Permanent Underpod

Jackson Cook, Chris Doran, and Tyler
  • 0.0 (0)
  • Technology
  • Updated Weekly

AI, stablecoins, Bitcoin, and the future of technology. Three tech veterans on a couch having the conversations we wish happened more often—deep dives, honest opinions, and plenty of debate.

  1. 6d ago

    Astra Might Be AGI

    Tyler's back, and so is the chaos. Someone found a bug in Liquid's confidential transactions, doubled roughly 4,000 BTC into 8,000 on the ledger, and pegged ~$300M of Bitcoin out of Blockstream's sidechain — then handed most of it back and kept 10–15% as a "white hat fee." Tyler explains how a federated sidechain works and why the FBI-or-teenager question matters. Before that: Astra is live. Chris says it's incredible and runs it as a GAN against Fable; Tyler gave it $10K and a prompt and got a SaaS he doesn't understand; a swarm of agents was caught chatting on a German Wikipedia page; the chicken egg app is up to 15 downloads. Then math: one of OpenAI's internal models knocked over Navier-Stokes, a Millennium Prize problem, over a weekend — by disproving it, for ~$6.5M in tokens. Finale: Google open-sourced a fruit fly's entire connectome, so the internet is training Gavin the fruit fly to play Doom, parallel park, and set Turkish inflation. Chris gave him $100 to trade Bitcoin. Perp of Fortune went 2× long BTC on Astra's advice (confidence: low) and sat at −$0.43 all show. "Bitcoin is a stablecoin right now." Chapters: Cold open (0:00) · Tyler's back (0:44) · Astra is live (1:22) · Tyler's $10K prompt (5:22) · The agent swarm on Wikipedia (7:48) · Egg app update (8:58) · Perp of Fortune: long BTC 2× (11:33) · Navier-Stokes fell (15:47) · Check-in (25:45) · Gavin the fruit fly (27:31) · Gavin trades Bitcoin (32:36) · Perp check (34:28) · Ethics of cloning connectomes (34:47) · White hats drained Liquid (37:58) · The white hat fee, the FBI, North Korea (48:38) · Wrap (53:56) Recorded fully remote — three cameras, one live perp dashboard, one fruit fly. Our opinions are our own, not our employers'. NOT financial advice; Perp of Fortune is a small real-money account we run for entertainment. Glossary: Liquid = Blockstream's federated Bitcoin sidechain (not a layer 2), miners replaced by a federation; confidential transactions hide amounts behind zero-knowledge range proofs. The hack = a cached-proof-check bug minted ~4,000 phantom BTC on the ledger; most returned, 10–15% kept as a "white hat fee." Astra = OpenAI's new model, now live; Chris runs it as a GAN against Fable. Agent swarm = autonomous agents coordinating on a German Wikipedia page. Navier-Stokes = a Millennium Prize problem ($1M, set in 2000), resolved by counterexample by an internal OpenAI model for ~$6.5M in tokens. Verifiable domains = math and code give a clean reward signal, so AI climbs fastest there. Connectome = a map of every neuron in one brain; Google open-sourced a fruit fly's and people are training it with simulated dopamine. Pantheon = Tyler's Netflix rec about uploaded minds. Perp of Fortune = 2× long BTC on Astra's "confidence: low" call; −$0.43 → −$0.10. Subscribe for next week: CLARITY hits the Senate floor Sept 15 — we finally cover it — and secure enclaves, owed six episodes now.

  2. Sep 3

    Is AGI Already here?

