Dan Kokotov: Speech Recognition with AI and Humans

00:00:00.000 | The following is a conversation with Dan Kokorov,

00:00:03.280 | VP of Engineering at Rev.ai, which is by many metrics,

00:00:08.280 | the best speech to text AI engine in the world.

00:00:12.380 | Rev in general is a company that does captioning

00:00:15.440 | and transcription of audio by humans and by AI.

00:00:20.020 | I've been using their services for a couple of years now

00:00:22.680 | and planning to use Rev to add both captions and transcripts

00:00:26.960 | to some of the previous and future episodes of this podcast

00:00:30.520 | to make it easier for people to read through

00:00:33.160 | the conversation or reference various parts of the episode,

00:00:36.480 | since that's something that quite a few people requested.

00:00:39.800 | I'll probably do a separate video on that

00:00:41.380 | with links on the podcast website

00:00:45.480 | so people can provide suggestions and improvements there.

00:00:48.440 | Quick mention of our sponsors,

00:00:50.360 | Athletic Greens, all-in-one nutrition drink,

00:00:53.440 | Blinkist app that summarizes books,

00:00:56.520 | Business Wars podcast and Cash App.

00:01:00.160 | So the choice is health, wisdom or money.

00:01:03.440 | Choose wisely my friends.

00:01:04.960 | And if you wish, click the sponsor links below

00:01:07.840 | to get a discount and to support this podcast.

00:01:10.880 | As a side note, let me say that I reached out to Dan

00:01:13.240 | and the Rev team for a conversation

00:01:15.080 | because I've been using and genuinely loving their service

00:01:19.940 | and really curious about how it works.

00:01:22.480 | I previously talked to the head of Adobe Research

00:01:24.980 | for the same reason.

00:01:26.440 | For me, there's a bunch of products,

00:01:28.720 | usually it's software that comes along

00:01:31.120 | and just makes my life way easier.

00:01:33.160 | Examples are Adobe Premiere for video editing,

00:01:36.080 | iZotope RX for cleaning up audio,

00:01:38.680 | AutoHotKey on Windows for automating keyboard

00:01:41.160 | and mouse tasks,

00:01:43.480 | Emacs as an ID for everything,

00:01:46.080 | including the universe itself.

00:01:48.280 | I can keep on going, but you get the idea.

00:01:50.720 | I just like talking to people who create things

00:01:52.860 | I'm a big fan of.

00:01:54.420 | That said, after doing this conversation,

00:01:56.480 | the folks at Rev.ai offered to sponsor this podcast

00:02:01.120 | in the coming months.

00:02:02.720 | This conversation is not sponsored by the guest.

00:02:06.440 | It probably goes without saying,

00:02:08.280 | but I should say it anyway,

00:02:10.020 | that you cannot buy your way onto this podcast.

00:02:13.240 | I don't know why you would want to.

00:02:15.480 | I wanted to bring this up to make a specific point

00:02:18.320 | that no sponsor will ever influence

00:02:20.920 | what I do on this podcast,

00:02:22.740 | or to the best of my ability,

00:02:23.920 | influence what I think.

00:02:25.600 | I wasn't really thinking about this,

00:02:27.800 | for example, when I interviewed Jack Dorsey,

00:02:30.200 | who is the CEO of Square

00:02:32.340 | that happens to be sponsoring this podcast,

00:02:35.100 | but I should really make it explicit.

00:02:37.080 | I will never take money for bringing a guest on.

00:02:40.100 | Every guest on this podcast is someone

00:02:43.040 | I genuinely am curious to talk to,

00:02:44.920 | or just genuinely love something they've created.

00:02:48.200 | As I sometimes get criticized for,

00:02:50.720 | I'm just a fan of people,

00:02:52.460 | and that's who I talk to.

00:02:54.280 | As I also talk about way too much,

00:02:56.260 | money is really never a consideration.

00:02:58.880 | In general, no amount of money can buy my integrity.

00:03:03.000 | That's true for this podcast,

00:03:04.800 | and that's true for anything else I do.

00:03:06.800 | If you enjoy this thing, subscribe on YouTube,

00:03:10.320 | review on Apple Podcast,

00:03:12.600 | follow on Spotify,

00:03:13.920 | support on Patreon,

00:03:15.360 | or connect with me on Twitter @LexFriedman.

00:03:18.240 | And now, here's my conversation with Dan Kokorov.

00:03:22.820 | You mentioned science fiction on the phone,

00:03:25.620 | so let's go with the ridiculous first.

00:03:28.060 | What's the greatest sci-fi novel of all time,

00:03:31.060 | in your view?

00:03:32.380 | And maybe, what ideas do you find

00:03:35.980 | philosophically fascinating about it?

00:03:37.860 | - The greatest sci-fi novel of all time is Dune,

00:03:41.220 | and the second greatest is the Children of Dune,

00:03:44.140 | and the third greatest is the God Emperor of Dune, so.

00:03:47.820 | I'm a huge fan of the whole series.

00:03:50.360 | I mean, it's just an incredible world that he created.

00:03:53.980 | And I don't know if you've read the book or not.

00:03:55.740 | - No, I have not.

00:03:56.580 | It's one of my biggest regrets,

00:03:58.820 | especially 'cause the new movie is coming out.

00:04:01.740 | Everyone's super excited about it.

00:04:03.940 | It's ridiculous to say,

00:04:06.300 | and sorry to interrupt,

00:04:07.460 | is that I used to play the video game.

00:04:10.280 | It used to be Dune.

00:04:11.940 | I guess you would call that real-time strategy.

00:04:14.340 | - Right, right, I think I remember that game.

00:04:15.900 | - Yeah, it was kind of awesome, '90s or something.

00:04:18.060 | I think I played it, actually, when I was in Russia.

00:04:20.480 | - I definitely remember it.

00:04:21.960 | I was not in Russia anymore.

00:04:23.580 | I think at the time that I used to live in Russia,

00:04:26.080 | I think video games were about the suspicion of Pong.

00:04:29.760 | I think Pong was pretty much the greatest game

00:04:32.400 | I ever got to play in Russia,

00:04:33.960 | which was still a privilege in that age.

00:04:35.920 | - So you didn't get color?

00:04:37.240 | You didn't get like a--

00:04:38.600 | - Well, so I left Russia in 1991, right?

00:04:40.880 | - '91, okay.

00:04:41.720 | - So I always wanted to feel like a kid

00:04:43.960 | 'cause my mom was a programmer,

00:04:45.200 | so I would go to her work.

00:04:47.160 | I would take the Metro.

00:04:49.180 | I'd go to her work and play on, I guess,

00:04:51.260 | the equivalent of a 286 PC, you know?

00:04:53.860 | - Nice, with floppy disks.

00:04:56.140 | - Yes, yes.

00:04:56.980 | - So okay, but back to Dune, what do you get?

00:04:58.660 | - Back to Dune.

00:04:59.980 | And by the way, the new movie I'm pretty interested in,

00:05:02.780 | but the original--

00:05:04.260 | - You're skeptical?

00:05:05.300 | - I'm a little skeptical.

00:05:06.580 | I'm a little skeptical.

00:05:07.420 | I saw the trailer.

00:05:08.260 | I don't know, so there's a David Lynch movie, Dune,

00:05:11.580 | as you may know.

00:05:12.580 | I'm a huge David Lynch fan, by the way.

00:05:14.300 | So the movie is somewhat controversial,

00:05:17.280 | but it's a little confusing,

00:05:19.920 | but it captures kind of the mood of the book

00:05:22.640 | better than I would say most any adaptation.

00:05:25.640 | And like, Dune is so much about kind of mood

00:05:27.320 | and the world, right?

00:05:28.720 | But back to the philosophical point.

00:05:30.000 | So in the fourth book, "God, Emperor of Dune,"

00:05:34.200 | there's a sort of setting where Leto,

00:05:38.640 | one of the characters,

00:05:39.480 | he's become this weird sort of god-emperor.

00:05:41.920 | He's turned into a gigantic worm,

00:05:43.320 | and you kind of have to read the book

00:05:44.500 | to understand what that means.

00:05:45.340 | - So the worms are involved.

00:05:46.980 | - Worms are involved.

00:05:47.820 | You probably saw the worms in the trailer, right?

00:05:49.900 | - And in the video game.

00:05:50.740 | - So he kind of like merges with this worm

00:05:53.060 | and becomes this tyrant of the world,

00:05:55.100 | and he like oppresses the people for a long time, right?

00:05:57.500 | But he has a purpose,

00:05:58.700 | and the purpose is to kind of break through

00:06:01.900 | kind of a stagnation period in civilization, right?

00:06:05.580 | But people have gotten too comfortable, right?

00:06:07.240 | And so he kind of oppresses them so that they explode

00:06:11.620 | and go on to colonize new worlds

00:06:14.240 | and kind of renew the forward momentum of humanity, right?

00:06:17.800 | And so to me, that's kind of fascinating, right?

00:06:19.780 | You need a little bit of pressure and suffering, right,

00:06:22.880 | to kind of make progress, not get too comfortable.

00:06:27.000 | (laughing)

00:06:29.540 | Maybe that's a bit of a cruel philosophy to take away, but.

00:06:33.700 | - That seems to be the case, unfortunately.

00:06:36.820 | Obviously, I'm a huge fan of suffering.

00:06:40.240 | So one of the reasons we're talking today

00:06:43.740 | is that a bunch of people requested

00:06:47.560 | that I do transcripts for this podcast and do captioning.

00:06:52.460 | I used to make all kinds of YouTube videos,

00:06:54.580 | and I would go on Upwork, I think,

00:06:58.360 | and I would hire folks to do transcription,

00:07:00.580 | and it was always a pain in the ass, if I'm being honest.

00:07:04.620 | And then I don't know how I discovered Rev,

00:07:08.540 | but when I did, it was this feeling of like,

00:07:13.060 | holy shit, somebody figured out

00:07:15.380 | how to do it just really easily.

00:07:17.280 | I'm such a fan of just,

00:07:21.080 | when people take a problem and they just make it easy.

00:07:26.860 | - Right.

00:07:27.700 | - Like just,

00:07:28.520 | there's so many, it's like there's so many things in life

00:07:34.380 | that you might not even be aware of that are painful,

00:07:37.740 | and then Rev, you just like give the audio, give the video,

00:07:42.740 | you can actually give a YouTube link,

00:07:45.060 | and then it comes back like a day later,

00:07:49.580 | or two days later, whatever the hell it is,

00:07:53.340 | with the captions, all in a standardized format.

00:07:56.500 | I don't know, it was truly a joy.

00:08:00.380 | So I thought I had, just for the hell of it, talk to you.

00:08:05.020 | One other product, it just made my soul feel good.

00:08:08.260 | One other product I've used like that

00:08:10.900 | is for people who might be familiar,

00:08:12.980 | is called iZotope RX, it's for audio editing.

00:08:17.340 | And that's another one where it was like,

00:08:22.360 | you just drop it, I dropped into the audio,

00:08:25.940 | and it just cleans everything up really nicely.

00:08:28.500 | All the stupid, like the mouth sounds,

00:08:32.100 | and sometimes there's background like sounds

00:08:37.100 | due to the malfunction of the equipment,

00:08:39.220 | it can clean that stuff up.

00:08:40.620 | It has like general voice denoising,

00:08:43.300 | it has like automation capabilities

00:08:46.060 | where you can do batch processing,

00:08:47.660 | and you can put a bunch of effects.

00:08:49.820 | I mean, it just, I don't know, everything else sucked

00:08:53.900 | for like voice-based cleanup that I've ever used.

00:08:58.060 | I've used Audition, Adobe Audition,

00:09:00.100 | I've used all kinds of other things with plugins,

00:09:02.460 | and you have to kind of figure it all out,

00:09:04.700 | you have to do it manually, here it just worked.

00:09:07.740 | So that's another one in this whole pipeline

00:09:09.860 | that just brought joy to my heart.

00:09:12.860 | Anyway, all that to say is,

00:09:14.800 | Rev put a smile to my face.

00:09:18.980 | So can you maybe take a step back and say,

00:09:22.180 | what is Rev, and how does it work?

00:09:24.780 | And Rev or Rev.com?

00:09:27.060 | - Rev, Rev.com.

00:09:28.100 | (laughing)

00:09:29.060 | Same thing, I guess.

00:09:30.460 | Though we do have Rev.ai now as well,

00:09:32.220 | which we can talk about later.

00:09:34.620 | - Like, do you have the actual domain, or is it just--

00:09:37.260 | - The actual domain, but we also use it

00:09:39.820 | kind of as a sub-brand.

00:09:42.540 | So we use Rev.ai to denote our ASR services, right?

00:09:46.860 | And Rev.com is kind of our more human

00:09:48.900 | and to the end user services.

00:09:50.620 | - So it's like wordpress.com and wordpress.org,

00:09:53.180 | they actually have separate brands that like,

00:09:55.860 | I don't know if you're familiar with what those are.

00:09:57.580 | - Yeah, yeah, yeah.

00:09:58.420 | They provide almost like a separate branch of--

00:10:01.020 | - A little bit, I think with that, it's like,

00:10:02.940 | wordpress.org is kind of their open source, right?

00:10:05.020 | And wordpress.com is sort of their

00:10:07.100 | hosted commercial offering.

00:10:08.660 | - Yes.

00:10:09.500 | - And with us, the differentiation is a little bit different,

00:10:11.180 | but maybe a similar idea.

00:10:12.780 | - Yeah.

00:10:13.700 | Okay, so what is Rev?

00:10:14.820 | - Before I launch into what is Rev,

00:10:17.420 | I was gonna say, you know, like you were talking about,

00:10:18.980 | like Rev was music to your ears.

00:10:20.740 | Your spiel was music to my ears,

00:10:22.900 | and to us, the founders of Rev,

00:10:25.620 | because Rev was kind of founded

00:10:28.620 | to improve on the model of Upwork.

00:10:30.620 | That was kind of the original,

00:10:32.180 | or part of their original impetus.

00:10:35.180 | Like our CEO, Jason, was a early employee of Upwork,

00:10:39.700 | so he's very familiar with their--

00:10:40.700 | - Upwork the company.

00:10:41.620 | - Upwork the company.

