1
0:00:02,345 --> 0:00:06,032
Mark Haney: Super excited today to be bringing on Ken Miller.

2
0:00:06,132 --> 0:00:19,308
Mark Haney: He is the co-founder and CTO of Fathom and there's other products associated with Fathom, but you guys are coming in and changing the game for podcasters and even people that listen to podcasts.

3
0:00:19,348 --> 0:00:24,523
Mark Haney: So, and for our audience, you're gonna meet a guy that's brilliant, right.

4
0:00:24,543 --> 0:00:29,518
Mark Haney: The first time I met Ken, I was like this guy is freaking brilliant.

5
0:00:29,899 --> 0:00:30,923
Mark Haney: So welcome to the show, ken.

6
0:00:30,944 --> 0:00:31,928
Mark Haney: I appreciate you coming on.

7
0:00:32,049 --> 0:00:33,114
Mark Haney: Thanks, mark, happy to be here.

8
0:00:33,335 --> 0:00:39,131
Mark Haney: Okay, maybe just give us a little bit of background on the company and then we can kind of dive deeper from there.

9
0:00:39,592 --> 0:00:39,913
Ken Miller: For sure.

10
0:00:40,100 --> 0:01:06,602
Ken Miller: Yeah, so you know, in early 2020 I was, you know, working in research at another startup in Trinio, primarily working in the financial space, financial data space and playing around a lot with AI, and I was listening to a lot of podcasts at the time and one thing I just generally noticed in the podcast space was that, like, transcript search really wasn't a thing and there's a lot of reasons for that cost-wise.

11
0:01:08,007 --> 0:01:29,477
Ken Miller: But I was really interested in, you know, like, what would happen if we just set the cost of transcribing all the podcast episodes aside and just said, hey, let's try to transcribe a bunch of stuff and create an AI powered search engine across these, these podcast transcripts, but then also incorporates some extractive AI question answering.

12
0:01:29,649 --> 0:01:48,189
Ken Miller: And you know, after a year of kind of like turning on that idea and doing research and development on the side, I had something, and so then I reached out to my co-founder, paul Block, who's a brilliant designer, brilliant artist, and I said, paul, I've got an AI powered search engine for podcast man, what's up with it?

13
0:01:48,209 --> 0:01:50,742
Ken Miller: And he's like let's go, you know.

14
0:01:51,365 --> 0:01:52,551
Ken Miller: And he it was interesting.

15
0:01:52,591 --> 0:01:54,279
Ken Miller: Paul and I have known each other since high school.

16
0:01:54,300 --> 0:02:11,154
Ken Miller: We've worked at like three different companies together worked at M train, a company here in Sacramento, but this is the first time that we've really kind of struck out and done, done the whole inch, a lot of that's so it's been fun so why do people want transcripts from podcasts?

17
0:02:11,194 --> 0:02:13,084
Mark Haney: who wants these and what kind of?

18
0:02:14,106 --> 0:02:15,349
Mark Haney: What kind of value do they bring?

19
0:02:15,429 --> 0:02:16,652
Mark Haney: Why do what does it help?

20
0:02:16,952 --> 0:02:17,213
Ken Miller: right.

21
0:02:17,273 --> 0:02:24,878
Ken Miller: Well, you know, when we look at audio right as a medium, it's fundamentally opaque, right?

22
0:02:24,918 --> 0:02:40,240
Ken Miller: So, like, think about when you're watching a movie, right, you can fast forward and as you're fast forwarding, you're getting shots along the way so you can tell what's within the content simply by fast forwarding.

23
0:02:40,260 --> 0:02:54,480
Ken Miller: But with audio you don't have that okay it's an opaque medium right so what having the transcript allows us to do is essentially lift the veil right and see what's what's within this particular content.

24
0:02:54,620 --> 0:02:56,588
Ken Miller: Now this has a lot of different cases.

25
0:02:57,431 --> 0:03:03,416
Ken Miller: From a podcasters perspective, you know, they're going to be interested in maybe posting those transcripts for seo purposes and whatnot.

26
0:03:03,918 --> 0:03:16,017
Ken Miller: From a listener's perspective, you know, really having having those transcripts within your podcast listening application really opens up the realm of search and in that's super important.