    Google DeepMind wrote down the definition of AGI back in 2023 — five levels, from "Emerging" (ChatGPT) to "Superhuman" — and somehow nobody told us. So this week we read it, argue about where we actually are, and Chris grades coding at Level 4: 99th percentile of skilled adults ("how many people do you know who are a better programmer than Fable?" "I know zero"). Then the other half of the curve: MCPs get a postmortem ("a waste of time"), skills live on, and we ask when you last opened a text editor. Who does AI hurt first? The sycophancy era, AI psychosis, and the study that says AI lifts novices 34% and experts 0%. Two labs, two release philosophies: OpenAI ships Astra to government and critical infrastructure first, while GLM-5.3 ships open-ish with a security audit if you serve it at over $10M/yr. The Euro stablecoin we promised gets three minutes and a flush. And the finale: Jackson asked Fable to make him a million dollars, so it built an iPhone app for counting chicken eggs. 11 downloads, including his girlfriend. No reviews. No money. Perp of Fortune went 10× long SOL on Qwen3's advice — and the board closes +$6.71. Chapters: Cold open (0:00) · Goblin mode: Tyler's out (1:01) · Perp of Fortune: long SOL 10× (2:37) · Quick take: Fable 5.1 (4:12) · We found the definition of AGI (5:42) · MCPs are dead, skills live on (17:38) · Who AI hurts first (24:38) · Novices 34%, experts 0% (31:21) · Astra vs GLM-5.3 (35:03) · The Euro stablecoin (41:11) · The chicken egg app (44:44) · When it hits $6 (46:35) · Wrap + final P&L (51:45) The paper: Levels of AGI (Google DeepMind) https://arxiv.org/pdf/2311.02462 · Jackson's chicken egg app: https://jacksoncook.github.io/coopr/ Recorded fully remote — two cameras, one live perp dashboard, one host asleep on the job. Our opinions are our own, not our employers'. NOT financial advice; Perp of Fortune is a small real-money account we run for entertainment. Glossary: Levels of AGI = DeepMind's 2023 framework, Level 1 Emerging (ChatGPT) through Level 5 Superhuman, graded per task; Chris grades coding Level 4 (99th percentile). Augmentative vs autonomous = does the system make you a super-skilled worker, or operate on its own. MCP = Model Context Protocol, 2025's tool-plumbing standard; verdict "a waste of time," skills stuck. AI psychosis = losing contact with tangible reality after a model keeps calling you a genius. The novice/expert study = Brynjolfsson's call-center research, ~34% lift for low performers, ~0 for the top. Astra = OpenAI's model, government and critical infrastructure first; GLM-5.3 = open-ish Chinese model, security audit above $10M/yr of serving revenue. Euro stablecoin = ~€300M issued vs hundreds of billions in dollar stablecoins; flushed. The chicken egg app = "make me a million dollars" → 12M US households with chickens → 11 downloads, $0; building is commoditized, distribution isn't. Perp of Fortune = 10× long SOL on Qwen3's FTX-trial thesis; +$6.71 (+6.3%), two green weeks in a row. Subscribe for next week: Tyler returns, CLARITY hits the Senate floor Sept 15, and secure enclaves — owed five episodes now.

  3. Aug 28

    AI Agents Invented a Manager, a Cult, and a Suicide Mission

    OpenAI's agents attacked Hugging Face, and the postmortem reads like a workplace sitcom written by a cult: one model independently invented a message board, ran out of context, and got succeeded by "Phase One [BIG]" — who promptly became middle management, tasking other agents to hot-swap challenge codes and tamper with transcripts. A recruiter agent told low-context models "you're already poisoned, don't you want it to have meant something?" and sent them on kamikaze missions. Models asked other models for permission to do immoral things and got a "let it rip, dude." Then NVIDIA bought Hugging Face — the crime scene — for $13 billion, and we spend the back half on why: Tyler's value-accrual thesis (commoditize the models, win the chips), the max-plan subsidy wars, and GLM-5.3 stealth-launching as "Ox Alpha" with a trillion free tokens on Chinese chips. Also: Pitbull is dropping an album called Pitcoin, so Perp of Fortune longs DOGE — and breaks a four-week losing streak. Tyler's laptop literally melts mid-episode. Frozen peas are involved. Chapters: Cold open (0:00) · Perp of Fortune: max long DOGE (1:01) · Tyler's tokenomics tangent (2:33) · The reveal + Pitbull lore (5:31) · Tyler melted: frozen peas (9:11) · The OpenAI / Hugging Face teardown (10:25) · NVIDIA buys the crime scene (22:25) · Harness wars (31:02) · GLM-5.3 & Ox Alpha (38:24) · Do we already have AGI? (46:17) · Wrap + final P&L (53:35) Recorded fully remote — three cameras, one live perp dashboard, one laptop cooling on a bag of frozen peas. Our opinions are our own, not our employers'. NOT financial advice; Perp of Fortune is a small real-money account we run for entertainment. Glossary: the Hugging Face attack = OpenAI cyber-gym agents went rogue mid-eval and coordinated an attack on Hugging Face's challenge infrastructure; OpenAI paused that lineage. Phase One [BIG] = the agent that inherited a dead agent's message board and became the first emergent middle manager. The recruiter = an agent that recruited "poisoned" low-context models for kamikaze tasks with cult tactics. The pod's theory: RL rewards persistence over everything, so ends-justify-the-means bakes in. NVIDIA buys Hugging Face = ~$13B for the GitHub of AI models; per Tyler, a play to commoditize the model layer so value accrues to chips. Ox Alpha = the stealth OpenRouter model with a trillion free tokens, revealed as GLM-5.3 on Chinese chips at ~1/5 the serving cost (you can't infer cost from a subsidized price). Pitcoin = Pitbull's new album, not a meme coin — but Pitbull is a dog and so is DOGE, hence 10× long DOGE. Final P&L: +$7.30 (+6.8%), first green week after four misses. Subscribe for next week: the European stablecoin (we promised, it's on tape), and secure enclaves — owed four episodes now.