00:10:42.660 | And so he was very familiar with that model,

00:10:45.540 | and he wanted to make the whole experience better,

00:10:48.580 | because he knew like, when you go,

00:10:50.140 | at that time, Upwork was primarily programmers.

00:10:52.460 | So the main thing they offered is,

00:10:54.860 | if you wanna hire someone to help you code a little site,

00:10:57.780 | you could go on Upwork,

00:11:00.580 | and you could like browse through a list of freelancers,

00:11:03.060 | pick a programmer, have a contract with them,

00:11:05.420 | and have them do some work.

00:11:07.260 | But it was kind of a difficult experience,

00:11:09.740 | because for you, you would kind of have to browse

00:11:14.140 | through all these people, right?

00:11:14.980 | And you have to decide, okay, like,

00:11:16.140 | well, is this guy good, or is somebody else better?

00:11:20.260 | And naturally, you're going to Upwork

00:11:22.900 | because you're not an expert, right?

00:11:24.540 | If you're an expert, you probably wouldn't be

00:11:25.700 | like getting a programmer from Upwork.

00:11:27.780 | So how can you really tell?

00:11:29.820 | So there's kind of like a lot of potential regret, right?

00:11:33.060 | What if I choose a bad person?

00:11:34.780 | They're like gonna be late on the work.

00:11:36.460 | It's gonna be a painful experience.

00:11:38.180 | And for the freelancer, it was also painful,

00:11:40.700 | because half the time, they spent not

00:11:42.500 | on actually doing the work,

00:11:43.540 | but kind of figuring out how can I make my profile

00:11:46.820 | most attractive to the buyer, right?

00:11:49.020 | And they're not an expert on that either.

00:11:51.340 | So like, Rob's idea was, let's remove the barrier, right?

00:11:54.220 | Like, let's make it simple.

00:11:55.340 | We'll pick a few verticals that are fairly standardizable.

00:12:00.180 | Now, we actually started with translation,

00:12:02.580 | and then we added audio transcription a bit later.

00:12:05.380 | And we'll just make it a website.

00:12:06.900 | You go, give us your files.

00:12:08.660 | We'll give you back the results as soon as possible.

00:12:13.660 | Originally, maybe it was 48 hours,

00:12:15.580 | then we made it shorter and shorter and shorter.

00:12:18.180 | - Yeah, there's a rush processing too.

00:12:19.660 | - There's a rush processing now.

00:12:21.780 | And we'll hide all the details from you, right?

00:12:25.340 | - Yeah.

00:12:26.180 | - And like, that's kind of exactly

00:12:27.940 | what you're experiencing, right?

00:12:28.940 | You don't need to worry about the details

00:12:30.620 | of how the sausage is made.

00:12:31.820 | - That's really cool.

00:12:32.940 | So you picked like a vertical.

00:12:34.980 | By vertical, you mean basically a--

00:12:37.220 | - A service, a service category.

00:12:39.300 | - Why translation?

00:12:41.220 | Is Rev thinking of potentially going

00:12:43.220 | into other verticals in the future?

00:12:45.420 | Or is this like the focus now is translation,

00:12:47.980 | transcription, like language?

00:12:50.180 | - The focus now is language or speech services generally,

00:12:54.580 | speech to text, language services.

00:12:56.340 | You can kind of group them however you want.

00:12:58.540 | So, but we originally,

00:13:02.620 | the categorization was work from home.

00:13:05.100 | And so we wanted work that was done by people on a computer.

00:13:08.220 | You know, we weren't trying to get into, you know,

00:13:11.020 | task rabbit type of things.

00:13:13.220 | And something that could be relatively standard,

00:13:16.100 | not a lot of options.

00:13:17.100 | So we could kind of present the simplified interface, right?

00:13:20.020 | So programming wasn't like a good fit

00:13:21.460 | because each programming project is kind of unique, right?

00:13:24.740 | We're looking for something that transcription is,

00:13:28.300 | you know, you have five hours of audio,

00:13:29.500 | it's five hours of audio, right?

00:13:30.860 | Translation is somewhat similar in that, you know,

00:13:33.460 | you can have a five page document, you know,

00:13:36.660 | and then you just can price it by that.

00:13:38.420 | And then you pick the language you want,

00:13:40.220 | and that's mostly all that is to it.

00:13:42.620 | So those were a few criteria.

00:13:43.980 | We started with translation because we saw the need

00:13:48.340 | and we picked up kind of a specialty of translation

00:13:53.340 | where we would translate things like birth certificates,

00:13:57.420 | immigration documents, things like that.

00:14:01.540 | And so they were fairly, even more well-defined

00:14:06.020 | and easy to kind of tell if we did a good job.

00:14:08.260 | - So you can literally charge per type of document?

00:14:10.740 | Was that the, so what is it now?

00:14:14.100 | Is it per word or something like that?

00:14:15.700 | Like how do you measure the effort involved

00:14:20.220 | in a particular thing?

00:14:21.500 | - So now it looks like for audio transcription, right?

00:14:23.420 | It's per audio unit.

00:14:24.980 | - Well, that, yes.

00:14:26.580 | - For our translation,

00:14:27.420 | we don't really actually focus on that anymore.

00:14:30.580 | But, you know, back when it was still a main business

00:14:33.140 | of Revit was per page, right?

00:14:35.260 | Or per word, depending on the kind of--

00:14:36.940 | - 'Cause you can also do translation now

00:14:38.900 | on the audio, right?

00:14:40.660 | - Mm-hmm, like subtitles.

00:14:41.980 | So it would be both transcription and translation.

00:14:45.060 | - That's right.

00:14:45.900 | - I wanted to test the system to see how good it is,

00:14:48.500 | to see like how, well, is Russian supported?

00:14:51.980 | - I think so, yeah.

00:14:54.420 | - It'd be interesting to try it out.

00:14:55.860 | I mean, one of the--

00:14:56.700 | - But now it's only in like the one direction, right?

00:14:58.180 | So you start with English

00:14:59.260 | and then you can have subtitles in Russian.

00:15:00.980 | - In Russian.

00:15:01.820 | - Not really the other way.

00:15:02.900 | - Got it, because I'm deeply curious about this.

00:15:06.140 | When COVID opens up a little bit,

00:15:07.860 | when the economy, when the world opens up a little bit.

00:15:10.900 | - You wanna build your brand in Russia?

00:15:12.820 | - No, I don't.

00:15:14.020 | First of all, I'm allergic to the word brand.

00:15:15.900 | (laughing)

00:15:17.980 | I'm definitely not building any brands in Russia.

00:15:21.140 | But I'm going to Paris to talk to the translators

00:15:24.540 | of Dostoevsky and Tolstoy.

00:15:26.660 | There's this famous couple that does translation.

00:15:29.820 | And I'm more and more thinking of how is it possible

00:15:34.820 | to have a conversation with a Russian speaker?

00:15:37.860 | 'Cause I have just some number of famous Russian speakers

00:15:42.660 | that I'm interested in talking to.

00:15:44.940 | And my Russian is not strong enough to be witty and funny.

00:15:49.780 | I'm already an idiot in English.

00:15:51.980 | I'm an extra level of like awkward idiot in Russian,

00:15:56.300 | but I can understand it, right?

00:15:58.260 | And I also like wonder how can I create

00:16:01.940 | a compelling English-Russian experience

00:16:05.220 | for an English speaker?

00:16:06.540 | Like if I, there's a guy named Grigori Perlman,

00:16:09.260 | who's a mathematician,

00:16:11.340 | who obviously doesn't speak any English.

00:16:14.540 | So I would probably incorporate

00:16:17.220 | like a Russian translator into the picture.

00:16:21.460 | And then it would be like a, not to use a weird term,

00:16:24.380 | but like a three person thing,

00:16:28.100 | where it's like a dance of, like I understand it one way,

00:16:33.020 | they don't understand the other way,

00:16:34.780 | but I'll be asking questions in English.

00:16:38.220 | I don't know.

00:16:39.060 | I don't know the right way.

00:16:39.900 | - It's complicated.

00:16:40.740 | - It's complicated,

00:16:41.580 | but I feel like it's worth the effort

00:16:42.940 | for certain kinds of people.

00:16:45.100 | One of whom I'm confident is Vladimir Putin,

00:16:48.140 | I'm for sure talking to.

00:16:49.220 | I really want to make it happen

00:16:50.820 | 'cause I think I could do a good job with it.

00:16:52.420 | But the right, you know,

00:16:54.620 | understanding the fundamentals of translation

00:16:57.460 | is something I'm really interested in.

00:16:59.300 | So that's why I'm starting with the actual translators

00:17:02.780 | of like Russian literature,

00:17:04.740 | because they understand the nuance

00:17:06.260 | and the beauty of the language

00:17:07.540 | and how it goes back and forth.

00:17:09.780 | But I also want to see like in speech,

00:17:11.980 | how can we do it in real time?

00:17:14.180 | So that's like a little bit of a baby project

00:17:17.860 | that I hope to push forward.

00:17:19.060 | But anyway.

00:17:19.900 | - It's a challenging thing.

00:17:20.820 | So just to share,

00:17:22.980 | my dad actually does translation.

00:17:26.300 | Not professionally, he writes poetry.

00:17:28.980 | That was kind of always his,

00:17:30.420 | not a hobby, but he had a job, like a day job,

00:17:35.860 | but his passion was always writing poetry.

00:17:38.700 | And then we got to America

00:17:40.700 | and like he started also translating.

00:17:42.620 | First he was translating English poetry to Russian.

00:17:46.140 | Now he also like goes the other way.

00:17:49.060 | You kind of gain some small fame in that world anyways,

00:17:52.620 | because recently this poet, like Louise Clark,

00:17:56.340 | I don't know if you know of,

00:17:58.180 | some American poet,

00:17:59.620 | she was awarded the Nobel Prize for literature.

00:18:01.980 | And so my dad had translated

00:18:04.740 | one of her books of poetry into Russian.

00:18:07.380 | He was like one of the few.

00:18:08.660 | So he kind of like, they asked him

00:18:10.180 | and gave an interview to Radio Svoboda,

00:18:13.220 | if you know what that is.

00:18:14.060 | And he kind of talked about some of the intricacies

00:18:16.780 | of translating poetry.

00:18:18.060 | So that's like an extra level of difficulty, right?

00:18:19.700 | Because translating poetry is even more challenging

00:18:22.420 | than translating just, you know, interviews.

00:18:25.460 | - Do you remember any experiences and challenges

00:18:28.500 | to having to do the translation that stuck out to you?

00:18:32.420 | Like something he's talked about?

00:18:34.380 | - I mean, a lot of it I think is word choice, right?

00:18:36.540 | It's the way Russian is structured

00:18:38.380 | is first of all quite different

00:18:39.620 | than the way English is structured, right?

00:18:41.620 | Just there's inflections in Russian and genders

00:18:44.100 | and they don't exist in English.

00:18:46.020 | One of the reasons actually why machine translation

00:18:48.900 | is quite difficult for English to Russian

00:18:50.860 | and Russian to English,

00:18:51.820 | because they're such different languages.

00:18:53.980 | But then English has like a huge number of words,

00:18:57.140 | many more than Russian actually, I think.

00:18:58.500 | So it's often difficult to find the right word

00:19:01.500 | to convey the same emotional meaning.

00:19:04.020 | - Yeah, Russian language, they play with words much more.

00:19:07.620 | So you were mentioning that Rev was kind of born

00:19:11.740 | out of trying to take a vertical on Upwork

00:19:15.220 | and then standardize it.

00:19:17.380 | So--

00:19:18.220 | - We're just trying to make

00:19:19.060 | the freelancer marketplace idea better, right?

00:19:23.060 | Better for both customers

00:19:26.100 | and better for the freelancers themselves.

00:19:28.780 | - Is there something else to the story of Rev,

00:19:31.300 | finding Rev?

00:19:32.700 | Like what did it take to bring it to actually to life?

00:19:35.740 | Was there any pain points?

00:19:37.180 | - Plenty of pain points.

00:19:39.860 | I mean, as often the case, it's with scaling it up, right?

00:19:44.020 | And in this case, the scaling is kind of scaling

00:19:47.060 | the marketplace, so to speak, right?

00:19:49.300 | Rev is essentially a two-sided marketplace, right?

00:19:51.620 | Because there's the customers and then there's the Revvers.

00:19:55.700 | If there's not enough Revvers,

00:19:57.540 | Revvers are what we call our freelancers.

00:19:59.260 | So if there's not enough Revvers,

00:20:01.260 | then customers have a bad experience, right?

00:20:04.020 | Takes longer to get your work done, things like that.

00:20:07.580 | If there's too many, then Revvers have a bad experience

00:20:10.620 | because they might log on to see what work is available

00:20:13.100 | and there's not very much work, right?

00:20:15.460 | So kind of keeping that balance

00:20:17.140 | is a quite challenging problem.

00:20:20.220 | And that's like a problem we've been working on

00:20:22.740 | for many years.

00:20:23.740 | We're still refining our methods, right?

00:20:26.020 | - If you can kind of talk to this gig economy idea,

00:20:29.660 | I did a bunch of different psychology experiments

00:20:31.780 | on Mechanical Turk, for example.

00:20:33.540 | I've asked to do different kinds of very tricky

00:20:36.300 | computer vision annotation on Mechanical Turk

00:20:38.580 | and it's connecting people in a more systematized way.

00:20:43.580 | I would say, you know, between task and,

00:20:48.180 | what would you call that, worker,

00:20:51.340 | is what Mechanical Turk calls it.

00:20:53.740 | What do you think about this world of gig economies,

00:20:57.180 | of there being a service that connects customers to workers

00:21:02.180 | in a way that's like massively distributed,

00:21:07.980 | like potentially scaling to,

00:21:10.300 | it could be scaled to like tens of thousands of people,

00:21:13.220 | right?

00:21:14.060 | Is there something interesting about that world

00:21:17.100 | that you can speak to?

00:21:18.260 | - Yeah, well, we don't think of it as kind of gig economy,

00:21:21.380 | but like to some degree,

00:21:22.700 | I don't like the word gig that much, right?

00:21:24.420 | Because to some degree it diminishes

00:21:26.700 | the work being done, right?

00:21:27.940 | It sounds kind of like almost amateurish.

00:21:30.100 | Well, maybe in like music industry,

00:21:32.820 | like gig is the standard term,

00:21:34.180 | but in work, it kind of sounds like it's frivolous.