27
0:03:16,057 --> 0:03:29,404
Ken Miller: You know, when you're like looking for ideas, if you just have the audio, you know there's no, no way to know what's what's actually within there beyond, say like a show notes summary or show notes description okay, so let's talk about the, the different products that you guys have.

28
0:03:29,444 --> 0:03:31,188
Mark Haney: So fathom is one.

29
0:03:31,248 --> 0:03:34,521
Mark Haney: That's what we use if we are a listener is that correct?

30
0:03:34,541 --> 0:03:41,868
Mark Haney: Yeah, and so I can go on to, let's say, the Joe Rogan show, and it breaks it up into chapters and that kind of thing tell me about that.

31
0:03:41,888 --> 0:03:51,493
Ken Miller: You know, uh Daniel, at covert spotify has uh pretty much taken that show off the off the uh oh, okay, yeah out of the the podcasting space.

32
0:03:51,760 --> 0:03:55,140
Mark Haney: Hooberman, I'm trying to think like yeah, I mean all of these.

33
0:03:55,240 --> 0:03:57,857
Ken Miller: You know any general podcast that you listen to.

34
0:03:57,954 --> 0:03:59,796
Ken Miller: Yeah, you can use the fathom player.

35
0:03:59,892 --> 0:04:01,164
Ken Miller: It's available on ios.

36
0:04:01,260 --> 0:04:02,745
Ken Miller: It's got a lot of different features.

37
0:04:03,527 --> 0:04:33,913
Ken Miller: Um, we extract highlights from every podcast that we, we transcribe so and then we extract actually multiple highlights and so when you follow podcasts, which you're gonna get is you're gonna get a feed of highlights from those episodes, but the highlights are further selected and customized for you okay, so as you're going through the feed of new episodes from stuff that you follow, you're gonna get the little tidbit that's most likely to be interesting, uh, to you.

38
0:04:34,521 --> 0:04:37,812
Ken Miller: To help you in the selection process, how do I want to spend my time?

39
0:04:38,836 --> 0:04:47,385
Ken Miller: Um, beyond that, we have question answering search, so you can go into the lex freeman podcast and go to the huberman lab podcast and ask a question.

40
0:04:48,066 --> 0:04:51,994
Ken Miller: You can say you know, uh, how can I optimize my mental performance?

41
0:04:52,500 --> 0:04:59,163
Ken Miller: And what you're gonna get are you're gonna get a series of clips that actually answer that question, which is really cool.

42
0:04:59,704 --> 0:05:03,361
Ken Miller: Um, we have full transcripts available that you can follow along with to your question.

43
0:05:03,381 --> 0:05:20,865
Ken Miller: Before you know what are tramps transcripts good for um, a lot of people like to actually watch the transcript as they're listening, because that's going to increase retention and then, of course, from an accessibility standpoint, right, that's huge, yeah, for the deaf community, um, so we have full transcripts you can follow along with those.

44
0:05:20,945 --> 0:05:21,627
Ken Miller: It's really great.

45
0:05:22,328 --> 0:05:45,630
Ken Miller: And, um, we also have clipping, so, as you're listening, you hear something interesting, right you know boom, click and save that, and what the ai is going to do is it's actually going to look at where you clipped and then it's going to go backwards and actually forwards in time and try to create a clip that represents like a coherent, cohesive thought, and then it's going to save that to your library.

46
0:05:45,670 --> 0:05:52,668
Mark Haney: Yeah, oh wow, okay, that's cool yeah um, and so from a uh, that's a direct to consumer model, right, okay.

47
0:05:53,109 --> 0:06:00,933
Mark Haney: So I know at one point you pivoted to a uh more of a b to b uh model, which we're using.

48
0:06:01,094 --> 0:06:02,798
Mark Haney: Podium my man, scott, over here.

49
0:06:02,818 --> 0:06:06,511
Mark Haney: I was asking him before the show um, what do you like about it?

50
0:06:06,551 --> 0:06:11,551
Mark Haney: He's like it saves me a ton of time, right, maybe describe that product, because there's a lot of people that have podcasts now.

51
0:06:11,617 --> 0:06:17,981
Ken Miller: So it's not like it's a tiny market, there's a lot of podcasters yeah, I mean just to speak to kind of like the total addressable market.

