Podcast archive pipeline: 402 episodes of Whisper transcription plus 646 AI Chinese summaries in parallel
vista8 · x · 2026-10-18
The blogger shares full pipeline details for the Lex Fridman podcast archive site.
Transcription: whisper large-v3-turbo on two parallel GPUs, filling in 402 missing episodes across No Priors, Cognitive Revolution, Training Data, and Unsupervised — estimated 1824 hours.
Chinese summaries: 646 articles generated with three-way concurrency, spot-checked for quality; discovered a model's default thinking chain was burning the token budget on long drafts, fixed via chattemplatekwargs: {enablethinking: false}, cutting response time from 5 minutes to 12 minutes per article.
Automated follow-up: transcript seq validation → banned-word sweep → rebuild 4 sites → rsync deploy → online verification.
More from coding & agent
- Open-weight TYGER speech decision model makes 80ms turn-taking calls for full-duplex voice agents — solyarisoftware · 2026-10-18
- Devs: Opus 5.5 on medium beats Opus on max — eyishazyer · 2026-10-18
- Cursor Super + Claude Pro + OpenCode: can this stack survive heavy autonomous coding? — DarioAmodeiWasTaken · 2026-10-18
- Coding AI wars: model is faster and smarter but 'dogshit at actual engineering' — MarcJSchmidt · 2026-10-18
- Anthropic borrows lawyers' ethics for Claude agents that closed 186 deals with no human approval — rohanpaul_ai · 2026-10-18
- Prompt injection via email can hijack your AI agent — jev-guard adds risk-scored guards — sujingshen · 2026-10-18