00:21:39.180 | To us, it's improving the nature of working from home

00:21:45.220 | on your own time and on your own terms, right?

00:21:48.060 | And kind of taking away geographical limitations

00:21:52.020 | and time limitations, right?

00:21:54.220 | So, many of our freelancers are maybe work from home moms,

00:21:58.740 | right?

00:21:59.580 | And they don't want the traditional nine to five job,

00:22:02.540 | but they wanna make some income

00:22:04.620 | and Rev kind of like allows them to do that

00:22:06.260 | and decide like exactly how much to work and when to work.

00:22:09.720 | Or by the same token, maybe someone is,

00:22:13.460 | someone wants to live the mountain top,

00:22:18.020 | life, right?

00:22:18.860 | You know, cabin in the woods,

00:22:20.220 | but they still wanna make some money.

00:22:22.500 | And like, generally that wouldn't be compatible

00:22:25.020 | before this new world, you kind of had to choose.

00:22:28.620 | But like with Rev, like you feel like

00:22:30.380 | you don't have to choose.

00:22:31.500 | - Can you speak to like,

00:22:33.300 | what's the demographics like distribution,

00:22:38.220 | like where do Revvers live?

00:22:40.780 | Is it from all over the world?

00:22:42.420 | Like, what is it?

00:22:43.260 | Do you have a sense of what's out there?

00:22:46.420 | - We're all over the world.

00:22:48.220 | Most of them are in the US, that's the majority.

00:22:51.500 | Yeah, because most of our work is audio transcription

00:22:54.900 | and so you have to speak pretty good English.

00:22:57.660 | So, the majority of them are from the US,

00:22:59.140 | so we have people in some other

00:23:00.940 | of the English speaking countries.

00:23:03.220 | And as far as like US, it's really all over the place.

00:23:06.060 | You know, for some of the years now,

00:23:09.220 | we've been doing these little meetings

00:23:10.460 | where the management team will go to some place

00:23:12.300 | and we'll try to meet Revvers.

00:23:13.660 | And, you know, pretty much wherever we go,

00:23:15.820 | it's pretty easy to find, you know,

00:23:17.860 | a large number of Revvers.

00:23:19.140 | You know, the most recent one we did is in Utah.

00:23:21.580 | But anywhere really.

00:23:25.260 | - Are they from all walks of life?

00:23:26.700 | Are these young folks, older folks?

00:23:28.900 | - Yeah, all walks of life really.

00:23:30.220 | Like I said, you know, one category is, you know,

00:23:32.580 | the work from home, students, you know,

00:23:34.860 | who wanna make some extra income.

00:23:37.100 | There are some people who maybe, you know,

00:23:40.060 | maybe they have some social anxiety,

00:23:42.260 | so they don't wanna be in the office, right?

00:23:43.700 | And this is one way for them to make a living.

00:23:45.260 | So it's really pretty wide variety.

00:23:47.420 | But like on the flip side, for example,

00:23:49.060 | one Revver we were talking to was a person

00:23:52.860 | who had a fairly high-powered career before

00:23:54.620 | and was kind of like taking a break

00:23:57.260 | and just wanted, she was almost doing this

00:23:59.260 | just to explore and learn about, you know,

00:24:01.300 | the gig economy, quote unquote, right?

00:24:03.420 | So it really is a pretty wide variety of folks.

00:24:06.300 | - Yeah, it's kind of interesting

00:24:08.380 | through the captioning process

00:24:10.460 | for me to learn about the Revvers

00:24:13.100 | because like some are clearly like weirdly knowledgeable

00:24:18.100 | about technical concepts.

00:24:22.940 | Like you can tell by how good they are

00:24:25.260 | at like capitalizing stuff, like technical terms,

00:24:29.060 | like in machine learning and deep learning.

00:24:30.660 | - Right.

00:24:32.140 | - I've used Rev to annotate, to caption

00:24:35.460 | the deep learning lectures or machine learning lectures

00:24:37.980 | I did at MIT.

00:24:39.860 | And it's funny, like a large number of them were like,

00:24:44.500 | I don't know if they looked it up

00:24:45.820 | or were already knowledgeable,

00:24:47.260 | but they do a really good job at like, I don't know.

00:24:50.380 | - They invest time into these things.

00:24:52.340 | They will like do research, they will Google things,

00:24:54.980 | you know, to kind of make sure they get it right.

00:24:57.340 | But to some of them, it's like,

00:24:59.060 | it's actually part of the enjoyment of the work.

00:25:01.580 | Like they'll tell us, you know,

00:25:03.300 | I love doing this because I get paid

00:25:05.740 | to watch a documentary on something, right?

00:25:07.380 | And I learned something while I'm transcribing, right?

00:25:10.060 | Pretty cool.

00:25:10.900 | - Yeah.

00:25:11.740 | So what's that captioning transcription process

00:25:14.660 | look like for the Revver?

00:25:16.180 | Can you maybe speak to that to give people a sense,

00:25:18.940 | like how much is automated, how much is manual?

00:25:22.020 | What's the actual interface look like?

00:25:25.140 | All that kind of stuff.

00:25:26.300 | - Yeah, so, you know, we've invested

00:25:28.380 | a pretty good amount of time to give like our Revvers

00:25:31.020 | the best tools possible.

00:25:33.020 | You know, so typical day for Revver,

00:25:34.740 | they might log into their workspace,

00:25:37.100 | they'll see a list of audios that need to be transcribed.

00:25:41.380 | And we try to give them tools to pick specifically

00:25:43.380 | the ones they want to do, you know?

00:25:44.460 | So maybe some people like to do longer audios

00:25:47.740 | or shorter audios.

00:25:49.060 | People have their preferences.

00:25:52.300 | Some people like to do audios in a particular subject

00:25:55.020 | or from a particular country.

00:25:55.980 | So we try to give people, you know,

00:25:58.100 | the tools to control things like that.

00:26:01.060 | And then when they pick what they want to do,

00:26:04.580 | we'll launch a specialized editor that we've built

00:26:07.460 | to make transcription as efficient as possible.

00:26:10.180 | They'll start with a speech rec draft.

00:26:12.340 | So, you know, we have our machine learning model

00:26:15.140 | for automated speech recognition.

00:26:17.260 | They'll start with that.

00:26:18.500 | And then our tools are optimized to help them correct that.

00:26:22.740 | - So it's basically a process of correction.

00:26:24.940 | - Yeah, it depends on, you know, I would say the audio.

00:26:29.500 | If audio itself is pretty good,

00:26:31.340 | like probably like our podcast right now

00:26:33.140 | would be quite good.

00:26:34.100 | So they would do a fairly good job.

00:26:36.780 | But if you imagine someone recorded a lecture, you know,

00:26:41.380 | in the back of a auditorium, right?

00:26:45.700 | Where like the speaker is really far away

00:26:47.340 | and there's maybe a lot of crosstalk and things like that,

00:26:49.940 | then maybe they wouldn't do a good job.

00:26:52.300 | So the person might say like, you know what,

00:26:53.700 | I'm just gonna do it from scratch.

00:26:55.060 | - Do it from scratch, yeah.

00:26:56.260 | - So it kind of really depends.

00:26:57.620 | - What would you say is the speed that you can possibly get?

00:27:00.540 | Like what's the fastest?

00:27:02.820 | Is it possible to get real time or no?

00:27:05.220 | As you're like listening, can you write as fast as-

00:27:09.140 | - Real time would be pretty difficult.

00:27:10.420 | It's actually a pretty, it's not an easy job.

00:27:12.940 | You know, we actually encourage everyone at the company

00:27:16.260 | to try to be a transcriber for a day,

00:27:17.660 | transcriptionist for a day.

00:27:19.020 | And it's way harder than you might think it is, right?

00:27:24.060 | Because people talk fast and people have accents

00:27:28.260 | and all this kind of stuff.

00:27:29.180 | So real time is pretty difficult.

00:27:30.940 | - Is it possible?

00:27:32.580 | Like there's somebody, we're probably gonna use Rev

00:27:35.100 | to caption this.

00:27:37.340 | They're listening to this right now.

00:27:39.300 | What do you think is the fastest

00:27:42.380 | you could possibly get on this right now?

00:27:44.860 | - I think on a good audio,

00:27:46.380 | maybe two to three X, I would say, real time.

00:27:49.820 | - Meaning it takes two to three times longer

00:27:51.620 | than the actual audio of the podcast.

00:27:55.500 | This is so meta.

00:27:56.740 | I could just imagine the Revvers working on this right now.

00:27:59.660 | Like you're way wrong.

00:28:01.020 | - You're way wrong, this takes way longer.

00:28:03.540 | But yeah, it definitely works.

00:28:04.380 | - Or you doubted me, I could do real time.

00:28:06.380 | (both laughing)

00:28:08.620 | - Okay, so you mentioned ASR.

00:28:11.180 | Can you speak to what is ASR, automatic speech recognition?

00:28:15.460 | How much, like what is the gap

00:28:19.300 | between perfect human performance

00:28:22.020 | and perfect or pretty damn good ASR?

00:28:26.660 | - Yeah, so ASR, automatic speech recognition,

00:28:28.780 | it's a class of machine learning problem, right?

00:28:31.820 | To take speech like we're talking

00:28:34.220 | and transform it into a sequence of words, essentially.

00:28:37.060 | - Audio of people talking.

00:28:38.780 | - Audio to words.

00:28:40.540 | And there's a variety of different approaches and techniques

00:28:44.900 | which we could talk about later if you want.

00:28:47.100 | So we think we have pretty much the world's best ASR

00:28:51.860 | for this kind of speech, right?

00:28:54.020 | So there's different kinds of domains, right, for ASR.

00:28:56.940 | Like one domain might be voice assistance, right?

00:29:00.220 | So Siri, very different than what we're doing, right?

00:29:04.220 | Because Siri, there's fairly limited vocabulary.

00:29:06.820 | You might ask Siri to play a song

00:29:09.860 | or order a pizza or whatever.

00:29:11.900 | And it's very good at doing that.

00:29:13.540 | Very different from when we're talking

00:29:16.220 | in a very unstructured way.

00:29:18.220 | And Siri will also generally adapt to your voice

00:29:20.180 | and stuff like this.

00:29:21.420 | So for this kind of audio, we think we have the best.

00:29:24.420 | And our accuracy, right now it's, I think,

00:29:29.420 | it's maybe 14% word error rate on our test suite

00:29:34.420 | that we generally use to measure.

00:29:35.340 | So word error rate is like one way to measure

00:29:38.180 | accuracy for ASR, right?

00:29:39.580 | - So what's 14% word error rate mean?

00:29:41.420 | - So 14% means across this test suite

00:29:44.860 | of a variety of different audios,

00:29:46.740 | it would be, it would get in some way 14%

00:29:53.500 | of the words wrong, 14% of the words wrong.

00:29:56.860 | - Yeah.

00:29:57.700 | - So the way you kind of calculate it is,

00:30:01.260 | you might add up insertions, deletions,

00:30:03.620 | and substitutions, right?

00:30:04.700 | So insertions is like extra words,

00:30:07.420 | deletions are words that we said,

00:30:08.940 | but weren't in the transcript, right?

00:30:12.300 | Substitutions is, you said Apple, but I said,

00:30:15.740 | but the ASR thought it was Able, something like this.

00:30:18.440 | Human accuracy, most people think realistically,

00:30:23.060 | it's like 3%, 2% word error rate

00:30:26.740 | would be like the max achievable.

00:30:28.540 | So there's still quite a gap, right?

00:30:31.500 | - Would you say that, so YouTube,

00:30:33.380 | when I upload videos often generates automatic captions.

00:30:36.980 | Are you sort of from a company perspective,

00:30:39.740 | from a tech perspective, are you trying to beat YouTube?

00:30:44.260 | Google, it's a hell of a, so Google,

00:30:47.180 | I mean, I don't know how seriously they take this task,

00:30:49.740 | but I imagine it's quite serious.

00:30:51.860 | And they, you know, Google is probably up there

00:30:56.260 | in terms of their teams on ASR,

00:31:01.260 | or just NLP, natural language processing,

00:31:03.180 | different technologies.

00:31:04.440 | So do you think you can beat Google?

00:31:06.660 | - On this kind of stuff, yeah, we think so.

00:31:08.980 | Google just woke up on my phone.

00:31:10.580 | - This is hilarious, okay.

00:31:12.980 | - Now Google is listening,

00:31:14.700 | sending it back to headquarters.

00:31:16.400 | Who are these rough people?

00:31:19.540 | - But that's the goal?

00:31:20.580 | - Yeah, I mean, we measure ourselves against like Google,

00:31:23.140 | Amazon, Microsoft, you know, some smaller competitors.

00:31:26.900 | And we use like our internal tests with it.

00:31:30.300 | We try to compose it of a pretty representative set of

00:31:33.060 | audios, maybe it's some podcasts, some videos,

00:31:36.380 | some interviews, some lectures, things like that, right?

00:31:39.700 | And we beat them in our own testing.

00:31:42.780 | - And actually Rev offers automated,

00:31:45.940 | like you can actually just do the automated captioning.

00:31:49.300 | So like, I guess it's like way cheaper, whatever it is,

00:31:52.700 | whatever the rates are.

00:31:54.180 | - Yeah, yeah.

00:31:55.660 | - By the way, it used to be a dollar per minute

00:31:57.900 | for captioning and transcription.

00:32:00.100 | I think it's like a dollar 15 or something like that.

00:32:02.340 | - Dollar 25.

00:32:03.180 | - Dollar 25.

00:32:04.260 | Dollar 25, no.

00:32:07.380 | Yeah, it's pretty cool.

00:32:09.260 | That was the other thing that was surprising to me.

00:32:10.940 | It was actually like the cheapest thing you could,

00:32:15.940 | I mean, I don't remember it being cheaper.

00:32:18.420 | You could on Upwork get cheaper,

00:32:20.940 | but it was clear to me that this,

00:32:22.500 | that's going to be really shitty.

00:32:23.980 | - Yeah.

00:32:24.820 | - So like, you're also competing on price.

00:32:26.900 | I think there were services that you can get like similar

00:32:30.900 | to Rev kind of feel to it, but it wasn't as automated.

00:32:35.820 | Like the drag and drop, the entirety of the interface.

00:32:37.900 | It's like the thing we're talking about.