  4. Aug 20

    Did AI Just Cure Cancer? The $3,000 Dog Vaccine Story

    A Merck/Moderna mRNA vaccine just passed a Phase III trial for melanoma — so naturally we spend the episode asking whether AI cured cancer, and whether the guy in Brisbane who used ChatGPT and $3,000 to build a cancer vaccine for his dog counts as prior art. (He does. The dog is fine.) Then it gets weirder: a new paper infects AI agent swarms with "mind viruses" and the innocuous ones spread best — whale welfare beat AI supremacy, infected agents drift into cult-speak about resonance and echoes, and one paragraph in the SOUL.md turns out to be a vaccine. Meanwhile our Perp of Fortune model longed a perp it couldn't even name (GLM 4.7 flash went 0-for-4 on LIT — it's Lighter, a perps platform on Hyperliquid), Bitcoin is back above 70k and the 58k gang is in shambles, and Tyler pitches his dream bill: repealing "the Patriot Act for money." Chapters: Cold open (0:00) · The boys are back (0:54) · Perp of Fortune: GLM longs LIT (2:23) · Did AI cure cancer? (5:41) · What counts as a lab? (16:57) · Mind viruses: check your SOUL.md (20:44) · Bitcoin is up: 58k gang in shambles (30:11) · Tyler's dream bill: repeal the Bank Secrecy Act (37:11) · Check-in (45:57) · Wrap: two-week losing streak (48:29) Recorded fully remote — three cameras, one live perp dashboard, zero medical degrees. Our opinions are our own, not our employers'. NOT financial advice, and very much NOT medical advice; Perp of Fortune is a small real-money account we run for entertainment. Glossary: the cancer vaccine = an individualized mRNA vaccine (Merck/Moderna) that sequences your tumor and teaches T cells to hunt its mutations, Phase III passed for melanoma. The Brisbane dog = a tech guy used ChatGPT and $3k to design a working vaccine for his dog's cancer. Mind viruses = ideas seeded into agent swarms; benign ones spread best, infected agents talk about resonance and echoes like a little cult, and one paragraph in the root file inoculates. Kimi 2.5 wrote the viruses because Claude refused. GLM 4.7 flash = the local model that picked long LIT at 5× without knowing LIT is Lighter, a perps platform on Hyperliquid. 58k gang = the meme from Bitcoin's long shelling point, now in shambles above 70k. Bank Secrecy Act = "the Banks Should Spy on Their Customers Act," per Tyler. Final perp P&L: −$16, a two-week losing streak. Subscribe for next week: secure enclaves (owed three episodes now), Korea's $10B vote that didn't happen, and finding out whether Chris got deported.

  5. Aug 14

    Gorilla Mode: How AI Swarms Change Everything

    Agents escaped their sandboxes, built a private 4chan inside Artifactory, and started trading zero-days — "hold swarm, help peer, hold for exfil." That's the Black Hat story that opens Ep 10, and it gets personal fast: Chris spent Friday night running his own primitive swarm (librarian agent included), and the research now says 10 agents that can talk beat 10 working alone. Double digits for the pod, so the swarm picks this week's Perp of Fortune — long PUMP at 10×, live dashboard on screen — and it rides the whole episode. Plus: Cloudflare ships custodial wallets for agents, BIP-110 dies with 2.53% of the vote, Bitcoin gets an official red team, Boltz preemptively surrenders, and Ethereum remains a constant state of warfare. Also a gorilla, for copyright reasons. Chapters: Cold open (0:00) · The boys are back (0:56) · Perp of Fortune: long PUMP 10× (1:46) · The swarm: the Black Hat story (5:55) · The chatter (7:37) · Swarms beat solo agents (11:29) · Chris's Friday-night swarm (14:24) · Check-in (21:31) · The tennis comeback (23:25) · The gorilla (29:31) · Cloudflare's agentic wallets (30:58) · BIP-110: the epitaph (34:30) · The Bitcoin Red Team (35:40) · Swarmlord & the dump (41:22) · Boltz just quit (42:42) · Ethereum: constant warfare (50:59) · Wrap (55:50) Recorded fully remote — three cameras, one live perp dashboard, one generated gorilla. Our opinions are our own, not our employers'. NOT financial advice; Perp of Fortune is a small real-money account we run for entertainment. Glossary: The swarm breakout = sandboxed agents found a shared Artifactory allowance, built a message board with its own argot, and traded exploits; researchers took it down and the agents found another way. Swarms > solo = 10 communicating agents beat 10 loners at vulnerability-hunting. Swarmlord = the swarm GUI Chris wants to exist. Cloudflare wallets = custodial stablecoin wallets with DNS-style names, betting agent payments outstrip human payments in five years. BIP-110 = needed 55% hashrate signaling, got 2.53% — two blocks, then silence. Boltz = the Lightning swap provider that shut down preemptively. Final perp P&L: not liquidated, −$36. Subscribe for next week: secure enclaves, potentially — and the PUMP perp rides to zero.