52
0:06:18,021 --> 0:06:21,528
Ken Miller: There's roughly like, uh, two and a half million podcasts out there.

53
0:06:22,029 --> 0:06:26,185
Ken Miller: Um 400 000 of them is are what they call like active, like they're.

54
0:06:26,225 --> 0:06:33,900
Ken Miller: They're very active, they're publishing monthly, um, so there's yeah, there's a tremendous amount of content, tremendous number of podcasters out there there.

55
0:06:34,943 --> 0:06:58,373
Ken Miller: Basically, what happened is, you know, we, we raised some money, we spent um a couple years building the fathom player, um, but what we came to realize, product wise, is that just the road to monetization for a podcast player it's really long, it's great from a strategic standpoint, but you're really looking at like six, eight years to gain any kind of significant market share.

56
0:06:59,321 --> 0:07:04,065
Ken Miller: And with, you know, the macro economics being the way that they are, we decided, hey, you know, we need to bring in some revenue.

57
0:07:04,828 --> 0:07:10,125
Ken Miller: So we kind of took a look at everything that we had and we had discussed, like things that we could do on the creator side.

58
0:07:10,185 --> 0:07:14,923
Ken Miller: Anyhow, we had built a ton of ai technology to to power this fathom player.

59
0:07:15,564 --> 0:07:23,127
Ken Miller: So we just we basically repurposed all of that ai and created podium, podium, launched this last february.

60
0:07:23,408 --> 0:07:25,154
Ken Miller: And essentially, what podium is it's?

61
0:07:25,294 --> 0:07:35,549
Ken Miller: It's ai tools for podcasters, um, the primary functionality that people use it for is, yes, to your point, to generate show notes to uh chapterizer episodes.

62
0:07:35,609 --> 0:07:46,831
Ken Miller: So it's going to break it into segments, give you all of those timestamps, give you summaries for all of the chapters, extract your keywords, give you title suggestions and, um, it also extracts highlights.

63
0:07:46,891 --> 0:07:48,558
Ken Miller: So the fathom player extracts highlights.

64
0:07:48,779 --> 0:07:49,901
Ken Miller: We also extract highlights.

65
0:07:50,642 --> 0:07:51,744
Ken Miller: Uh, a podium for you.

66
0:07:51,784 --> 0:07:54,693
Ken Miller: So super useful for podcasters.

67
0:07:54,765 --> 0:08:00,189
Ken Miller: The reception has been absolutely amazing and, yes, people love it because it saves them a tremendous amount of time.

68
0:08:00,985 --> 0:08:01,367
Mark Haney: How much.

69
0:08:01,948 --> 0:08:08,811
Mark Haney: And I was talking to Scott earlier and he says that not only does it save him time, it's as good as what he could write.

70
0:08:10,306 --> 0:08:15,514
Mark Haney: He tweaks so awarding here or there on the show notes, but pretty much great writing.

71
0:08:15,565 --> 0:08:19,094
Mark Haney: I mean it's written so that it's like ready to post.

72
0:08:19,455 --> 0:08:19,776
Ken Miller: Right.

73
0:08:20,165 --> 0:08:24,255
Ken Miller: So another feature within podium is what we call podium GPT.

74
0:08:25,425 --> 0:08:39,689
Ken Miller: So beyond all of the standard stuff that we generate by default show notes, chapters, keyword extraction, all that jazz we also have kind of almost a free form chat, right, that fully understands your episode and you can ask it.

75
0:08:39,970 --> 0:08:51,622
Ken Miller: Write me a Twitter thread about this episode, write me a blog post, write me an email out to my newsletter audience and it fully understands the episode.

76
0:08:52,587 --> 0:08:54,213
Ken Miller: So it'll then write that for you.

77
0:08:55,347 --> 0:09:03,725
Ken Miller: The whole thing with Fathom as a company is that and Paul and I it's just kind of a vibe we're on.

78
0:09:04,287 --> 0:09:05,352
Ken Miller: We like to push the limits.

79
0:09:06,067 --> 0:09:10,364
Ken Miller: So I wanna see how good can we make this content?

80
0:09:10,385 --> 0:09:10,485
Ken Miller: Right?

81
0:09:12,007 --> 0:09:13,874
Ken Miller: How hard can we push this AI?