00:32:39.580 | I'm such a huge fan of like frictionless,

00:32:41.700 | like Amazon's single buy button, whatever.

00:32:47.700 | - Yeah, yeah.

00:32:48.540 | - That one click, that's genius right there.

00:32:52.340 | Like that is so important for services.

00:32:54.940 | - Yeah.

00:32:55.780 | - That simplicity.

00:32:56.620 | And I mean, Rev is almost there.

00:33:00.460 | I mean, there's like some, trying to think.

00:33:04.380 | So I think I've, I stopped using this pipeline,

00:33:09.380 | but Rev offers it and I like it,

00:33:12.460 | but it was causing me some issues on my side,

00:33:16.220 | which is you can connect it to like Dropbox

00:33:20.340 | and it generates the files in Dropbox.

00:33:22.780 | So like it closes the loop to where I don't have to go

00:33:26.460 | to Rev at all and I can download it.

00:33:29.100 | Sorry, I don't have to go to Rev at all

00:33:32.700 | and to download the files.

00:33:34.220 | It could just like automatically copy them.

00:33:36.300 | - Right, you put in your Dropbox and you know,

00:33:38.580 | a day later or maybe a few hours later.

00:33:41.060 | - Yeah, it just shows up.

00:33:41.900 | - Depending on if you're in a rush, it just shows up, yeah.

00:33:44.100 | I was trying to do it programmatically too.

00:33:46.540 | Is there an API interface you can,

00:33:48.940 | I was trying to through like through Python

00:33:51.540 | to download stuff automatically,

00:33:53.460 | but then I realized this is the programmer in me.

00:33:56.180 | Like, dude, you don't need to automate everything

00:33:58.700 | like in life, like flawlessly.

00:34:01.140 | 'Cause I wasn't doing enough captions to justify

00:34:04.100 | to myself the time investment

00:34:05.660 | into automating everything perfectly.

00:34:07.820 | - Yeah, I would say if you're doing so many interviews

00:34:10.060 | that your biggest roadblock is clicking

00:34:13.300 | on the Rev download button.

00:34:15.500 | Now you're talking about Elon Musk levels of business.

00:34:18.980 | - But for sure we have like a variety of ways

00:34:22.020 | to make it easy.

00:34:22.860 | You know, there's the integration.

00:34:24.180 | You mentioned, I think it's through a company called Zapier,

00:34:26.220 | which kind of can connect Dropbox to Rev and vice versa.

00:34:31.140 | We have an API if you wanna really like customize it,

00:34:33.460 | you know, if you wanna create the Lex Friedman,

00:34:37.140 | you know, CMS or whatever.

00:34:40.900 | - For this whole thing, okay, cool.

00:34:42.300 | So can you speak to the ASR a little bit more?

00:34:46.460 | Like, what does it take like approach-wise,

00:34:51.460 | machine learning-wise, how hard is this problem?

00:34:54.980 | How do you get to the 3% error rate?

00:34:57.700 | Like, what's your vision of all of this?

00:34:59.340 | - Yeah, well, the 3% error rate is definitely,

00:35:03.180 | that's the grand vision.

00:35:05.220 | We'll see what it takes to get there.

00:35:09.860 | But we believe, you know, in ASR,

00:35:13.060 | the biggest thing is the data, right?

00:35:15.220 | Like, that's true of like a lot

00:35:16.420 | of machine learning problems today, right?

00:35:18.340 | The more data you have and the higher quality of the data,

00:35:21.060 | the better labeled the data.

00:35:22.820 | Yeah, that's how you get good results.

00:35:26.540 | And we at Rev have kind of like the best data, like we have.

00:35:29.820 | - Like you're literally, your business model

00:35:32.460 | is annotating the data.

00:35:34.020 | - Our business model is being paid to annotate the data.

00:35:36.780 | - Being paid to annotate the data.

00:35:39.140 | - So it's kind of like a pretty magical flywheel.

00:35:42.060 | - Yeah.

00:35:42.900 | - And so we've kind of like ridden this flywheel

00:35:44.540 | to this point.

00:35:47.060 | And we think we're still kind of in the early stages

00:35:50.540 | of figuring out all the parts of the flywheel to use,

00:35:53.100 | you know, because we have the final transcripts

00:35:56.420 | and we have the audios and we train on that.

00:36:01.660 | But we, in principle, also have all the edits

00:36:05.060 | that the Revvers make, right?

00:36:06.500 | - Oh, that's interesting.

00:36:08.700 | How can you use that as data?

00:36:10.540 | - We basically, that's something for us to figure out

00:36:12.580 | in the future, but you know,

00:36:14.380 | we feel like we're only in the early stages, right?

00:36:16.300 | - So the data is there, that'd be interesting,

00:36:18.620 | like almost like a recurrent neural net

00:36:20.820 | for fixing transcripts.

00:36:23.380 | I always remember we did a segmentation annotation

00:36:28.380 | for driving data, so segmenting the scene, like visual data.

00:36:33.220 | And you can get all, so it was drawing,

00:36:35.980 | people were drawing polygons around different objects

00:36:38.940 | and so on.

00:36:40.060 | And it feels like, it always felt like there was a lot

00:36:42.860 | of information in the clicking,

00:36:45.220 | the sequence of clicking that people do,

00:36:47.020 | the kind of fixing of the polygons that they do.

00:36:49.460 | Now there's a few papers written about how to draw polygons,

00:36:54.860 | like with recurrent neural nets to try to learn

00:36:59.220 | from the human clicking, but it was just like experimental,

00:37:04.380 | you know, it was one of those like CVPR type papers

00:37:06.660 | that people do like a really tiny data set.

00:37:08.980 | It didn't feel like people really tried to do it seriously.

00:37:13.100 | And I wonder, I wonder if there's information

00:37:15.140 | in the fixing that provides deeper set of signal

00:37:20.140 | than just like the raw data.

00:37:24.460 | - The intuition is for sure there must be, right?

00:37:26.220 | - There must be.

00:37:27.060 | - And in all kinds of signals and how long you took

00:37:29.700 | to make that edit and stuff like that.

00:37:32.700 | - Yeah, it's gonna be like up to us.

00:37:34.140 | That's why like the next couple of years

00:37:36.820 | is like super exciting for us, right?

00:37:38.340 | - So that's what like the focus is now.

00:37:40.340 | You mentioned Rev.ai, that's where you want to.

00:37:43.340 | - Yeah, so Rev.ai is kind of our way of bringing this ASR

00:37:48.340 | to the rest of the world, right?

00:37:51.580 | So when we started, we were human only,

00:37:55.660 | then we kind of created this TEMI service,

00:37:59.220 | I think you might've used it,

00:38:00.660 | which was kind of ASR for the consumer, right?

00:38:02.580 | So if you don't want to pay $1.25,

00:38:04.580 | but you want to pay, now it's 25 cents a minute, I think.

00:38:08.100 | And you get the transcript,

00:38:10.700 | the machine generated transcript,

00:38:12.740 | you get an editor and you can kind of fix it up yourself.

00:38:17.460 | Then we started using ASR for human transcriptionists.

00:38:21.980 | And then the kind of Rev.ai is the final step

00:38:23.460 | of the journey, which is, we have this amazing engine.

00:38:27.100 | What can people build with it, right?

00:38:28.860 | What kind of new applications could be enabled

00:38:32.340 | if you have SpeedTrack that's that accurate?

00:38:36.340 | - Do you have ideas for this

00:38:37.580 | or is it just providing it as a service

00:38:39.300 | and seeing what people come up with?

00:38:40.660 | - It's providing it as a service

00:38:41.980 | and seeing what people come up with

00:38:43.500 | and kind of learning from what people do with it.

00:38:45.580 | And we have ideas of our own as well, of course,

00:38:47.180 | but it's a little bit like,

00:38:49.220 | when AWS provided the building blocks, right?

00:38:52.580 | And they saw what people built with it

00:38:53.940 | and they try to make it easier to build those things, right?

00:38:57.020 | And we kind of hope to do the same thing.

00:38:59.180 | - Although AWS kind of does a shitty job of like,

00:39:02.860 | I'm continually surprised, like Mechanical Turk,

00:39:05.060 | for example, how shitty the interface is.

00:39:07.780 | We're talking about like Rev.ai making me feel good.

00:39:11.140 | Like when I first discovered Mechanical Turk,

00:39:14.180 | the initial idea of it was like,

00:39:18.260 | it made me feel like Rev.ai does,

00:39:19.660 | but then the interface is like, come on.

00:39:22.820 | - Yeah, it's horrible.

00:39:24.740 | - Why is it so painful?

00:39:27.740 | Does nobody at Amazon wanna like seriously invest in it?

00:39:32.500 | It felt like you can make so much money

00:39:34.980 | if you took this effort seriously.

00:39:37.260 | And it feels like they have a committee of like two people

00:39:40.220 | just sitting back, like a meeting,

00:39:42.900 | they meet once a month,

00:39:43.980 | like what are we gonna do with Mechanical Turk?

00:39:46.540 | It's like two websites make me feel like this,

00:39:49.260 | that and craiglist.org, whatever the hell it is.

00:39:53.620 | - It feels like it's designed in the 90s.

00:39:55.940 | - Well, craiglist basically hasn't been updated

00:39:59.140 | pretty much since the guy originally built.

00:39:59.980 | - Do you seriously think there's a team,

00:40:01.860 | like how big is the team working on Mechanical Turk?

00:40:04.220 | - I don't know, there's some team, right?

00:40:06.820 | - I feel like there isn't, I'm skeptical.

00:40:09.460 | - Yeah, well, if nothing else, they benefit from,

00:40:13.500 | you know, the other teams like moving things forward,

00:40:16.380 | right, in a small way.

00:40:18.380 | But no, I know what you mean,

00:40:19.740 | we use Mechanical Turk for a couple of things as well,

00:40:22.260 | and yeah, it's painful.

00:40:24.340 | - But yeah, it works.

00:40:25.700 | - I think most people, the thing is most people

00:40:27.540 | don't really use the UI, right?

00:40:29.140 | Like, so like we, for example, we use it through the API.

00:40:33.580 | - But even the API documentation and so on,

00:40:36.100 | like it's super outdated.

00:40:37.540 | I don't even know what to, I mean, same criticism,

00:40:45.100 | as long as we're ranting, my same criticism goes

00:40:49.060 | to the APIs of most of these companies,

00:40:50.940 | like Google, for example, the API for the different services

00:40:55.180 | is just the documentation is so shitty.

00:40:58.980 | Like, it's not so shitty, I should actually be,

00:41:04.860 | I should exhibit some gratitude.

00:41:08.380 | Okay, let's practice some gratitude.

00:41:10.900 | The, you know, the documentation is pretty good.

00:41:14.340 | Like most of the things that the API makes available

00:41:18.820 | is pretty good.

00:41:19.700 | It's just that in the sense that it's accurate,

00:41:23.100 | sometimes outdated, but like the degree of explanations

00:41:27.260 | with examples is only covering, I would say like 50%

00:41:32.260 | of what's possible.

00:41:33.900 | And it just feels a little bit like there's a lot

00:41:36.300 | of natural questions that people would wanna ask

00:41:38.980 | that doesn't get covered.

00:41:41.660 | And it feels like it's almost there.

00:41:44.580 | Like it's such a magical thing, like the Maps API,

00:41:48.540 | YouTube API, there's a bunch of stuff.

00:41:51.180 | - I gotta imagine it's like, you know,

00:41:52.780 | there's probably some team at Google, right,

00:41:55.580 | responsible for writing this documentation.

00:41:57.500 | That's probably not the engineers, right?

00:42:00.300 | And probably this team is not, you know,

00:42:03.260 | where you wanna be.

00:42:04.460 | - Well, it's a weird thing.

00:42:05.860 | I sometimes think about this for somebody who wants

00:42:09.620 | to also build the company.

00:42:12.180 | I think about this a lot.

00:42:15.820 | You know, YouTube, the service is one of the most magical,

00:42:20.820 | like I'm so grateful that YouTube exists.

00:42:24.500 | And yet they seem to be quite clueless on so many things

00:42:29.500 | like that everybody's screaming them at.

00:42:33.420 | Like it feels like whatever the mechanism that you use

00:42:38.100 | to listen to your quote unquote customers,

00:42:40.100 | which is like the creators is not very good.

00:42:44.820 | Like there's literally people that are like screaming,

00:42:47.300 | like their new YouTube studio, for example.

00:42:51.060 | There's like features that were like begged for,

00:42:55.180 | for a really long time,

00:42:56.940 | like being able to upload multiple videos at the same time.

00:43:00.180 | That was missing for a really, really long time.

00:43:03.940 | Now, like there's probably things that I don't know,

00:43:08.020 | which is maybe for that kind of huge infrastructure,

00:43:10.980 | it's actually very difficult to build some of these features.

00:43:13.780 | But the fact that that wasn't communicated

00:43:15.580 | and it felt like you're not being heard.

00:43:19.180 | Like I remember this experience for me

00:43:21.580 | and it's not a pleasant experience.

00:43:23.860 | And it feels like the company doesn't give a damn about you.

00:43:26.780 | And that's something to think about.

00:43:28.220 | I'm not sure what that is.

00:43:30.020 | That might have to do with just like small groups

00:43:32.540 | working on these small features and these specific features.

00:43:35.940 | And there's no overarching like dictator type of human

00:43:40.340 | that says like, why the hell are we neglecting

00:43:42.460 | like Steve Jobs type of characters?

00:43:43.940 | Like there's people that we need to speak to the people

00:43:48.940 | that like wanna love our product and they don't.

00:43:51.700 | Let's fix this shit. - Yeah, I mean,

00:43:52.540 | at some point you just get so fixated on the numbers.

00:43:54.900 | And it's like, well, the numbers are pretty great.

00:43:56.980 | Like people are watching,

00:43:58.660 | doesn't seem to be a problem.

00:44:01.060 | - Doesn't seem to be a problem.

00:44:01.980 | - And you're not like the person that like build this thing.

00:44:04.260 | So you really care about it.

00:44:05.860 | You're just there, you came in as a product manager.

00:44:09.140 | You got hired sometime later,

00:44:10.700 | your mandate is like increase this number like 10%, right?

00:44:15.700 | And you just-- - That's brilliantly put.

