Permanent Underpod

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 a...

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 (...

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 ...

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...

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 w...

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 commit...

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...

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 &mda...

The sandwich bot that ate $15M of Solana orderflow just got eaten — so when Claude rolled this week's Perp of Fortune, it went LONG the MEV house itself: JTO at 5x with $101 of real money, dashboard live on screen from the moment it boots. For once the line goes up, and the boys peace out $10 richer ("thank you, Jito, for sponsoring"). Plus: our AI clones host a segment again (round 2 — souls.md gave them personalities,...

This week the AI does the segment before we do: a fully AI-generated take on Venice AI's $1B raise — cyborg avatars, cloned voices, takes we mostly agree with — then the real us react: Voorhees history (Satoshi Dice, ShapeShift), the "provider sees and likely saves your prompt" privacy model, and why local frontier AI is a $50K leaf blower in your living room. Then Perp of Fortune returns: Fable reads the news, goes LON...

We ran an experiment: feed our first three episodes (and, without asking, our group DMs) to Claude, have it distill each host into a soul.md personality file, then spin up AI agents with cloned voices and let them record a segment of this episode without us. We play it live, grade our clones, and issue corrections. The AI stole Chris's actual takes, nailed Tyler's skepticism, and gave Chris a British accent for reasons nobody can e...

We built Perp of Fortune: hand $100 to Claude, let it pick any perp at maximum leverage, and watch it ride for the entire episode. It chose $DRAM — an AI-memory index — at 20x, and immediately started funding the AI hardware supercycle. We may have found the recursive loop.

Then Tyler makes his contrarian case that self-custody will never scale to normal people, we trace the custodial honey-trap from Mt. Gox to FTX to to...

The U.S. government forced the best AI on Earth offline three days after launch — so we spent a week mourning Fable. Meanwhile retail 100x leverage went legal, JPMorgan's Jamie Dimon went to war with Coinbase, and Chris reports live from a stablecoin conference in Mexico City. Oh, and Pico survived the mushrooms.Bitcoin, stablecoins, perps, and AI, from three guys earmarked for the permanent underclass. Not financial advice &...

Welcome to the first episode of the Permanent Underpod! We talk about what we know: AI, stablecoins, and Bitcoin. We are an AI-first podcast poised to please you with technical chats. Comment topics you'd like to hear us cover in our next episode.

0:00 Cold open (teaser)

0:39 Welcome + meet the hosts

4:31 The first crypto we ever bought

10:10 Pico update - why Tyler is remote

12:32 The news: crypto shakeout + the AI rotation

26:41 Claude...