The Synergy of AI and Human Intelligence in Speech-to-Text Services with Dan Kokotov of Rev.ai

This episode features a conversation with Dan Kokotov, VP of Engineering at Rev.ai, delving into the company's innovative approach to speech-to-text services. Lex Fridman, a long-time user and admirer of Rev's offerings, highlights the profound impact of user-friendly software that simplifies complex tasks, drawing parallels with other beloved tools like Adobe Premiere and Izotope RX. The discussion explores Rev's origins, born from a desire to improve upon the traditional freelancer marketplace model (like Upwork) by standardizing services and abstracting away the complexities for both customers and freelancers, whom they affectionately call 'Revvers'.

Dan explains that Rev initially focused on translation and later expanded into audio transcription and captioning, prioritizing 'work from home' tasks that are relatively standardizable. The company operates as a two-sided marketplace, meticulously balancing the supply of Revvers with customer demand to ensure efficient service delivery and a positive experience for all. A key distinction is made between Rev.com, which offers human-powered services, and Rev.ai, which focuses on automated speech recognition (ASR). The conversation touches upon the diverse demographics of Revvers, ranging from work-from-home parents and students to individuals seeking flexible work arrangements or even those exploring the gig economy out of curiosity, many of whom find enjoyment and learning in the transcription process.

The technical core of Rev.ai's offering is its advanced ASR engine, which provides an initial draft for human transcribers to refine. Dan details the concept of Word Error Rate (WER) as a metric for ASR accuracy, stating Rev.ai's current WER at 14% for unstructured speech, compared to an estimated human benchmark of 2-3%. He emphasizes that Rev.ai aims to outperform major tech giants like Google, Amazon, and Microsoft in this specific domain. The process for Revvers involves using specialized tools to correct AI-generated transcripts, adapting their approach based on audio quality, sometimes even transcribing from scratch for challenging recordings.

Beyond the technical and business aspects, the podcast touches upon broader philosophical themes, notably Dan's admiration for Frank Herbert's 'Dune' series. He extracts a profound idea from 'God Emperor of Dune': the necessity of pressure and suffering to break societal stagnation and drive human progress. This philosophical interlude provides a unique lens through which to view the challenges and advancements in technology and work, suggesting that even in the pursuit of simplification and efficiency, an underlying struggle can be a catalyst for innovation and growth, ultimately benefiting humanity by enabling new forms of work and access to knowledge.

rev in general is a company that does captioning and transcription of audio by humans and by ai

holy somebody figured out how to do it just really easily I I'm I'm such a fan of just when people take a problem and they just make it easy

you cannot buy your way onto this podcast

no sponsor will ever influence what I do on this podcast or to the best of my ability influence what I think

the greatest sci-fi novel of all time is dune and the second greatest is the children of dune and the third greatest is the god emperor of doom

you need a little bit of pressure and suffering right to kind of like make progress not not not get too comfortable

rev was kind of founded to improve on the model of upwork that was kind of the original um or part of their original impetus

we don't think of it as kind of gig economy like to some degree I don't like the word gig that much right because to some degree diminishes the works being done

to us it's um improving the nature of working from home on your own time and on your own terms right and kind of taking away geographical limitations and time limitations

I love doing this because I get paid to watch a documentary on something right and I learn something while I'm transcribing

our accuracy right now it's I think it's maybe 14 word error rate on on um our test test suite that we generally use to measure

human accuracy most people think realistically it's like three percent two percent word error rate would be like the max achievable so there's still quite a gap

we measure ourselves against like google amazon microsoft you know some of the some smaller competitors