Continuing the series on the various ways one can learn about the technical aspects of Bitcoin, in this article we will focus on transcripts and contributing to or reading the archive of transcripts maintained by Bryan Bishop (kanzure).
In the early years of Bitcoin’s history, all communication involving Satoshi Nakamoto occurred online on mailing lists, IRC and the BitcoinTalk forum. These years are well archived by the Satoshi Nakamoto Institute. There are no recordings of Satoshi speaking, presumably as they could have been used to identify him. However, once in-person meetups, conferences and meetings of core developers started to be organized, there was a danger of content from verbal presentations and discussions disappearing and being forgotten.
In the last decade, Bishop has transcribed over 600 transcripts racking up over a million-and-a-half words. The transcripts can be accessed here and pull requests to add or edit a transcript can be submitted to this GitHub repository. A small selection of highlights include a transcript on choosing safe curves for elliptic curve cryptography from 2014, a transcript of Greg Maxwell presenting confidential transactions from 2017 and the transcripts from the Bitcoin Core developer meetings that are not filmed or otherwise recorded.
Typing at the Speed of Lightning
At the CES Summit 2019, Bishop explained why all talks should have transcripts. These reasons include facilitating further discussion after the talk, distributing the content beyond the attendees in the room, and text being easier to parse and search than video and audio. His presentation spurred others to attempt to transcribe Bishop’s talk in real time.
Bishop takes pride in publishing the transcript before the speaker has sat down. He believes the immediate availability of the transcript is the most critical factor for those who utilize the transcripts, even more important than the quality of the content. It is certainly true that having a transcript available immediately at the conclusion is extremely valuable for supporting further in-person discussions and for bringing up to speed those who are not present but interested in what was discussed.
Granted, Bishop is an extremely fast typist. He started transcribing in high school when he sought to prove to his high school principal that the classes were a waste of his time. After four years of transcribing the classes’ content, he realized no one cared.
However, one upside of the experience is that Bishop was ranked 30th for typing speed out of 5 million competitors. He can type up to 200 words per minute. Court stenographers can typically type faster than this but they take advantage of special keyboards called stenotypes and a system of abbreviations called shorthand. If it wasn’t for his high-paying career in software development, Bishop could try to join the ranks of court stenographers earning around $200,000 per year.
The fastest speaker in the Bitcoin ecosystem is undoubtedly Laolu Osuntokun (roasbeef), CTO of Lightning Labs. He has become almost as renowned for his pace of verbal delivery as his weighty contributions to the lnd Lightning implementation and his work on Neutrino, the privacy-preserving light client. So if anyone in the Bitcoin ecosystem would be able to defeat Bishop, it would be him.
However, Bishop, with his ability to type up to 200 words per minute, has risen to the challenge on a number of occasions and conquered this particular human adversary. (The rivalry is obviously entirely good-natured and other individuals in the Bitcoin community have got involved in the fun on Twitter  and )
AI: Not a Complete Alternative
So no human speaker in the Bitcoin ecosystem has been able to defeat Bishop. But what about artificial intelligence? As it did in chess and the board game Go, is AI able to overpower the best humanity can offer and type at least as fast as Bishop but with even greater accuracy? The answer to this question is not yet.
The Stephan Livera Podcast is one of the most popular Bitcoin podcasts. Livera has experimented with transcripts on his show. Initially, a sponsor of the show (GiveBitcoin) paid for human transcription on a small subset of episodes and they are available on Livera’s site. Some of them have since been added to the transcript repository maintained by Bishop. These “polished” transcripts were purchased from rev.com. They are high quality in terms of accuracy, they promise to be 99 percent accurate but they cost $1 per audio minute.
Livera has also tried machine-generated transcripts from rev.com. These cost only $0.10 per audio minute but are only promised to be 80 percent accurate. Therefore, they require Livera or somebody else to edit them afterward.
The Challenge of ‘Searchability’ in Transcripts
On the Software Engineering Daily podcast, Wenbin Fang — the founder of ListenNotes, a podcast search engine — discussed with Jeff Meyerson the latest state of podcast transcripts….