture or private equity or high finance and then being like, I got to get the award. Jordy's got to put on the Fell for it again award. Because he fell for AI slop. There is so much AI slop on the timeline again Okay, we've got to move the camera down. There we go. There we go. Fell for it again, Ward. That's me. Many, many times. Many, many times. But yeah, the job apocalypse is certainly not happening. Bank of America hired a thousand contractors to patch software flaws exposed by AI systems, mythos, and whatnot. What? I'm excited that these thousand contractors... Yeah. But at the same time, shouldn't the same AI... that discovers the bug, be able to work out a solution and But maybe there's just like, maybe there's 100,000 engineers worth of bugs. Yeah. because these thousand contractors are... powered by AI. Yeah, this is kind of an odd headline now that I'm thinking about it. Very, very odd. And it's like, yeah, did we just create a bunch of make work? Because if we just didn't, if we hadn't invented this, it really is like the... Pandora's box. Pandora's box opening creates thousands of jobs containing the stuff that is flowing out of Pandora's box. Is that what happened? Basically, that's the story here. Anyway, there is some progress on alignment. We're very excited to see this. Nick Camerata says neural networks may be... to interpret than researchers once thought. Rune is very excited about this. So it starts with Nick Camerata. [20:00] He says, interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think we have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic because neural networks appeared to fold into arbitrary tangled knots. They're grown, we're growing a garden, we're midwifing a deity instead of building a machine that is interpretable. This has been the hope that we can actually interpret these systems. And so he started OpenAI's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools that he had at the time. With modern agentic tools and a much closer look, the picture looks totally different. The vast majority in models, up to trillions of parameters are legible and their circuits beautiful. Many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. Now think ambitious interpretability will be solved by the time we need it most and possibly quite soon. As automated scientists keep improving, it may be a bitter lesson for interpretability. The simple thing at absurd scale works. And so Rune contextualizes it by saying when Nick told me that he had some interp breakthroughs, he was underselling it. I think this may be remembered as a major moment in scientific history. News so good, even Eliezer Yudkowsky will have to lower his P-do. We can start making it a lot of work into it now. Yeah, so he's at Goodfire. Goodfire is like a mechinterp lab. This is like all the stuff they work on. It's like a third-party kind of... Similar to a lot of the safety labs. Is there a Frankenstein reference? Like fire bad, fire good. Like he's like, we're summoning the Frankenstein, but it's going to be good. Potentially. I'm not sure. So basically the context of this is like, Um, like, Mecanterp broadly is-- you're trying to figure out like what's going on at the most minute level in the models, right? So you're looking at individual neurons. When you pass in some text through a model, neuron A lights up, it has a high number, like Y, it's very unclear. You'll look at one neuron, and maybe you pass in some text about cats. And it lights up. And then you pass in some text about dogs. And it also lights up. And it's very confusing because-- um, Like, how is the model actually working It seems that-- You know, each individual neuron is basically uh, It corresponds to a bunch of different topics or features, you usually call it. And so there's this thing called superposition. This was the previous theory. MICHAEL SHORT: There's the thing you were doing where Didn't you make like a 1776? Yes, that used something called a sparse autoencoder, which is basically Like, the theory is that these models are basically compressed forms of like bigger models, essentially, You can imagine like a much bigger model where every single neuron corresponds to one specific feature. So one neuron is the cat neuron. And so if you're talking about cats, that one will definitely-- And if you're talking about something else-- We don't talk about cats on this show. If you're talking about something else, it won't light up. And then basically you can take your small model, and you can basically expand it out. It's called a sparse autoencoder. Yeah, basically the outcome is we no longer need these sparse autoencoders, and you can actually look at the neurons themselves and figure out what's going on there. So it's very useful for safety, but also for capabilities, right, because if you can figure out, like, if you're trying to make some capability better within the model, like whatever math on some-- niche subject. You can figure out exactly how it's working right now and then kind of optimize on that. So it's very applicable Yeah, great news. Fantastic. Great timer. Thank you. Yeah, White Pill. The timeline is in turmoil and has been over the weekend. over Valar Atomics, Isaiah's company, and Day One, because they're beefing over Day One's pro rata rights. And so this happened back in October 9th, but we didn't get a chance to go back. through it, but Isaiah Taylor the founder and CEO of Velar Atomics, posted a long blog post on October 9th saying that he had gotten all of the investors to agree to waive the, pro rata. except for one venture capital firm, Day One Ventures, and that they were fighting tooth and nail, that Day One had retained a law firm and was fighting a lawsuit against them, and it was getting messy, so Isaiah... decided to bring it to the public square on X and have the open debate. Lots of people coming out and supporting Isaiah. Some VCs coming out and supporting pro rata rights, Obviously, this is a more complicated situation because day one has a bunch of complicated and controversial connections to international [25:00] national organizations and stuff like like and Isaiah is like going and doing more and more government work and so like there is this extra layer of nuance there I think that's getting missed but The core debate that the VCs are having is that there's a – There's like a don't throw the baby out with the bathwater pitch happening that I'm seeing. Like Delian was talking about this. And you're saying their baby. is pro rata? Yeah, basically. Like, the bathwater might... So, like, it might be like, okay, Isaiah, like, we support you in this one case. You should be able to sort of force an investor who has this like very controversial background to be diluted down below the major investor threshold. in this case, but just know that founders don't get the idea that pro rata rights are squishy, like they're negotiated