82
0:09:14,751 --> 0:09:17,965
Ken Miller: And don't worry about the cost It'll get cheaper over time, right?

83
0:09:19,229 --> 0:09:24,965
Ken Miller: Let's see how accurate, how nuanced we can get this content.

84
0:09:25,807 --> 0:09:32,295
Ken Miller: So we spend a tremendous amount of time really doing a lot of testing, a lot of like.

85
0:09:33,209 --> 0:09:36,602
Ken Miller: Sometimes I remember we'll take one episode and just generating show notes.

86
0:09:37,208 --> 0:09:51,017
Ken Miller: I might generate 100 versions of those show notes with different prompts, different tweaks and whatnot, and then read them all and try to arrive at, you know, a prompt for the underlying AI.

87
0:09:51,165 --> 0:09:55,777
Ken Miller: That really makes something that sounds nearly human.

88
0:09:56,145 --> 0:10:08,713
Mark Haney: Okay, so you actually, in your testing, you read the show you're the guy that reads the show notes, I guess and then you say, well, how could we make this a little more accurate?

89
0:10:08,753 --> 0:10:11,466
Mark Haney: I guess a little more engaging, right?

90
0:10:11,647 --> 0:10:11,987
Mark Haney: So?

91
0:10:13,150 --> 0:10:15,717
Ken Miller: unlike some other products out there.

92
0:10:15,765 --> 0:10:21,293
Ken Miller: I mean, as a team we have fundamental AI skills so we can build and train neural networks from scratch.

93
0:10:21,785 --> 0:10:32,376
Ken Miller: So we know how all of the open AI GPT style models work and what you're really doing when you're writing a prompt.

94
0:10:32,505 --> 0:10:37,673
Ken Miller: Like many of the people in this audience will be using chat GPT, right?

95
0:10:37,785 --> 0:10:55,115
Ken Miller: So when you're writing a prompt, you probably noticed that you know if you change that prompt slightly you can get significantly different output, and it's because when that prompt is going in as input, it's literally almost like an MRI or an FMRI.

96
0:10:55,685 --> 0:11:02,071
Ken Miller: It's gonna the words that you're putting in are gonna light up different portions of that neural network, right?

97
0:11:02,865 --> 0:11:37,154
Ken Miller: So part of the trick, when you're trying to take, say, an existing piece of content and then adapt it into another form, right, which you can almost think of that show notes summary as a form of content adaptation, right From the conversational transcript into adapted into a summary you wanna make sure that, as that adaptation process is occurring, that you're lighting up the right portions of the neural network, right, tonally, stylistically, right.

98
0:11:37,345 --> 0:11:41,668
Ken Miller: So, but in order to do that, it really does take a lot of work and a lot of testing.

99
0:11:42,345 --> 0:11:45,950
Mark Haney: Okay, so you must be a speed reader, because I mean, if you're having, you must.

100
0:11:46,105 --> 0:11:48,173
Mark Haney: I mean, this is like heavy duty.

101
0:11:48,265 --> 0:11:49,629
Mark Haney: You're reading this stuff yourself.

102
0:11:50,091 --> 0:11:53,411
Mark Haney: Yeah, right, I mean, and I have to comprehend it too.

103
0:11:53,671 --> 0:11:54,995
Mark Haney: Yeah, well, I mean, and put creativity to it.

104
0:11:55,526 --> 0:12:00,274
Ken Miller: My AI background is in computational linguistics, so you know, I mean, yeah, I'd love to read.

105
0:12:01,385 --> 0:12:02,751
Mark Haney: Okay, very interesting.

106
0:12:02,805 --> 0:12:05,374
Mark Haney: Okay, talk to us about the product roadmap.

107
0:12:07,971 --> 0:12:08,653
Mark Haney: What's next?

108
0:12:09,094 --> 0:12:09,976
Mark Haney: What are you most excited about?

109
0:12:10,925 --> 0:12:14,413
Ken Miller: Well, what I'm most excited about is this new product that we just released.

110
0:12:14,545 --> 0:12:15,851
Ken Miller: It's absolutely fantastic.

111
0:12:15,965 --> 0:12:16,809
Ken Miller: It's called Podbook.

112
0:12:18,305 --> 0:12:25,854
Ken Miller: So whereas Podium today works with a singular episode right, so you upload in your episode, you just recorded it.