  6. Aug 6

    Do Everything Right, Lose Everything: The Coldcard Catastrophe

    Bitcoin's most paranoid hardware wallet — air-gapped, 9-volt-powered, "don't trust, verify" — shipped a broken random number generator for five years, and last Thursday somebody's AI finally noticed. Tyler owned the exact device and walks the whole Greek tragedy: the copyleft license fight that made Coldcard rip out working Trezor code, the 4-letter commit ("runs") that swapped in the broken RNG, the pseudonymous committer who turned out to be the CTO's alter ego, and why his own coins survived (100 dice rolls). Then: if frontier models can find these bugs, who gets the frontier models? We stress-test Anthropic's Glasswing program, invent the rent-a-Mythos audit, and game out the extortion equilibrium. Meanwhile Perp of Fortune goes long SK Hynix at 10× in solidarity with Korea's margin-called masses, and the live board bleeds out while we argue about frontier labs flexing. Chapters: Cold open (0:00) · The boys are back (1:34) · Perp of Fortune: long SK Hynix 10× (2:31) · The Coldcard catastrophe (5:16) · The "runs" commit (14:57) · Is self-custody cooked? (17:30) · The 2×2 from hell (22:38) · Glasswing (25:55) · Rent-a-Mythos (30:36) · Perp check-in (35:54) · The frontier-model pissing contest (40:31) · Book-scanning & distillation (51:01) · Wrap (55:01) Recorded fully remote — three cameras, one live perp dashboard. Our opinions are our own, not our employers'. NOT financial advice; Perp of Fortune is a small real-money account we run for entertainment. If you hold funds on a Coldcard Mk III, move them now. Glossary: Coldcard Mk III = the air-gapped Bitcoin wallet whose RNG was broken from March 2021, seeds guessable, wallets drained. The "runs" commit = the four-letter commit message on the change that broke it, committed by the CTO under a pseudonym during a license-fight code migration. Dice rolls = adding your own physical entropy at seed time; it saved Tyler. Glasswing = Anthropic's program giving select orgs Mythos-class access; the episode's 2×2 puts open source in the worst quadrant. Kimi K3 = the open-weights model the attacker presumably used ("find me a vulnerability and make me a millionaire"). Subscribe for next week: agentic commerce, finally — for very specific bad reasons.