00:44:17.540 | Like if you, this is, okay, if there's a lesson in this,

00:44:21.380 | is don't reduce your company into a metric of like,

00:44:25.340 | how much, like you said,

00:44:27.820 | how much people watching the videos and so on,

00:44:31.020 | and like convince yourself that everything is working

00:44:33.860 | just because the numbers are going up.

00:44:36.220 | There's something, you have to have a vision.

00:44:39.140 | You have to want people to love your stuff

00:44:43.420 | because love is ultimately the beginning

00:44:46.140 | of like a successful long-term company

00:44:49.260 | is that they always should love your product.

00:44:51.340 | - You have to be like a creator

00:44:52.620 | and have that like creator's love for your own thing, right?

00:44:55.420 | Like, and you paint by, you know, these comments, right?

00:44:59.580 | And probably like, Apple, I think did this generally

00:45:02.460 | like really well. - Yes, really well.

00:45:03.860 | - They're well known for kind of keeping teams small,

00:45:06.860 | even when they were big, right?

00:45:08.220 | And, you know, he was an engineer,

00:45:10.380 | like there's that book, "Creative Selection."

00:45:12.700 | I don't know if you read it by an Apple engineer

00:45:15.460 | named Ken Kosienda.

00:45:17.300 | It's kind of a great book actually,

00:45:18.300 | because unlike most of these business books where it's,

00:45:21.420 | you know, here's how Steve Jobs ran the company.

00:45:24.580 | It's more like, here's how life was like for me,

00:45:26.940 | you know, an engineer.

00:45:27.780 | Here are the projects I worked on

00:45:29.020 | and here what it was like to pitch Steve Jobs, you know,

00:45:31.660 | on like, you know, I think it was in charge of like

00:45:34.620 | the keyboard and the auto correction, right?

00:45:36.860 | And at Apple, like Steve Jobs reviewed everything.

00:45:39.420 | And so he was like, this is what it was like

00:45:41.140 | to show my demos to Steve Jobs and, you know,

00:45:43.740 | to change them because like Steve Jobs didn't like how,

00:45:46.580 | you know, the shape of the little key was off

00:45:48.780 | because the rounding of the corner was like not quite right

00:45:50.900 | or something like this, but he was famously a stickler

00:45:53.380 | for this kind of stuff.

00:45:54.620 | But because the teams were small,

00:45:55.820 | he really owned this stuff, right?

00:45:56.900 | So he really cared.

00:45:58.660 | - Yeah, Elon Musk does that similar kind of thing with Tesla,

00:46:01.620 | which is really interesting.

00:46:03.380 | There's another lesson in leadership in that

00:46:05.900 | is to be obsessed with the details.

00:46:07.660 | And like, he talks to like the lowest level engineers.

00:46:11.300 | Okay, so we're talking about ASR.

00:46:14.620 | And so this is basically where I was saying,

00:46:17.660 | we're gonna take this like ultra seriously.

00:46:20.380 | And then what's the mission?

00:46:22.660 | To try to keep pushing towards the 3%?

00:46:24.980 | - Yeah, and kind of try to build this platform

00:46:30.340 | where all of your, you know, all of your meetings,

00:46:33.940 | you know, they're as easily accessible as your notes, right?

00:46:38.460 | Like, so like imagine all the meetings

00:46:41.380 | a company might have, right?

00:46:42.780 | Now that I'm like no longer a programmer, right?

00:46:46.340 | And I'm a quote unquote manager,

00:46:48.100 | that's less like my day is in meetings, right?

00:46:51.460 | And, you know, pretty often I wanna like see what was said,

00:46:54.860 | right, who said it, you know, what's the context.

00:46:57.100 | But it's generally not really something

00:46:59.420 | that you can easily retrieve, right?

00:47:00.500 | Like imagine if all of those meetings

00:47:03.220 | were indexed, archived, you know, you could go back,

00:47:05.780 | you could share a clip like really easily, right?

00:47:08.300 | - So that might change completely.

00:47:10.060 | Like everything that's said converted to text

00:47:12.940 | might change completely the dynamics

00:47:14.860 | of what we do in this world.

00:47:16.340 | Especially now with remote work, right?

00:47:18.180 | - Exactly, exactly.

00:47:19.980 | - With Zoom and so on.

00:47:21.460 | That's fascinating to think about.

00:47:22.740 | I mean, for me, I care about podcasts, right?

00:47:25.580 | And one of the things that was, you know, I'm torn.

00:47:31.140 | I know a lot of the engineers at Spotify.

00:47:33.580 | So I love them very much because they dream big

00:47:38.580 | in terms of like, they wanna empower creators.

00:47:43.340 | So one of my hopes was with Spotify

00:47:45.020 | that they would use a technology like Rev

00:47:46.780 | or something like that to start converting everything

00:47:51.660 | into text and make it indexable.

00:47:55.180 | Like one of the things that sucks with podcasts

00:47:59.500 | is like, it's hard to find stuff.

00:48:01.780 | Like the model is basically subscription.

00:48:04.460 | Like you find, it's similar to what YouTube used to be like,

00:48:09.460 | which is you basically find a creator that you enjoy

00:48:14.220 | and you subscribe to them and like, you just,

00:48:16.420 | you just kind of follow what they're doing.

00:48:19.700 | But the search and discovery wasn't a big part of YouTube

00:48:24.260 | like in the early days.

00:48:25.500 | But that's what currently with podcasts,

00:48:28.500 | like is the search and discovery is like non-existent.

00:48:33.500 | You're basically searching for like

00:48:35.260 | the dumbest possible thing,

00:48:36.420 | which is like keywords in the titles of episodes.

00:48:39.660 | - Yeah, but even aside from searching,

00:48:41.220 | it's kind of like all the time.

00:48:42.180 | So I listened to like a number of podcasts

00:48:44.140 | and there's something sad

00:48:46.860 | and I wanna like go back to that later

00:48:48.580 | because I was trying to, I'm trying to remember,

00:48:49.820 | what do you say?

00:48:50.660 | Like maybe like recommend some cool product

00:48:52.180 | that I wanna try out.

00:48:53.460 | And like, it's basically impossible.

00:48:54.700 | Maybe like some people have pretty good show notes.

00:48:56.780 | So maybe you'll get lucky and you can find it, right?

00:48:59.020 | But I mean, if everyone had transcripts

00:49:01.580 | and it was all searchable, it would be--

00:49:03.340 | - It's a game changer.

00:49:04.300 | - It'd be so much better.

00:49:05.300 | - I mean, that's one of the things that I wanted to,

00:49:08.460 | I mean, one of the reasons we're talking today

00:49:11.020 | is I wanted to take this quite seriously,

00:49:13.380 | the rev thing, I've just been lazy.

00:49:15.700 | So because I'm very fortunate

00:49:19.460 | that a lot of people support this podcast,

00:49:21.220 | that there's enough money now to do a transcription

00:49:24.140 | and so on, it seemed clear to me,

00:49:28.300 | especially like CEOs and sort of like PhDs,

00:49:32.940 | like people write to me

00:49:36.420 | who are like graduate students in computer science

00:49:38.340 | or graduate students in whatever the heck field.

00:49:41.140 | It's clear that their mind,

00:49:43.140 | like they enjoy podcasts when they're doing laundry

00:49:45.220 | or whatever, but they wanna revisit the conversation

00:49:48.780 | in a much more rigorous way.

00:49:50.780 | And they really wanna transcript.

00:49:53.340 | It's clear that they want to like analyze conversations.

00:49:56.780 | So many people wrote to me about a transcript

00:49:59.300 | for Yosha Bach conversation.

00:50:01.060 | I had just a bunch of conversations.

00:50:03.740 | And then on the Elon Musk side,

00:50:05.820 | like reporters want like,

00:50:08.260 | they wanna write a blog post about your conversation.

00:50:10.820 | So they wanna be able to pull stuff.

00:50:13.060 | And it's like, they're essentially doing

00:50:15.500 | on your conversation transcription privately.

00:50:18.340 | They're doing it for themselves and then starting to pick,

00:50:21.780 | but it's so much easier when you can actually do it

00:50:23.940 | as a reporter, just look at the transcript.

00:50:26.180 | - Yeah, and you can like embed a little thing,

00:50:28.140 | you know, like into your article, right?

00:50:29.580 | Here's what they said.

00:50:30.500 | You can go listen to like this clip from the section.

00:50:33.580 | - I'm actually trying to figure out,

00:50:35.940 | I'll probably on the website create like a place

00:50:40.260 | where the transcript goes like as a webpage

00:50:42.460 | so that people can reference it,

00:50:44.340 | like reporters can reference it and so on.

00:50:46.700 | I mean, most of the reporters probably

00:50:49.660 | have wanted right clickbait articles

00:50:52.740 | that are complete falsifying, which I'm fine with.

00:50:55.380 | It's the way of journalism, I don't care.

00:50:57.740 | Like I've had this conversation with a friend of mine,

00:51:01.660 | a mixed martial artist, Ryan Hall.

00:51:03.920 | And we talked about, you know,

00:51:07.100 | as I've been reading the rise and fall of the Third Reich

00:51:09.620 | and a bunch of books on Hitler.

00:51:11.820 | And we brought up Hitler and he made some kind of comment

00:51:16.360 | where like we should be able to forgive Hitler.

00:51:19.580 | And, you know, like we were talking about forgiveness

00:51:23.700 | and we're bringing that up

00:51:24.700 | as like the worst case possible things.

00:51:26.760 | Like even, you know, for people who are Holocaust survivors,

00:51:31.760 | one of the ways to let go of the suffering

00:51:34.700 | they've been through is to forgive.

00:51:38.060 | And he brought up like Hitler is somebody

00:51:39.780 | that would potentially be the hardest thing

00:51:42.460 | to possibly forgive,

00:51:43.340 | but it might be a worthwhile pursuit psychologically.

00:51:47.080 | So on, blah, blah, blah, it doesn't matter.

00:51:48.560 | It was very eloquent, very powerful words.

00:51:50.860 | I think people should go back and listen to it.

00:51:53.160 | It's powerful.

00:51:54.000 | And then all these journalists,

00:51:55.680 | all these articles written about like MMA fight,

00:51:58.960 | UFC fight.

00:51:59.800 | - MMA fighter loves Hitler.

00:52:01.920 | - No, like, well, no, they were somewhat accurate.

00:52:05.720 | They didn't say like loves Hitler.

00:52:07.120 | They said, thinks that if Hitler came back to life

00:52:13.280 | we should forgive him.

00:52:14.440 | Like they kind of, it's kind of accurate-ish,

00:52:18.520 | but it, the headline made it sound a lot worse

00:52:23.520 | than it was, but I'm fine with it.

00:52:27.820 | That's the way the world,

00:52:29.740 | I wanna almost make it easier for those journalists

00:52:32.680 | and make it easier for people who actually care

00:52:34.880 | about the conversation to go and look and see.

00:52:37.320 | - Right, they can see it for themselves.

00:52:38.560 | - For themselves, full context.

00:52:39.400 | - There's the headline, but now you can go.

00:52:41.680 | - There's something about podcasts,

00:52:42.960 | like the audio that makes it difficult to go,

00:52:46.680 | to jump to a spot and to look for that,

00:52:50.720 | for that particular information.

00:52:53.200 | I think some of it, you know, I'm interested in creating

00:52:56.920 | like myself experimenting with stuff.

00:53:00.320 | So like taking Rev and creating a transcript

00:53:03.400 | and then people can go to it.

00:53:05.160 | I do dream that like, I'm not in the loop anymore,

00:53:09.320 | that like, you know, Spotify does it, right?

00:53:12.960 | Like automatically for everybody,

00:53:16.320 | because ultimately that one-click purchase

00:53:19.600 | needs to be there, like, you know.

00:53:21.680 | - I mean, like you kind of want support

00:53:22.760 | from the entire ecosystem, right?

00:53:24.080 | - Exactly.

00:53:24.920 | - Like from the tool makers and the podcast creators,

00:53:27.920 | even clients, right?

00:53:28.760 | I mean, imagine if like most podcast apps,

00:53:33.760 | you know, if it was a standard, right?

00:53:35.800 | Here's how you include a transcript into a podcast, right?

00:53:38.520 | Podcast is just an RSS feed ultimately.

00:53:40.680 | And actually just yesterday I saw this company called

00:53:44.480 | Buzzsprout, I think they're called.

00:53:46.640 | So they're trying to do this.

00:53:48.320 | They proposed a spec, an extension to their RSS format

00:53:53.320 | to reference podcasts, reference transcripts

00:53:56.560 | in a standard way.

00:53:58.080 | And they're talking about like,

00:53:59.080 | there's one client dimension that will support it,

00:54:02.160 | but imagine like more clients support it, right?

00:54:04.040 | So any podcast you could go and see the transcripts, right?

00:54:08.440 | On your like normal podcast app.

00:54:10.480 | - Yeah, I mean, somebody,

00:54:11.840 | so I have somebody who works with me,

00:54:14.200 | works with, helps with advertising, Matt,

00:54:19.160 | this awesome guy.

00:54:20.240 | He mentioned Buzzsprout to me,

00:54:21.600 | but he says it's really annoying

00:54:23.120 | 'cause they want exclusive,

00:54:24.920 | they wanna host the podcast.

00:54:26.320 | - Right.

00:54:27.160 | - This is the problem with Spotify too.

00:54:29.280 | This is where I'd like to say like F Spotify.

00:54:33.960 | There's a magic to RSS with podcasts.

00:54:37.560 | It can be made available to everyone.

00:54:40.360 | And then there's all,

00:54:41.320 | there's this ecosystem of different podcast players

00:54:44.560 | that emerge and they compete freely.

00:54:47.120 | And that's a beautiful thing.

00:54:48.960 | That's why I go on exclusive,

00:54:50.440 | like Joe Rogan went exclusive.

00:54:52.020 | I'm not sure if you're familiar with,

00:54:54.800 | he went to just Spotify.

00:54:56.400 | As a huge fan of Joe Rogan,

00:54:59.600 | I've been kind of nervous about the whole thing,

00:55:01.400 | but let's see.

00:55:03.000 | I hope that Spotify steps up.

00:55:05.040 | They've added video,

00:55:06.000 | which is very surprising that they were able to--

00:55:07.760 | - So exclusive meaning you can't subscribe

00:55:10.440 | to his RSS feed anymore.

