How to digest 36 weekly podcasts without spending 36 hours listening | Tomasz Tunguz (Theory Ventures)
How I AI
2025/08/25
How to digest 36 weekly podcasts without spending 36 hours listening | Tomasz Tunguz (Theory Ventures)
How to digest 36 weekly podcasts without spending 36 hours listening | Tomasz Tunguz (Theory Ventures)
How I AI
2025/08/25
In this episode, Claire Vo sits down with Tomasz Tunguz, founder of Theory Ventures, to explore his innovative approach to podcast consumption and content extraction. Tomasz shares insights into his self-built 'Parakeet Podcast Processor,' a terminal-based system that automates the downloading, transcribing, and summarizing of podcasts—allowing him to process up to 36 episodes per week efficiently. The conversation dives into how AI tools like Whisper, Parakeet, and Claude are leveraged to extract investment theses, generate blog post drafts, and refine writing through iterative feedback loops.
Tomasz walks through the components of his podcast processing pipeline, including transcript cleaning with Gemma 3, structured summarization with DuckDB, and named entity recognition for identifying companies and themes. He emphasizes the efficiency of terminal-based tools and the power of AI in enabling personalized software workflows. A key focus is his method for generating blog posts using an 'AP English teacher' grading system, where AI models critique and refine drafts across multiple iterations to match his writing style. The discussion also touches on AI's broader role in writing education, productivity, and the future of hyper-personalized software experiences.
03:32
03:32
Tomasz built a podcast ripper system that downloads, transcribes, and summarizes podcasts automatically.
09:53
09:53
Focus shifted to output prompting and structure
10:21
10:21
The guest praises Cursor for blog post generation and modification.
12:38
12:38
The marginal friction to achieve a good fit with small utilities is now low.
15:32
15:32
AI is being used to generate blog post drafts from podcast content.
17:34
17:34
The speaker finds grading AI-generated blog posts like an AP English teacher effective.
18:25
18:25
Personal voice in AI writing requires preserving intentional imperfections and stylistic quirks like starting paragraphs with conjunctions.
22:00
22:00
Gemini critiques AI-generated blog posts in a student-teacher model
26:13
26:13
AI-generated blog posts evaluated using AP English grading show scores ranging from C-minuses to 91%
28:16
28:16
AI can help with grammar, logic checks, and refining ideas in academic writing
33:38
33:38
Tomasz suggests using Gemini and Claude to polish scripts and improve AI output quality through competition
