Podcasting 2.0 July 17th 2026 Episode 265 - "Jihobbyist" Slop or Not! Lightning or Frightening! Adam and Dave go deep with the Boardroom after a 2 week break ShowNotes ------------------------------------------------------------------------------------------------------------------------------------- 00 - PODPING GOSSIP-WRITER → PRODUCTION + INDEX PLUMBING ------------------------------------------------------------------------------------------------------------------------------------- 01 - THE OPEN INDEX TURNS UP INSIDE AN AI SCRAPER (SUNO) <b>@js flagged it:</b> a hack at Suno reveals the AI music generator used <b>PodcastIndex</b> to do its targeting — "with the help of an online tool called PodcastIndex, Suno identified 420,000 different podcasts that had at least five 30-minute episodes and sought to download roughly 1 million hours of podcasts." <b>Board question:</b> the Index exists to be "free, for any use" — and here "any use" is an AI company bulk-harvesting a million hours to train on. Is that the deal working as designed, or does anything change? A live test of the open-index ethos. Ties straight to the on-device/namespace fight below and to the spam/AI-slop lane — the same extraction pressure from three directions. Hack reveals Suno scraped via PodcastIndex (404 Media) ------------------------------------------------------------------------------------------------------------------------------------- 02 - ON-DEVICE AI vs NAMESPACE TAGS — "DIDN'T YOU INVENT THE NAMESPACE?" <b>Your own post set it off:</b> Adam — Apple in talks to run a startup's compressed AI models on-device (PrisMML) — "More fodder for on-device features vs new tags." The core board fight of the week. <b>🎧 CLIPS TO PLAY — a 3-part PNWR sequence (Cridland and Sethi discussing your position; your own SOT is NOT in their show, so play + respond):</b> <b>1. The jab (44.5s):</b> recap of the Car Curious "podcast annotations" (≈ Apple timed links) debate — Nathan Gathright: LLM-from-transcript data stays the purview of players, not publishers; Sam Sethi: "Adam Curry also came in… he sees LLMs as the way for on-device abstraction of data rather than tags. And I'm like — didn't you come up with the namespace?" File: <i>01 - Cridland Sethi - Adam annotations, didnt you invent the namespace - PNWR.mp3</i> <b>2. Sethi's thesis (55s):</b> his actual position — fine, copy Apple, let LLMs do transcripts as a FALLBACK "if you can't be bothered," but if a creator wants to put real data in the feed, respect what the creator wants FIRST. "An LLM is a fancy autocomplete system… I'd much rather the creator is given the opportunity to do something good first." The publisher-vs-LLM argument, self-contained. File: <i>02 - Sethi - an LLM is a fancy autocomplete, creator first - PNWR.mp3</i> <b>3. The Spurlock correction (36s):</b> Sam apologizes for mischaracterizing @js — "I don't think we should leave it up to LLMs at all… but in practice it's going to be the app crawlers that mark shows up using hard work and drive for differentiation, not the publishers themselves." It might not be an AI LLM, but someone still has to mark the metadata up. File: <i>03 - Sethi - Spurlock correction, app crawlers not publishers - PNWR.mp3</i> <b>Ammo for your side:</b> @manton — his daughter's voicemail "Hi dad, this isn't an emergency" was transcribed by Apple as "this is an emergency." Player-side LLM inference is lossy; publisher-authored tags are authoritative. The ca