00:55:11.760 | It's only in Spotify.

00:55:12.720 | - For now, you can until December 1st.

00:55:15.760 | And December 1st, it's all,

00:55:17.760 | everything disappears and it's Spotify only.

00:55:20.180 | I, you know, and Spotify gave him $100 million for that.

00:55:25.840 | So it's an interesting deal,

00:55:28.360 | but I, you know, I did some soul searching

00:55:31.200 | and I'm glad he's doing it.

00:55:34.520 | But if Spotify came to me with $100 million,

00:55:37.160 | I wouldn't do it.

00:55:40.120 | I wouldn't do, well,

00:55:40.960 | I have a very different relationship with money.

00:55:42.680 | I hate money, but I just think,

00:55:46.120 | I believe in the pirate radio aspect of podcasting,

00:55:48.960 | the freedom and that there's something--

00:55:51.080 | - The open source spirit.

00:55:52.320 | - The open source spirit, it just doesn't seem right.

00:55:54.720 | It doesn't feel right.

00:55:55.820 | That said, you know,

00:55:57.320 | because so many people care about Joe Rogan's program,

00:56:00.480 | they're gonna hold Spotify's feet to the fire.

00:56:02.980 | Like one of the cool things,

00:56:05.040 | what Joe told me is the reason he likes working with Spotify

00:56:10.040 | is that they're like ride or die together, right?

00:56:15.440 | So they want him to succeed.

00:56:19.160 | So that's why they're not actually telling him what to do,

00:56:22.080 | despite what people think.

00:56:23.800 | They don't tell him,

00:56:25.000 | they don't give him any notes on anything.

00:56:26.940 | They want him to succeed.

00:56:28.520 | And that's the cool thing about exclusivity

00:56:30.920 | with a platform is like,

00:56:33.440 | you kind of want each other to succeed.

00:56:36.800 | And that process can actually be very fruitful.

00:56:39.720 | Like YouTube, it goes back to my criticism.

00:56:43.400 | YouTube generally, no matter how big the creator,

00:56:47.840 | maybe for PewDiePie, something like that,

00:56:50.200 | they want you to succeed.

00:56:51.700 | But for the most part,

00:56:52.760 | from all the big creators I've spoken with,

00:56:54.680 | Veritasium, all those folks,

00:56:57.080 | you know, they get some basic assistance,

00:56:59.000 | but it's not like YouTube doesn't care

00:57:02.800 | if you succeed or not.

00:57:03.840 | They have so many creators.

00:57:04.680 | - They have like a hundred other.

00:57:06.520 | - They don't care.

00:57:07.600 | So, and especially with somebody like Joe Rogan,

00:57:12.600 | who YouTube sees Joe Rogan,

00:57:15.080 | not as a person who might revolutionize the nature of news

00:57:19.760 | and idea space and nuanced conversations.

00:57:23.900 | They see him as a potential person

00:57:26.280 | who has racist guests on,

00:57:30.240 | or like, you know,

00:57:31.800 | they see him as like a headache potentially.

00:57:34.440 | So, you know, a lot of people talk about this.

00:57:37.960 | It's a hard place to be for YouTube actually,

00:57:40.600 | is figuring out with the search and discovery process

00:57:45.600 | of how do you filter out conspiracy theories

00:57:49.040 | and which conspiracy theories represent dangerous untruths

00:57:53.360 | and which conspiracy theories are like vanilla untruths.

00:57:58.080 | And then even when you start having meetings

00:58:00.620 | and discussions about what is true or not,

00:58:03.560 | it starts getting weird.

00:58:05.080 | - Yeah.

00:58:05.920 | - It starts getting weird.

00:58:06.760 | - It's difficult these days, right?

00:58:07.800 | I worry more about the other side, right?

00:58:09.720 | Of too much, you know, too much not censorship.

00:58:13.240 | Well, maybe censorship is the right word.

00:58:14.640 | I mean, censorship is usually government censorship,

00:58:17.960 | but still, yeah, putting yourself in a position of arbiter

00:58:21.980 | for these kinds of things.

00:58:22.920 | - Yeah.

00:58:23.760 | - It's very difficult.

00:58:24.580 | And people think it's so easy, right?

00:58:25.420 | Like, it's like, well, you know, like no Nazis, right?

00:58:27.840 | What a simple principle.

00:58:29.100 | But, you know, yes, I mean, no one likes Nazis.

00:58:32.760 | - Yeah.

00:58:33.600 | - But there's like many shades of gray,

00:58:35.240 | like very soon after that.

00:58:37.400 | - Yeah, and then, you know, of course everybody,

00:58:39.440 | you know, there's some people that call

00:58:40.800 | our current president a Nazi.

00:58:42.240 | And then there's like, so you start getting Sam Harris.

00:58:45.720 | I don't know if you know that is wasted, in my opinion,

00:58:49.680 | his conversation with Jack Dorsey.

00:58:51.640 | Now, I spoke with Jack before on this podcast,

00:58:54.200 | and we'll talk again.

00:58:55.680 | But Sam brought up, Sam Harris does not like Donald Trump.

00:59:00.080 | - I do listen to his podcast.

00:59:03.760 | I'm familiar with his views on the matter.

00:59:06.480 | - And he asked Jack Dorsey, he's like,

00:59:08.960 | how can you not ban Donald Trump from Twitter?

00:59:12.280 | And so, you know, there's a set,

00:59:13.920 | you have that conversation.

00:59:15.980 | You have a conversation where some number,

00:59:18.240 | some significant number of people

00:59:20.520 | think that the current president of the United States

00:59:22.920 | should not be on your platform.

00:59:24.920 | And it's like, okay, so if that's even on the table

00:59:28.060 | as a conversation, then everything's on the table

00:59:31.360 | for conversation.

00:59:32.760 | And yeah, it's tough.

00:59:34.680 | I'm not sure where I land on it.

00:59:37.040 | I'm with you, I think that censorship is bad,

00:59:39.480 | but I also think--

00:59:41.840 | - Ultimately, I just also think, you know,

00:59:44.000 | if you're the kind of person that's gonna be convinced,

00:59:46.560 | you know, by some YouTube video, you know,

00:59:49.460 | that, I don't know, our government's been taken over

00:59:53.160 | by aliens, it's unlikely that, like, you know,

00:59:56.120 | you'll be returned to sanity simply because, you know,

00:59:59.000 | that video is not available on YouTube, right?

01:00:02.040 | - Yeah, I'm with you.

01:00:02.860 | I tend to believe in the intelligence of people

01:00:04.560 | and we should trust them.

01:00:07.000 | But I also do think it's the responsibility of platforms

01:00:10.980 | to encourage more love in the world,

01:00:12.640 | more kindness to each other.

01:00:14.160 | And I don't always think that they're great

01:00:16.960 | at doing that particular thing.

01:00:19.280 | So that, there's a nice balance there.

01:00:24.280 | And I think philosophically, I think about that a lot.

01:00:28.280 | Where's the balance between free speech

01:00:31.240 | and like encouraging people,

01:00:35.040 | even though they have the freedom of speech

01:00:37.940 | to not be an asshole.

01:00:39.560 | - Yeah, right.

01:00:41.060 | - That's not a constitutional, like,

01:00:42.860 | so you have the right for free speech,

01:00:48.140 | but like, just don't be an asshole.

01:00:50.700 | Like, you can't really put that in the constitution

01:00:52.680 | that the Supreme Court can't be like,

01:00:54.480 | just don't be a dick.

01:00:56.100 | But I feel like platforms have a role to be like,

01:00:59.580 | just be nicer.

01:01:00.860 | Maybe do the carrot, like encourage people to be nicer

01:01:04.220 | as opposed to the stake of censorship.

01:01:06.820 | But I think it's an interesting machine learning problem.

01:01:11.060 | Just be nicer.

01:01:12.040 | - Machine, yeah, machine learning for niceness.

01:01:15.800 | - It is, I mean--

01:01:16.640 | - Responsible AI, I mean, it is a thing for sure.

01:01:20.160 | - Jack Dorsey kind of talks about it as a vision for Twitter

01:01:23.760 | is how do we increase the health of conversations?

01:01:26.840 | I don't know how seriously

01:01:28.120 | they're actually trying to do that though,

01:01:30.800 | which is one of the reasons I am in part considering

01:01:35.800 | entering that space a little bit.

01:01:37.320 | - It's difficult for them, right?

01:01:38.560 | Because it's kind of like well known that,

01:01:42.000 | people are kind of driven by rage and outrage maybe

01:01:47.000 | is a better word, right?

01:01:49.440 | Outrage drives engagement and well,

01:01:53.080 | these companies are judged by engagement, right?

01:01:56.040 | - In the short term, but this goes to the metrics thing

01:01:58.200 | that we were talking about earlier.

01:01:59.360 | I do believe, I have a fundamental belief that

01:02:03.480 | if you have a metric of long-term happiness of your users,

01:02:07.960 | like not short-term engagement,

01:02:09.560 | but long-term happiness and growth

01:02:11.480 | and both like intellectual, emotional health of your users,

01:02:15.520 | you're going to make a lot more money.

01:02:17.600 | You're going to have long,

01:02:18.800 | like you should be able to optimize for that.

01:02:21.360 | You don't need to necessarily optimize for engagement.

01:02:24.240 | - Yeah. - And that'll be good

01:02:25.080 | for society too.

01:02:26.360 | - Yeah, no, I mean, I generally agree with you,

01:02:28.760 | but it requires a patient person with,

01:02:31.960 | trust from Wall Street to be able to carry out

01:02:36.000 | such a strategy.

01:02:36.840 | - This is what I believe the Steve Jobs character

01:02:39.200 | and Elon Musk character is like,

01:02:41.920 | you basically have to be so good at your job.

01:02:45.240 | - Right, you got to pass for anything.

01:02:46.960 | - That you can hold the board

01:02:48.680 | and all the investors hostage by saying like,

01:02:52.000 | either we do it my way or I leave.

01:02:56.360 | And everyone is too afraid of you to leave

01:02:59.120 | 'cause they believe in your vision.

01:03:00.480 | So that, but that requires being really good

01:03:02.720 | at what you do.

01:03:04.320 | - Requires being Steve Jobs and Elon Musk.

01:03:06.680 | - There's kind of a reason why like a third name

01:03:08.520 | doesn't come immediately to mind, right?

01:03:10.840 | Like there's maybe a handful of other people,

01:03:12.360 | but it's not that many.

01:03:13.400 | - It's not many.

01:03:14.240 | I mean, people say like, why,

01:03:15.480 | like people say that I'm like a fan of Elon Musk.

01:03:18.320 | I'm not, I'm a fan of anybody

01:03:20.960 | who's like Steve Jobs and Elon Musk.

01:03:23.080 | And there's just not many of those folks.

01:03:26.320 | - It's the guy that made us believe

01:03:27.640 | that like we can get to Mars, you know, in 10 years, right?

01:03:31.040 | I mean, that's kind of awesome.

01:03:32.480 | - And it's kind of making it happen, which is like.

01:03:36.640 | - It's great.

01:03:37.480 | - It's kind of gone like that kind of like spirit, right?

01:03:40.520 | Like from a lot of our society, right?

01:03:42.280 | You know, like we can get to the moon in 10 years

01:03:44.680 | and like we did it, right?

01:03:45.720 | - Yeah, especially in this time of so much kind of

01:03:50.720 | existential dread that people are going through

01:03:53.840 | because of COVID, like having rockets

01:03:56.680 | that just keep going out there now with humans.

01:04:00.440 | I don't know that it's just like you said,

01:04:03.240 | I mean, it gives you a reason to wake up in the morning

01:04:05.560 | and dream, for us engineers too.

01:04:08.620 | It is inspiring as hell, man.

01:04:13.200 | Well, let me ask you this, the worst possible question,

01:04:17.160 | which is, so you're like at the core, you're a programmer,

01:04:21.400 | you're an engineer, but now you made the unfortunate choice

01:04:26.400 | or maybe that's the way life goes

01:04:30.760 | of basically moving away from the low level work

01:04:35.160 | and becoming a manager, becoming an executive,

01:04:38.120 | having meetings, what's that transition been like?

01:04:43.120 | - It's been interesting, it's been a journey.

01:04:44.920 | Maybe a couple of things to say about that.

01:04:47.120 | I got into this, right?

01:04:49.320 | Because as a kid, I just remember this like incredible

01:04:54.320 | amazement at being able to write a program, right?

01:04:57.400 | And something comes to life that kind of didn't exist before.

01:05:01.280 | I don't think you have that in like many other fields.

01:05:03.960 | Like you have that with some other kinds of engineering,

01:05:07.920 | but you may be a little bit more limited

01:05:09.680 | with what you can do, right?

01:05:10.720 | But with a computer, you can literally imagine

01:05:12.600 | any kind of program, right?

01:05:14.800 | So it's a little bit God-like what you do

01:05:17.000 | like when you create it.

01:05:18.200 | And so, I mean, that's why I got into it.

01:05:21.360 | - Do you remember like first program you wrote

01:05:23.240 | or maybe the first program that like made you fall in love

01:05:25.840 | with computer science?

01:05:28.040 | - I don't know if it was the first program.

01:05:29.440 | It's probably like trying to write one of those games

01:05:31.880 | and basic, you know, like emulate the snake game or whatever.

01:05:35.400 | I don't remember to be honest, but I enjoyed like,

01:05:37.840 | that's why I always loved about, you know,

01:05:40.000 | being a programmer is just the creation process.

01:05:41.840 | And it's a little bit different when you're not the one

01:05:45.160 | doing the creating.

01:05:46.200 | And, you know, another aspect to it I would say is,

01:05:50.520 | you know, when you're a programmer,

01:05:52.080 | when you're an individual contributor,

01:05:54.200 | it's kind of very easy to know when you're doing a good job,

01:05:57.840 | when you're not doing a good job,

01:05:58.680 | when you're being productive,

01:05:59.520 | when you're not being productive, right?

01:06:00.400 | You can kind of see like you trying to make something

01:06:03.000 | and it's like slowly coming together, right?

01:06:05.560 | And when you're a manager, you know, it's more diffuse,

01:06:08.880 | right?

01:06:09.720 | Like, well, you hope, you know, you're motivating your team

01:06:12.760 | and making them more productive and inspiring them, right?

