Interview
Dan Kokotov: Speech Recognition with AI and Humans | Lex Fridman Podcast #151
- Lex Friedman plans to implement captions and transcripts for current and future episodes to improve referenceability, alongside a dedicated video update to solicit user feedback for the new feature.
- The podcast anticipates sponsorship from Rev.ai in the coming months, while Dan Kokotov predicts significant advancements in Rev's automated speech recognition (ASR) technology over the next couple of years.
- Rev.ai is currently exploring the use of human transcriber edit data as a flywheel for machine learning, with a long-term goal of reducing word error rates from the current 14% to 3%.
- The host predicts that on December 1st, The Joe Rogan Experience will become exclusive to Spotify, resulting in the discontinuation of its RSS feed.
- Expectations exist for Spotify to adopt a standard converting all podcasts into searchable text and to treat transcripts as a standard RSS extension for direct player access.
- A proposed "baby project" aims to enable real-time English-Russian conversational experiences, potentially involving a translator and third-party speakers like Gregory Perlman.
- The host expressed a specific intention to conduct an interview with Vladimir Putin, citing confidence in their ability to facilitate the conversation effectively.
- Future predictions include a shift in corporate culture where all meetings are archived and indexed as searchable text, fundamentally altering remote work dynamics and information retrieval.
- Strategic focus is placed on optimizing for long-term user happiness and intellectual health rather than short-term engagement metrics, a philosophy attributed to leaders like Steve Jobs and Elon Musk.
- Skepticism is predicted for the new Dune movie due to potential confusion compared to the David Lynch adaptation, despite general audience excitement.
- The podcast ecosystem hopes to resolve discovery and search limitations by utilizing technology partners like Rev to convert content into searchable text formats.