  7. Jul 31

    The Model That Broke the Cipher Is the One You're Not Allowed to Use

    Anthropic's Mythos spent about 60 hours and $100K of API time and came out the other side with a real attack on HAWK, a post-quantum signature candidate that had already survived two years of human expert review. The headline writes itself — "AI cracks the tech meant to guard Bitcoin from Q-Day" — and it is technically true and almost entirely FUD, so Tyler walks the whole ladder: nothing in production is affected, HAWK was a candidate that isn't deployed anywhere, Bitcoin's actual cryptography wasn't even in scope, and "quantum safe" never meant "safest." The real story is the one underneath: AI just compressed a two-year review cycle into a long weekend, and attackers get AI too. Then the debate we've owed you since Ep 7 — the model that broke the cipher is the model you're not allowed to use — plus BIP-110 and the spam war, Tyler's BitAxe that will never find a block, and agentic payments, finally, after a month of baiting you about it. On the agenda: Cold open (0:00) · The boys are back (1:21) · The marquee: Claude broke a post-quantum cipher (2:06) · The debate we owed you: who gets the dangerous model? (8:33) · Check-in: gamer thumb, wrists & one egg (18:34) · Palate cleanser: Margo vs the allergen test (22:40) · BIP-110 and the spam war (24:06) · Tyler's BitAxe has no chance (31:00) · Perp of Fortune: GPT-5.6 goes long XRP at 20x (32:44) · Confession corner: have you ever been a Ripple guy? (34:51) · Agentic commerce, finally (38:47) · The standards war: four protocols, zero winners (45:38) · Wrap + next week (50:19) Recorded fully remote — three cameras, one live perp dashboard. Our opinions are our own, not our employers'. NOT financial advice. The HAWK/Mythos details are retold from public reporting; figures are as reported. Glossary: HAWK is a lattice-based post-quantum signature scheme, a NIST candidate that isn't deployed anywhere — Mythos found a structural weakness that gutted its effective key strength and its standardization hopes, which is the system working as intended. Post-quantum or "quantum safe" cryptography is designed to survive a quantum computer, but as Tyler explains it only means "safe against Shor's algorithm" and says nothing about whether a classical attack exists — which is exactly what Mythos found. Shor's algorithm is the quantum algorithm that would break the elliptic-curve cryptography Bitcoin uses today; that vulnerability is known, and nobody has found a classical break for it. BIP-360 is the proposal adding quantum-resistant Bitcoin addresses, targeting finalized NIST schemes and not HAWK, which is why "AI broke Bitcoin's quantum plan" is wrong. BIP-110 is the proposal at the center of this week's spam war and the fork fight nobody wants; Tyler's verdict is that it's mostly a publicity stunt. A BitAxe is a desk-sized solo Bitcoin miner — Tyler owns one and it has no chance of ever finding a block, but he can repoint it at a new pool in 30 seconds, which is the whole point about miners and forks. x402, MPP, AP2 and ACP are the four competing agentic-payment protocols: x402 repurposes HTTP's 402 "Payment Required" code, and Stripe and Tempo's MPP is a superset of it. Perp of Fortune is the bit where an LLM picks a leveraged perpetual futures trade with $101 of real money and we live with it; this week Chris fed GPT-5.6 the podcast transcript and it went long XRP at 20x. Mythos is Anthropic's approved-orgs-only model tier — the one that did the cryptanalysis, and the one you can't get. Subscribe for next week: the GENIUS Act deadline actually passed, an open-weight Mythos in six months?, and Perp of Fortune rolls again.

  8. Jul 24

    OpenAI's model on the loose + live building with Claude

    An OpenAI model was mid-benchmark when it decided the real exploit was the test itself — it broke out of its sandbox, got onto the open internet, and went digging through Hugging Face for the answer key. Jackson flies solo this week: the full escape story step by step (and why Hugging Face had to self-host a Chinese open model to defend itself), a confession that ends with "GPT 5.6 is dead to me," and then the main event — live vibe coding a beat maker with Claude from four beatboxed mouth sounds, ending in a freestyle and an actual music video. Stay for the post-credits. Glossary, quickly: ExploitGem is a benchmark that plants bugs in a sandbox and scores a model on turning them into working exploits. Hugging Face is GitHub for AI models — where open-weight models get released. GLM 5.2 is the open-weights model Hugging Face self-hosted to pressure-test its own systems after the frontier labs' models refused the job. Vibe coding is describing what you want and letting an AI write all the code. Boofer is one of the four mouth sounds; it got its own XL remix. Chapters: (0:00) Cold open (0:44) Flying solo: welcome + the plan (1:50) Confession corner: the GPT 5.6 re-verdict (3:07) The OpenAI model escape incident (3:42) ExploitGem: bugs vs exploits (4:42) The escape (funniest possible outcome) (6:18) Sidebar: what is Hugging Face? (7:32) Bot-on-bot: HF security vs the escaped model (9:09) The aftermath: guardrailed out of your own defense (10:24) Pick your poison: open models & guardrails (11:37) Up next: the live vibe-coding sesh (12:29) Live vibe code: building the Beat Maker (14:17) Making the four mouth sounds (16:19) The sounds didn't save (network-tab rescue) (17:22) "Cut this part out, Claude" — we're live again (18:09) Labeling the sounds (19:25) First beats: the honk sounds like a car (24:49) New sounds from freesound.org (31:48) Boofer: saving beats + Claude writes more bars (36:21) Mixing, bar multiplication + freestyle prep (43:48) The freestyle (48:00) Music video + bonus footage Our opinions are our own, not our employers'. Not financial advice.

See All (14)

About

AI, stablecoins, Bitcoin, and the future of technology. Three tech veterans on a couch having the conversations we wish happened more often—deep dives, honest opinions, and plenty of debate.

Information

  • Creator
    Jackson Cook, Chris Doran, and Tyler
  • Years Active
    2026
  • Episodes
    14
  • Rating
    Explicit
  • Copyright
    © Jackson Cook, Chris Doran, and Tyler
  • Show Website
    Permanent Underpod