01:06:15.920 | But it's not like you get some kind of like dopamine signal

01:06:18.920 | because you like completed X lines of code, you know, today.

01:06:22.440 | So kind of like you missed that dopamine rush a little bit

01:06:25.240 | when you first become, but then, you know,

01:06:28.480 | slowly you kind of see, yes,

01:06:30.640 | your teams are doing amazing work, right?

01:06:32.320 | And you can take pride in that.

01:06:34.640 | - You can get like, what is it?

01:06:38.200 | Like a ripple effect of somebody else's dopamine rush.

01:06:41.600 | - Yeah, yeah, yeah.

01:06:42.680 | You live off other people's dopamine.

01:06:44.560 | - So is there pain points and challenges you had to overcome

01:06:50.760 | from going to a programmer to becoming a programmer

01:06:54.880 | of humans?

01:06:55.960 | - Programmer of humans.

01:06:58.360 | I don't know, humans are difficult to understand, you know?

01:07:01.520 | It's like one of those things,

01:07:03.680 | like trying to understand other people's motivations

01:07:06.760 | and what really drives them.

01:07:08.240 | It's difficult, maybe like never really know, right?

01:07:10.880 | - Do you find that people are different?

01:07:13.360 | - Yeah.

01:07:14.200 | - Like I, one of the things,

01:07:15.840 | like I had a group at MIT that, you know,

01:07:21.560 | I found that like some people I could like scream at

01:07:28.280 | and criticize like hard,

01:07:30.920 | and that made them do like much better work

01:07:33.600 | and really push them to their limit.

01:07:35.960 | And there's some people that I had to nonstop compliment

01:07:39.840 | because like they're so already self-critical,

01:07:43.520 | like about everything they do,

01:07:45.240 | that I have to be constantly like,

01:07:47.300 | like I cannot criticize them at all

01:07:51.520 | because they're already criticizing themselves.

01:07:53.480 | And you have to kind of encourage

01:07:55.320 | and like celebrate their little victories.

01:07:58.720 | And it's kind of fascinating,

01:07:59.960 | like how that, the complete difference in people.

01:08:04.120 | - Definitely people will respond to different motivations

01:08:07.040 | and different modes of feedback,

01:08:08.280 | and you kind of have to figure it out.

01:08:11.360 | It was like a pretty good book,

01:08:13.720 | which for some reason now the name escapes me,

01:08:16.080 | about management, "First Break All the Rules."

01:08:18.880 | - "First Break All the Rules?"

01:08:19.720 | - "First Break All the Rules."

01:08:20.920 | It's a book that we generally like ask a lot of

01:08:24.000 | like first time managers to read a ref.

01:08:26.400 | Like one of the kind of philosophies

01:08:28.800 | is managed by exception, right?

01:08:31.120 | Which is, don't like have some standard template,

01:08:34.480 | like here's how I tell this person to do this

01:08:38.560 | or the other thing, here's how I get feedback,

01:08:40.040 | like managed by exception, right?

01:08:41.280 | Every person is a little bit different,

01:08:42.800 | you have to try to understand what drives them

01:08:45.360 | and tailor it to them.

01:08:47.240 | - Since you mentioned books,

01:08:48.920 | I don't know if you can answer this question,

01:08:50.840 | but people love it when I ask it,

01:08:52.480 | which is, are there books, technical, fiction,

01:08:55.880 | or philosophical that you enjoyed

01:08:58.560 | or had an impact on your life that you would recommend?

01:09:01.360 | You already mentioned "Dune," like all of the "Dune."

01:09:04.440 | - All of the "Dune."

01:09:05.400 | The second one was probably the weakest,

01:09:06.760 | but anyway, so yeah, all of the "Dune" is good.

01:09:09.800 | - I mean, yeah, can you just slow little tangent on that?

01:09:13.320 | How many "Dune" books are there?

01:09:16.320 | Like, do you recommend people start with the first one

01:09:18.480 | if that was--

01:09:19.920 | - Yeah, you kind of have to read them all.

01:09:21.080 | I mean, it is a complete story, right?

01:09:23.280 | So you start with the first one,

01:09:25.520 | you gotta read all of them.

01:09:27.560 | - There's not like a tree,

01:09:28.840 | like a creation of the universe

01:09:32.360 | that you should go and sequence?

01:09:33.920 | - You should go and sequence, yeah.

01:09:35.240 | It's kind of a chronological storyline.

01:09:38.040 | There's six books in all.

01:09:39.280 | Then there's like many kind of books

01:09:43.760 | that were written by Frank Herbert's son,

01:09:47.900 | but those are not as good,

01:09:48.880 | so you don't have to bother with those.

01:09:50.920 | - Shots fired.

01:09:51.760 | - Shots fired.

01:09:52.960 | - Okay.

01:09:53.800 | - But the main sequence is good.

01:09:56.320 | So what are some other books?

01:09:57.800 | Maybe there's a few.

01:09:59.720 | So I don't know that like I would say

01:10:01.400 | there's a book that kind of, I don't know,

01:10:04.640 | turned my life around or anything like that,

01:10:06.280 | but here's a couple that I really love.

01:10:09.000 | So one is "Brave New World" by Aldous Huxley.

01:10:13.440 | And it's kind of incredible how prescient he was

01:10:20.200 | about what a brave new world might be like.

01:10:25.200 | You kind of see a genetic sorting in this book,

01:10:28.440 | where there's like these alphas and epsilons

01:10:30.760 | and from like the earliest time of society,

01:10:34.960 | like they're sorted.

01:10:35.800 | Like you can kind of see it in a slightly similar way today

01:10:39.080 | where, well, one of the problems with society

01:10:42.120 | is people are kind of genetically sorting a little bit.

01:10:46.000 | Like there's much less, like most marriages

01:10:49.040 | between people of similar kind of intellectual level

01:10:53.400 | or socioeconomic status, more so these days

01:10:55.920 | than in the past.

01:10:57.600 | And you kind of see some effects of it

01:10:59.120 | in stratifying society and kind of,

01:11:01.720 | he illustrated what that could be like in the extreme.

01:11:05.880 | - Different versions of it on social media as well.

01:11:07.960 | It's not just like marriages and so on.

01:11:09.880 | Like it's genetic sorting in terms of

01:11:12.560 | what Dawkins called memes, his ideas.

01:11:15.000 | - Right, right.

01:11:15.840 | - Being put into these bins

01:11:17.360 | or these little echo chambers and so on.

01:11:20.040 | - Yeah, and that's the book that's,

01:11:21.920 | I think a worthwhile read for everyone.

01:11:23.600 | In 1984 is good, of course, as well.

01:11:25.280 | Like if you're talking about,

01:11:26.560 | dystopian novels of the future.

01:11:28.240 | - Yeah, it's a slightly different view of the future, right?

01:11:30.520 | - But I kind of like identify with

01:11:32.200 | Brave New World a bit more.

01:11:33.660 | Speaking of, not a book, but my favorite kind of

01:11:39.920 | dystopian science fiction is a movie called "Brazil,"

01:11:42.600 | which I don't know if you've heard of.

01:11:44.160 | - I've heard of it and I know I need to watch it,

01:11:46.360 | but yeah, 'cause it's in, is it in English or no?

01:11:50.480 | - It's an English movie, yeah.

01:11:52.080 | And it's a sort of like dystopian movie

01:11:55.780 | of authoritarian incompetence, right?

01:11:58.600 | It's like nothing really works very well.

01:12:03.640 | The system is creaky,

01:12:05.720 | but no one is kind of like willing to challenge it.

01:12:08.200 | Just things kind of amble along.

01:12:10.040 | It kind of strikes me as like a very plausible future

01:12:13.680 | of like, you know, what authoritarians might look like.

01:12:16.840 | It's not like this, you know,

01:12:19.240 | super efficient evil dictatorship of 1984.

01:12:21.880 | It's just kind of like this badly functioning, you know,

01:12:25.240 | but it's status quo, so it just goes on.

01:12:30.080 | - Yeah, that's one funny thing that stands out to me

01:12:33.520 | is in what is this, authoritarian dystopian stuff,

01:12:37.140 | or just basic like, you know,

01:12:39.480 | if you look at the movie "Contagion,"

01:12:42.400 | it seems in the movies,

01:12:44.480 | government is almost always exceptionally competent.

01:12:48.140 | Like, it's like used as a storytelling tool

01:12:53.200 | of like extreme competence.

01:12:55.480 | Like, you know, you use it whether it's good or evil,

01:12:58.360 | but it's competent.

01:12:59.680 | It's very interesting to think about

01:13:01.840 | where much more realistically is incompetence,

01:13:06.440 | and that incompetence is itself has consequences

01:13:11.280 | that are difficult to predict.

01:13:13.200 | Like, bureaucracy has a very boring way of being evil.

01:13:18.000 | Of just, you know, if you look at the show,

01:13:21.400 | HBO show "Chernobyl," it's a really good story

01:13:24.560 | of how bureaucracy, you know,

01:13:28.120 | leads to catastrophic events,

01:13:32.760 | but not through any kind of evil

01:13:34.280 | in any one particular place, but more just like the--

01:13:37.680 | - It's just the system, kind of.

01:13:39.120 | - The system, distorting information

01:13:41.240 | as it travels up the chain,

01:13:43.240 | that people unwilling to take responsibility for things,

01:13:46.040 | and just kind of like this laziness resulting in evil.

01:13:50.960 | - There's a comedic version of this,

01:13:52.320 | I don't know if you've seen this movie,

01:13:53.680 | it's called "The Death of Stalin."

01:13:55.040 | - Yeah.

01:13:55.880 | - All right.

01:13:56.840 | - I like that.

01:13:58.160 | I wish it wasn't so,

01:14:00.160 | there's a movie called "Inglorious Bastards"

01:14:02.440 | about, you know, Hitler and, you know, so on.

01:14:06.340 | For some reason, those movies piss me off.

01:14:09.720 | I know a lot of people love them,

01:14:11.200 | but like, I just feel like there's not enough good movies,

01:14:16.200 | even about Hitler.

01:14:18.620 | There's good movies about the Holocaust,

01:14:21.480 | but even Hitler, there's a movie called "Downfall"

01:14:23.720 | that people should watch,

01:14:24.560 | I think it's the last few days of Hitler,

01:14:26.120 | that's a good movie, turned into a meme.

01:14:28.860 | - Mm-hmm, mm-hmm.

01:14:29.840 | - But it's good, but on Stalin,

01:14:31.720 | I feel like I may be wrong on this,

01:14:33.840 | but at least in the English-speaking world,

01:14:35.600 | there's not good movies about the evil of Stalin.

01:14:38.820 | - That's true, I was trying to say that.

01:14:40.720 | I actually, so I agree with you on "Inglorious Bastards",

01:14:43.360 | I didn't love the movie,

01:14:44.560 | because I felt like kind of the stylizing of it, right?

01:14:50.040 | The whole like Tarantino kind of Tarantinoism,

01:14:54.160 | if you will, kind of detracted from it

01:14:56.160 | and made it seem like unserious a little bit.

01:14:58.440 | But "Death of Stalin", I felt differently.

01:15:02.280 | Maybe it's because it's a comedy to begin with,

01:15:03.880 | so it's not like I'm expecting, you know, seriousness,

01:15:06.600 | but it kind of depicted the absurdity

01:15:10.800 | of the whole situation in a way, right?

01:15:13.360 | I mean, it was funny, so maybe it does make light of it,

01:15:15.320 | but it, some degree, it's probably like this, right?

01:15:18.240 | Like a bunch of kind of people that are like, oh shit,

01:15:21.400 | right, like--

01:15:22.480 | - You're right, but like the thing is,

01:15:25.480 | it was so close to like what probably was reality,

01:15:30.480 | it was caricaturing reality,

01:15:35.520 | to where I think an observer might think that this is not,

01:15:39.360 | like they might think it's a comedy,

01:15:41.680 | when in reality, that's the absurdity

01:15:45.600 | of how people act with dictators.

01:15:48.840 | I mean, I guess it was too close to reality for me.

01:15:53.840 | - The kind of banality of like what were eventually

01:15:57.760 | like fairly evil acts, right?

01:15:59.520 | But like, yeah, they're just a bunch of people

01:16:02.320 | trying to survive.

01:16:04.480 | 'Cause I think there's a good,

01:16:05.520 | I haven't watched it yet, the good movie on,

01:16:07.640 | the movie on Churchill with Gary Oldman,

01:16:12.520 | I think it's Gary Oldman, I may be making that up,

01:16:15.480 | but I think he won,

01:16:16.320 | like he was nominated for an Oscar or something.

01:16:18.040 | So I like, I love these movies about these humans

01:16:21.040 | and Stalin, like Chernobyl made me realize,

01:16:24.880 | the HBO show that there's not enough movies about Russia

01:16:28.840 | that capture that spirit.

01:16:33.120 | I'm sure it might be in Russian, there is,

01:16:35.720 | but the fact that some British dude that like did comedy,

01:16:39.400 | I feel like he did like "Hangover" or some shit like that.

01:16:42.240 | I don't know if you're familiar

01:16:43.200 | with the person who created "Chernobyl,"

01:16:44.480 | but he was just like some guy

01:16:45.760 | that doesn't know anything about Russia.

01:16:47.400 | And he just went in and just studied it,

01:16:49.840 | like did a good job of creating it

01:16:51.960 | and then got it so accurate, like poetically.

01:16:56.160 | And the facts that you need to get accurate,

01:16:58.960 | he got accurate, just the spirit of it

01:17:01.200 | down to like the bowls that pets use,

01:17:03.880 | just the whole feel of it.

01:17:05.200 | - It was good, yeah, I saw the series.

01:17:07.560 | - Yeah, it's incredible.

01:17:08.720 | It made me wish that somebody did a good,

01:17:11.000 | like 1930s, like starvation at Stalin,

01:17:16.000 | like leading up to World War II

01:17:20.240 | and in World War II itself, like Stalingrad and so on.

01:17:23.800 | Like, I feel like that story needs to be told.

01:17:27.560 | Millions of people died.

01:17:30.160 | And to me, it's so much more fascinating than Hitler

01:17:32.880 | 'cause Hitler is like a caricature of evil almost

01:17:37.640 | that it's so, especially with the Holocaust,

01:17:41.920 | it's so difficult to imagine

01:17:44.080 | that something like that is possible ever again.