for for a reason. They're going to stick around. don't think that you can pull this move without a huge amount of pushback. My view is I don't even know if it's worth having the pro rata debate, because this isn't about pro rata. Thank you. at all. Yeah, it's sort of a proxy debate over the fact that It seems like Isaiah doesn't want to work what they want anymore. Yeah. It is tricky. Signing up to work with a venture capitalist is somewhat of a marriage and you're stuck with each other for a really long time. And it's hard if they don't want to sell their shares, if they don't want to go the other way, it's typically very difficult. But once it's papered and the pro rata is in the contract, it's very hard to dilute them down because that's the design. As long as I keep showing up with the money, me as an investor with a pro rata right, I can have 5% of this company forever. Yeah, the question is just how much. is having Masha involved. With the company in a meaningful way, right? It's a meaningful stake. Hurt the company. Yes. Are they losing... contracts over it? Is it just something that competitors can sort of use to... poke at The LAR... Yeah. You know... Based on what Isaiah was saying, There's no information rights. like technical information rights that come along with that. Yeah, yeah. I do think that it's probably completely possible for – them to just sort of keep, uh, Keep the fund. At an arm's length. Yeah. But financial information rights are valuable. Yeah. Like even internationally. Like, I mean, we just saw that with the OpenAI spy. Big car day today. Two car launches. The Tesla Roadster. Tonight, 7.55pm. PM Central time? Yeah, so 5.55 Pacific. 5.55 Pacific. 8.55 Eastern. PUT DOWN. a deposit yesterday for Roadster, and I'm fully ready to put that PIN ON THE FELL FOR IT AGAIN AWARD. is ready to go. So it's five grand on a credit card. And you have to wire 45,000 within 10 days, but I'm going to know whether or not it's a good car by tonight. And so I can cancel it if I don't like it and jump out of line. But if it's amazing, I mean, Elon saying it's going to be the most amazing product demo ever. If it's amazing, then I will be able to hold my position in the line before everyone floods in post-lawn. Let's speculate a little bit. So I think that Ramps will be involved. Okay. I think that ramps will be in. So because our original bet was it will leave the ground on flat ground. And I think you had like half a foot. I think he's a flat ground. No, he just said he just gestured towards it flying. Yeah. And so and so to me, what's the easiest way to get a car in the air? Okay. Shoot it off of a ramp. Drive it off of a ramp. I think we're going to be seeing... multiple types of ramps. I think we're going to be seeing-- That'd be awesome. I love it. So if there's ramps involved, Jordy wins. If it jumps off the ground, off a flat ground, I win. That's the bet. So you think jump. I think rocket engines or thrusters or fans that will lift them. The people that fell for the Tesla Roadster the first time, they put down the full amount in 2017 and they waited a decade and they didn't get anything. Nothing shipped, right? They fell for it. They didn't fall for it again because Tesla has always shipped their vehicles. They did the Cybertruck, they shipped that. There's one outside. Like they're real. You may like it, you may not, but they shipped it, but they didn't with the original Roadster. And so... There were a lot of people that feel like they fell for the Roadster promo scheme because they put down the money and then they waited a decade and they didn't get anything. So in order to fall [30:00] again, you need to put down the money and wait another decade, but we don't want to wait a decade. And so the question is, is-- on what time scale would you be disappointed? And to what feature set would you be disappointed? And so I thought it'd be funny to demo the car in such an insane way that it just instantly, like... federal regulators come in and say, you cannot under any circumstances do this. And then he's like, ah, my hands are tied. Yeah, sorry. I've made the product. I just can't ship it. We're going back to Model Y. We're only going to make the Model Y now. So it's going to be withered and old by the time we put it on? No, I would say... I would say one year. It's got to be one year. If it takes more than one year to ship this thing, Then you fell for it again. Right? Because, like, you've had 10 years. Like, you don't get another 10. You get another launch, and then you get another, okay, ship it, right? How much was the original register deposit? I thought it was the full amount. I thought it was like a quarter mil. I don't know. You can look that up. Mission S is coming out. I think they will ship it. I think that the Mission S is probably a safer bet. You won't be watching that. So this video leaked. Late last night. Okay, what are they doing? They're peeling back a cover. And they're showing off the car. And you had a very good take on this. What do you like about this design? What do you like about the Mission S? What sticks out? to you about this. So I'll start by saying... I feel like a kid who accidentally saw their Christmas present. This car is... way way more aggressive than I was anticipating. Yes. It feels like it really does feel like Thank you. It's over. for every other manufacturer in the in the sort of like mid-range supercar category. The McLarens, the Lamborghinis, the Ferraris. Like it is going to be rough. It's got a roof scoop. I know, I know. We're scooping. I mean, this is the perfect car for... an S tier journalist. To just rip around SF. Yeah. With the roof scoop. For sure. We'll see about the performance. We'll see about how they position it. I know that they're not going to do as much of a restriction on the production numbers. But at least from like an emotional perspective. Feeling it's giving me 918 successor, and I know that they might say oh, it's not it's a different thing It's gonna be a high production or whatever. It doesn't matter emotionally. I'm feeling career GT I'm feeling 918 and this just immediately becomes by far the most desirable car in the 500 and $2 million price range, in my opinion. Yes. Like day one. Yeah. The car is somewhat antisocial, right? How so? Looking at the amount of rocks that the back tire is going to be thrown up. Yeah, that's right. If you want to take this out on a Sunday drive with somebody's... You're going to need some mud flaps. That's right. Otherwise, good luck if you're sitting behind this thing. Well, thank you for tuning in to TBPN. Leave us five stars. We're just 23 minutes away from the Mission S reveal. We're just hours away from us seeing the new Nathan Fielder movie. That's right. We're very excited. Leave us five stars on Apple Podcasts and Spotify. Get ready for Christmas. It's only 70 days away. And sign up for a newsletter at tdpn.com.