Crypto Going Bunker Mode, AI Mathpocalypse, Starbucks Explores Chipotle Takeover | Diet TBPN

Summary · 4 min read instead of 32 min

Coogan walks through warnings from Vitalik Buterin and cryptographer Matthew Green that AI-accelerated math could undermine public key cryptography, pushing crypto into "bunker mode." He argues the threat is harder to plan for than quantum computing because its shape is unknown, and that it could force libertarian crypto users to weigh limits on open source models. Hays contends that published math breakthroughs are the visible part of a larger wave, since commercially valuable discoveries are less likely to be shared. The hosts also cover John Urschel's AI-assisted proof, a report that Starbucks explored buying Chipotle, and how inconsistent leaks of OpenAI and Anthropic revenue now move public markets.

Minute by minute

1. 0:10

   AI math may threaten cryptography. Coogan reports that Vitalik Buterin and others warn AI-driven math could threaten blockchain security, with crypto going "bunker mode." He cites Johns Hopkins cryptographer Matthew Green saying "I think we might lose public key cryptography." Coogan notes cryptography was absent from hundreds of published AI math papers, suggesting labs either skipped it or are withholding results until fixes exist.

2. 2:50

   Breaking Bitcoin would destroy its value. Coogan argues no one would rationally steal all Bitcoin, since breaking every wallet would erode trust and make the coins worthless, so the risk is more about long-tail targets. Hays counters that some people would do it anyway, and Coogan agrees a Joker-style AI agent swarm would be a bad situation.

3. 3:37

   Open models trail closed ones by months. Coogan uses roughly six months as the gap for open source models to catch closed ones. He notes Anthropic released Mythos Preview on 04/07/2026, and CrowdStrike reported on October 7 that a 26-year-old in China used an open source Chinese AI agent to attack South Korean banks. He says the breach exposed client data such as names and addresses rather than money, as banks have redundancies like rollbacks and tape storage.

4. 6:04

   An unknown threat is harder than quantum. Coogan argues quantum risk was formalized for years, with a known target in Shor's algorithm, while AI math may yield some unknown algorithm running on commodity hardware. He notes Bitcoin was down only 3% on the day and up 3% for the month. Buterin advised against scrambling to move funds but warned that lattices could take serious hits from the next two years of AI math, and recommended fresh addresses.

5. 8:41

   Crypto faces a regulation dilemma. Coogan argues the AI safety push to register clusters and license inference would counter open source attacks, but clashes with crypto's libertarian, small-government outlook. Hays separates cryptography attacks from bio risk, saying digital hacking is something millions have long done. Coogan adds that a rogue agent able to crack wallets could amass capital to buy compute and hire people.

6. 11:10

   Crypto can govern itself. Tyler argues the crypto industry is fairly good at self-governance and can change protocols, citing Bitcoin Cash as a fork. He mentions a Bitcoin Policy Institute open letter calling for stronger defenses and heavy inference spending, and says most people oppose limiting open source models.

7. 11:47

   Math breakthroughs are the visible tip. Hays argues math is a field where results are shared because there is little immediate commercial value. For every math breakthrough, he says, assume others in commercially useful fields are being kept private, including by people using frontier models in their own fields. He expects an explosion of progress toward the end of the year.

8. 13:38

   Little incentive to cheat at online chess. Coogan calls OpenAI's release of hundreds of new proofs the Mathpocalypse and raises online chess cheating. Tyler says an 800-rated player making 2,000-rated moves is easy to spot. Coogan predicts models will soon mimic human play at a chosen rating, but argues there is little reason to cheat since anyone can play Stockfish and nothing is won.

9. 16:18

   Former NFL lineman publishes AI-assisted proof. Coogan reads a Wall Street Journal story on John Urschel, who left the Baltimore Ravens for MIT and is now an assistant professor there. After reading OpenAI's proofs, Urschel published a paper titled On the Growth Factor of Random Matrices proving a decades-old conjecture. His paper credits OpenAI models, and he compares math to chess, saying AI does not ruin his search for the why.

10. 20:44

    AI doesn't spoil curiosity. Coogan agrees with Urschel that "AI doesn't ruin the why." He says that despite years of making YouTube documentaries about Silicon Valley companies, ChatGPT knowing more history has helped him, because he looks things up constantly and enjoys the journey. Hays adds that people benefit from holding knowledge themselves.

11. 21:48

    Starbucks explored a Chipotle takeover. Coogan reports a Financial Times scoop that Starbucks worked with advisers on a takeover proposal for Chipotle, valued at $41,000,000,000, which would be the largest restaurant acquisition ever. Its CEO left Chipotle in August 2024, and Chipotle has since lost half its value. Hays attributes the drop to the broader fast casual run ending, not the departure.

12. 24:33

    PepsiCo admits soda trouble. Coogan cites PepsiCo's CEO saying the company does not feel good about its beverage business and is urgently working to improve sodas. He floats an alcoholic Pepsi, noting Topo Chico made an alcoholic seltzer, and Hays jokes about a Pepsi Pro Max Ultra.

13. 25:40

    AI lab revenue leaks don't match. Coogan notes one leak put OpenAI near 70,000,000,000 in revenue and another near 50,000,000,000. He explains that on AWS, OpenAI keeps the sale as revenue while Anthropic books the share paid to Amazon as a cost, inflating its gross figure, and says attempts to compare them were sloppy. Hays says the labs are now large enough that public markets depend on their metrics.

14. 27:31

    IPOs would spread gains and curb manipulation. Coogan argues IPOs align with better governance and would put lab gains into retirement accounts through ETFs. He notes that leaks about private companies like Stripe never moved public stocks, but OpenAI and Anthropic have proxies like SpaceX, Oracle and SoftBank. That makes leaking inflated numbers to trade on a new possibility, which he notes would be securities fraud.

Published Oct 9, 2026.