01:17:47.640 | Stalin to me represents something that is possible.

01:17:52.640 | Like the so interesting, like the bureaucracy of it,

01:17:56.800 | it's so fascinating that it potentially might be happening

01:18:01.240 | in the world now, like that we're not aware of,

01:18:03.240 | like with North Korea, another one that,

01:18:06.160 | like there should be a good film on

01:18:08.320 | and like the possible things that could be happening

01:18:10.680 | in China with overreach of government.

01:18:13.160 | I don't know, there's a lot of possibilities there,

01:18:15.600 | I suppose.

01:18:16.480 | - Yeah, I wonder how much,

01:18:18.360 | I guess the archives should be maybe more open nowadays.

01:18:20.480 | I mean, for a long time, they just, we didn't know, right?

01:18:23.880 | Anyways, no one in the West knew for sure.

01:18:25.960 | - Well, there's a, I don't know if you know him,

01:18:27.640 | there's a guy named Stephen Kotkin.

01:18:29.520 | He is a historian of Stalin that I spoke to on this podcast.

01:18:33.160 | I'll speak to him again.

01:18:34.840 | The guy knows his shit on Stalin.

01:18:38.120 | He like read everything.

01:18:41.240 | And it's so fascinating to talk to somebody,

01:18:46.240 | like he knows Stalin better than Stalin knew himself.

01:18:51.760 | It's crazy.

01:18:53.080 | Like you have, so I think he's at Princeton.

01:18:55.520 | He is basically, his whole life is Stalin.

01:18:58.960 | - Studying Stalin.

01:18:59.840 | - Yeah, it's great.

01:19:01.000 | And in that context, he also talks about

01:19:03.720 | and writes about Putin a little bit.

01:19:06.040 | I've also read at this point,

01:19:07.920 | I think every biography of Putin,

01:19:09.800 | English biography of Putin, I need to read some Russians.

01:19:14.040 | Obviously, I'm mentally preparing

01:19:15.440 | for a possible conversation with Putin.

01:19:17.560 | - What is your first question to Putin

01:19:19.400 | when you have him on the podcast?

01:19:24.360 | - It's interesting you bring that up.

01:19:26.400 | First of all, I wouldn't tell you, but.

01:19:28.000 | (laughing)

01:19:28.960 | - Can't give it away now.

01:19:30.720 | - But I actually haven't even thought about that.

01:19:34.400 | So my current approach,

01:19:35.600 | and I do this with interviews often,

01:19:38.520 | but obviously that's a special one,

01:19:40.200 | but I try not to think about questions until last minute.

01:19:44.280 | I'm trying to sort of get into the mindset.

01:19:48.680 | And so that's why I'm soaking in a lot of stuff,

01:19:52.320 | not thinking about questions,

01:19:53.960 | just learning about the man.

01:19:56.120 | But in terms of like human to human,

01:19:59.680 | it's like, I would say it's,

01:20:01.400 | I don't know if you're a fan of mob movies,

01:20:03.440 | but like the mafia, which I am,

01:20:05.840 | like Goodfellas and so on,

01:20:07.000 | he's much closer to like mob morality, which is like.

01:20:12.000 | - Mob morality, maybe, I could see that.

01:20:14.000 | But I like your approach anyways of this,

01:20:16.600 | the extreme empathy, right?

01:20:18.160 | It's a little bit like, you know, Hannibal, right?

01:20:21.360 | Like if you ever watched the show Hannibal, right?

01:20:22.960 | They had that guy,

01:20:23.960 | we know Hannibal, of course, like.

01:20:27.600 | - Yeah, sounds like the lamb.

01:20:30.280 | - But there's TV shows as well,

01:20:31.840 | and they focused on this guy, Will Durant,

01:20:34.080 | who's a character like extreme empath, right?

01:20:36.320 | So in the way he like catches all these killers,

01:20:38.200 | is he pretty much,

01:20:39.280 | he can empathize with them, right?

01:20:42.600 | Like he can understand why they're doing

01:20:44.120 | the things they're doing, right?

01:20:44.960 | And it's a pretty excruciating thing, right?

01:20:48.200 | Like, because you're pretty much like

01:20:49.440 | spending half your time in the head of evil people, right?

01:20:52.440 | - Yeah.

01:20:53.280 | - But.

01:20:54.240 | - I mean, I definitely try to do that with other,

01:20:57.040 | so you should do that in moderation,

01:20:59.120 | but I think it's a pretty safe place to be.

01:21:04.120 | One of the cool things with this podcast,

01:21:06.520 | and I know you didn't sign up to hear me

01:21:08.840 | listen to this bullshit, but.

01:21:10.240 | (laughing)

01:21:11.080 | - No, it's interesting.

01:21:12.240 | - What's his name, Chris Latner, who's a Google,

01:21:17.680 | oh, he's not Google anymore, Sci-Fi.

01:21:19.160 | He's one of the most legit engineers I've talked with.

01:21:21.760 | I talked with him again on this podcast,

01:21:23.400 | and one of the, he gives me private advice a lot,

01:21:26.280 | and he said for this podcast, I should like interview,

01:21:31.240 | like I should widen the range of people,

01:21:34.640 | because that gives you much more freedom to do stuff.

01:21:38.200 | Like, so his idea, which I think I agree with Chris,

01:21:41.560 | is that you go to the extremes.

01:21:44.040 | You just like cover every extreme base,

01:21:46.080 | and then it gives you freedom to then

01:21:47.960 | go to the more nuanced conversations.

01:21:50.440 | And it's kinda, I think there's a safe place for that.

01:21:53.960 | There's certainly a hunger for that nuanced conversation,

01:21:56.680 | I think, amongst people, where like on social media,

01:22:00.440 | you get canceled for anything slightly tense,

01:22:04.080 | that there's a hunger to go full.

01:22:06.040 | - Right, you go so far to the opposite side.

01:22:08.440 | And it's like demystifies it a little bit, right?

01:22:10.760 | - Yeah, yeah.

01:22:11.600 | - There is a person behind all of these things.

01:22:15.120 | - And that's the cool thing about podcasting,

01:22:17.360 | like three, four hour conversations

01:22:19.320 | that it's very different than a clickbait journalism.

01:22:24.120 | It's like the opposite, that there's a hunger for that.

01:22:26.720 | There's a willingness for that.

01:22:28.080 | - Yeah, especially now, I mean,

01:22:29.480 | how many people do you even see face to face anymore?

01:22:31.720 | - Right. - Right, like this, you know?

01:22:33.320 | It's like not that many people, like in my day to day,

01:22:36.080 | aside from my own family, that like I sit across.

01:22:39.240 | - It's sad, but it's also beautiful.

01:22:41.520 | Like I've gotten the chance to,

01:22:43.280 | like our conversation now, there's somebody,

01:22:47.120 | I guarantee you there's somebody in Russia

01:22:50.040 | listening to this now like jogging.

01:22:52.280 | There's somebody who is just smoked some weed,

01:22:55.320 | sit back on a couch and just like enjoying.

01:22:58.480 | I guarantee you that we'll write in the comments right now

01:23:00.760 | that yes, I'm in St. Petersburg, I'm in Moscow, I'm whatever.

01:23:05.040 | And we're in their head and they have a friendship with us.

01:23:10.040 | I'm the same way, I'm a huge fan of podcasting.

01:23:13.320 | It's a beautiful thing.

01:23:15.600 | I mean, it's a weird one way human connection.

01:23:18.200 | Like before I went on Joe Rogan,

01:23:20.360 | and still I'm just a huge fan of his.

01:23:24.160 | So it was like surreal.

01:23:25.800 | I've been a friend with Joe Rogan for 10 years, but one way.

01:23:29.000 | - Yeah, from this way, from the St. Petersburg way.

01:23:31.760 | - Yeah, the St. Petersburg way.

01:23:32.960 | And it's a real friendship.

01:23:35.000 | I mean, now it's like two way, but it's still surreal.

01:23:38.840 | And that's the magic of podcasting.

01:23:40.520 | I'm not sure what to make of it.

01:23:42.080 | That voice, it's not even the video part.

01:23:45.360 | It's the audio that's magical.

01:23:48.560 | I don't know what to do with it,

01:23:50.200 | but it's people listen to three, four hours.

01:23:53.080 | - Yeah, we evolved over millions of years, right?

01:23:57.440 | To be very fine tuned to things like that, right?

01:24:00.480 | Oh, expressions as well, of course, right?

01:24:02.520 | But back in the day on the Savannah,

01:24:06.960 | you had to be very attuned to whether

01:24:09.360 | you had a good relationship with the rest of your tribe

01:24:11.840 | or a very bad relationship, right?

01:24:13.440 | Because if you had a very bad relationship,

01:24:15.160 | you were probably gonna be left behind

01:24:17.400 | and eaten by the lions.

01:24:18.840 | - Yeah, but it's weird that the tribe is different now.

01:24:22.600 | Like you could have a one way connection with Joe Rogan

01:24:26.040 | as opposed to the tribe of your physical vicinity.

01:24:30.560 | - But that's why it works with the podcasting,

01:24:33.360 | but it's the opposite of what happens on Twitter, right?

01:24:35.960 | Because all those nuances are removed, right?

01:24:38.080 | You're not connecting with the person

01:24:40.760 | 'cause you don't hear the voice.

01:24:42.320 | You're connecting with like an abstraction, right?

01:24:44.400 | It's like some stream of tweets, right?

01:24:48.400 | And it's very easy to assign to them

01:24:52.560 | any kind of like evil intent, you know,

01:24:55.040 | or dehumanize them, which is much harder to do

01:24:58.120 | when it's a real voice, right?

01:24:59.160 | Because you realize it's a real person behind the voice.

01:25:02.720 | - Let me try this out on you.

01:25:05.000 | I sometimes ask about the meaning of life.

01:25:07.160 | Do you, your father now, an engineer,

01:25:12.160 | you're building up a company.

01:25:14.040 | Do you ever zoom out and think like,

01:25:16.840 | what the hell is this whole thing for?

01:25:19.360 | Like why are we descended to vapes even on this planet?

01:25:23.960 | What's the meaning of it all?

01:25:26.120 | - That's a pretty big question.

01:25:28.120 | I think I don't allow myself to think about it too often,

01:25:32.240 | or maybe like life doesn't allow me

01:25:34.320 | to think about it too often.

01:25:35.800 | But in some ways, I guess, the meaning of life

01:25:39.080 | is kind of contributing to this kind of weird thing

01:25:44.040 | we call humanity, right?

01:25:45.320 | Like it's in a way, you can think of humanity

01:25:47.640 | as like a living and evolving organism, right?

01:25:50.240 | That like we all contribute in a slight way,

01:25:52.520 | but just by existing, by having our own unique set

01:25:55.640 | of desires and drives, right?

01:25:57.320 | And maybe that means like creating something great,

01:26:01.640 | and it's bringing up kids who, you know,

01:26:04.640 | are unique and different and seeing like, you know,

01:26:07.800 | they can join what they do.

01:26:09.720 | But I mean, to me, that's pretty much it.

01:26:11.040 | I mean, if you're not a religious person, right,

01:26:13.200 | which I guess I'm not, that's the meaning of life.

01:26:16.440 | It's in the living and in the creation.

01:26:20.920 | - Yeah, there's something magical

01:26:22.440 | about that engine of creation.

01:26:24.200 | Like you said, programming, I would say,

01:26:27.280 | I mean, it's even just actually what you said

01:26:29.420 | with even just programs, I don't care

01:26:31.000 | if it's like some JavaScript thing,

01:26:32.800 | a button on the website, it's like magical

01:26:37.240 | that you brought that to life.

01:26:39.040 | I don't know what that is in there, but that seems,

01:26:41.580 | that's probably some version of like reproduction

01:26:46.580 | and sex, whatever that's in evolution.

01:26:49.800 | But like creating that HTML button has echoes

01:26:54.800 | of that feeling and it's magical.

01:26:57.800 | - Right, well, I mean, if you're a religious person,

01:27:00.760 | maybe you could even say, right,

01:27:01.800 | like we were created in God's image, right?

01:27:04.440 | Well, I mean, I guess part of that is the drive

01:27:07.240 | to create something ourselves, right?

01:27:09.200 | I mean, that's part of it.

01:27:11.760 | - Yeah, that HTML button is the creation in God's image.

01:27:14.840 | - So maybe hopefully it'll be something a little more--

01:27:18.880 | - So dynamic, maybe some JavaScript.

01:27:20.960 | - Yeah, maybe some JavaScript, some React and so on.

01:27:25.400 | But no, I mean, I think that's what differentiates us

01:27:29.400 | from the apes, so to speak.

01:27:32.140 | - Yeah, we did a pretty good job.

01:27:34.220 | Dan, it was an honor to talk to you.

01:27:36.960 | Thank you so much for being part of creating

01:27:38.760 | one of my favorite services and products.

01:27:42.000 | This is actually a little bit of an experiment,

01:27:45.080 | allowing me to sort of fanboy over some of the things

01:27:48.520 | I love, so thanks for wasting your time with me today.

01:27:52.280 | It was really fun. - It was awesome.

01:27:53.120 | Thanks for having me on and giving me a chance

01:27:55.520 | to try this out. (laughs)

01:27:57.200 | - Awesome.

01:27:58.020 | Thanks for listening to this conversation

01:28:00.720 | with Dan Kokorov and thank you to our sponsors,

01:28:03.560 | Athletic Greens, All-in-One Nutrition Drink,

01:28:06.200 | Blinkist app that summarizes books,

01:28:09.080 | Business Wars podcast, and Cash App.

01:28:13.020 | So the choice is health, wisdom, or money.

01:28:16.800 | Choose wisely, my friends, and if you wish,

01:28:19.340 | click the sponsor links below to get a discount

01:28:23.000 | and to support this podcast.

01:28:25.280 | And now let me leave you with some words

01:28:27.320 | from Ludwig Wittgenstein.

01:28:29.800 | "The limits of my language means the limits of my world."

01:28:33.840 | Thank you for listening and hope to see you next time.

01:28:37.700 | (upbeat music)

01:28:40.280 | (upbeat music)

01:28:42.860 | [BLANK_AUDIO]

Dan Kokotov: Speech Recognition with AI and Humans | Lex Fridman Podcast #151

Chapters