Knowledge · 00:02:38 · 373 words
file
knowledge.index
speakers
vocateca
address
vocateca.com/en/knowledge
duration
00:02:38 · 373 words
Knowledge base

Turning audio into knowledge

00:00:00vocateca

Field notes and essays on the method: pull a transcript, compile it into a knowledge network, query it, build from it. Start here.

00:00:10

Method & workflow

  • The knowledge network/knowledge-networkPull a transcript, compile it into a linked network, then work every question and every post out of that network. The loop we run, and why it compounds.
  • The compile step/knowledge/the-compile-stepHow a model turns a raw transcript into a linked encyclopedia page: one concept, sources in the header, every claim traceable to a line.
  • Compiled once, not retrieved every time/knowledge/compiled-not-ragWhy we don't RAG our own knowledge - and the cases where retrieval still wins.
  • Cross-domain links are the product/knowledge/cross-domain-linksThe insight lives where a podcast claim connects to an article claim. Only a linked network surfaces it.
  • From wiki to content/knowledge/from-wiki-to-contentWriting a post the thinking is already done for. The artefact is a view onto the network.
00:01:03

Foundations

  • Read, don't listen/transcriptsAudio is opaque. Text is searchable, quotable, and the only form an LLM can actually read. The case for transcripts, anchored on Karpathy.
  • Timestamps make a transcript navigable/knowledge/timestampsA transcript without per-line timestamps is half a transcript. Jump to the moment, quote the sentence.
  • Who said what/knowledge/who-said-whatSpeaker diarization, and why attributing a claim to a speaker is what makes a transcript usable as knowledge.
  • A transcript you own vs one you rent/knowledge/own-your-transcriptsAudio behind an app disappears. Plain text you keep is durable, greppable, and portable.
00:01:42

Sources & ingestion

  • Lexicap at home/knowledge/lexicap-at-homeTranscribe a whole back-catalogue overnight - the effort Karpathy spent by hand in 2022, automated on your Mac.
  • Scraped captions vs real transcription/knowledge/captions-vs-transcriptionWhy YouTube's auto-captions aren't enough: no punctuation, no speakers, rolling duplicates.
  • Follow creators, don't chase links/knowledge/follow-creatorsSubscribe once; new episodes and reels arrive already transcribed.
00:02:04

Formats & interoperability

00:02:13

Privacy & local-first

  • On-device by default/knowledge/on-device-by-defaultWhy sensitive recordings shouldn't touch the cloud - and what changes when transcription is local.
00:02:22

LLM & agent workflows