113
0:12:26,225 --> 0:12:28,654
Ken Miller: Everybody loves talking into these microphones, right?

114
0:12:28,785 --> 0:12:30,370
Ken Miller: Super fun, super easy.

115
0:12:31,685 --> 0:12:35,052
Ken Miller: Nobody likes writing, so you know, podium's great for that.

116
0:12:36,565 --> 0:12:39,746
Ken Miller: But we started to really think, paul and I, around.

117
0:12:40,228 --> 0:12:41,914
Ken Miller: What can we do with your whole library?

118
0:12:42,045 --> 0:12:43,531
Ken Miller: I mean, you'll resonate with this.

119
0:12:43,665 --> 0:12:50,533
Ken Miller: You know you spend years of time investing into this content, right, recording all these episodes, editing them.

120
0:12:52,525 --> 0:12:57,291
Ken Miller: You know, what can we do, not, with an hour, an hour and a half of audio?

121
0:12:57,905 --> 0:13:00,353
Ken Miller: What can we do with 200 hours of audio?

122
0:13:01,226 --> 0:13:04,015
Ken Miller: And we thought we can make a book, you know.

123
0:13:04,165 --> 0:13:16,416
Ken Miller: And so what would that look like if we took 200 hours of audio and then adapted that content into, say, a 60,000 to 90,000 word readable book?

124
0:13:18,145 --> 0:13:25,010
Ken Miller: Incidentally, not super easy, but we managed to pull it off and we've put it out there.

125
0:13:25,085 --> 0:13:29,946
Ken Miller: The reception has been absolutely mind boggling, you know.

126
0:13:30,247 --> 0:13:33,813
Ken Miller: The interest in it has far exceeded what we thought, you know.

127
0:13:33,985 --> 0:13:36,534
Mark Haney: It isn't the same clients the podcaster who wants to write a book.

128
0:13:36,705 --> 0:13:40,230
Mark Haney: I'm a podcaster, I've been doing this thing for five years and, man, I'm gonna make a book.

129
0:13:40,445 --> 0:13:44,796
Ken Miller: Right now, we're super focused on the podcast space.

130
0:13:46,045 --> 0:13:47,871
Ken Miller: As a company, we do have an API.

131
0:13:48,985 --> 0:13:51,112
Ken Miller: Interestingly, there's another company.

132
0:13:51,825 --> 0:13:57,551
Ken Miller: We actually have startups that are building off of our startup tech, so one of them is called Pulpit AI.

133
0:13:58,585 --> 0:14:08,135
Ken Miller: They actually recently just made Fox News, but they're geared at pastors right, so all pastors record their sermons now.

134
0:14:08,355 --> 0:14:16,651
Ken Miller: Right, so you have four to six hours of audio content per church per month.

135
0:14:17,765 --> 0:14:23,796
Ken Miller: They're actually connected with the Tidely guys, so, and Tidely is using about 37,000 different churches now.

136
0:14:24,465 --> 0:14:32,577
Ken Miller: So Pulpit AI is essentially an offshoot of Tidely, and they're building everything off of the podium platform.

137
0:14:32,665 --> 0:14:49,231
Ken Miller: So, while we're directly targeting podcasters, there's a lot of opportunity in other audio spaces, right, both for kind of like singular one shot, like summarization, and chapterization for, like, let's say, a singular sermon.

138
0:14:49,565 --> 0:14:55,811
Ken Miller: But what if you take 10 years worth of sermons and then produce a book out of it?

139
0:14:56,112 --> 0:14:58,470
Ken Miller: Right, for your congregation?

140
0:14:59,145 --> 0:15:00,591
Ken Miller: So that becomes really interesting.

141
0:15:01,568 --> 0:15:08,749
Ken Miller: So, while as a company, right now we're kind of laser focused on the podcast space, we definitely see a tremendous amount of opportunity in other audio spaces.

142
0:15:09,025 --> 0:15:11,694
Mark Haney: Yeah, I'm thinking that pod book thing sounds really cool.

143
0:15:12,065 --> 0:15:21,190
Mark Haney: And another idea came to me just as we were talking here that might be interesting for a future roadmap and I don't know if this is possible.

144
0:15:21,285 --> 0:15:26,990
Mark Haney: But Scott and I have talked about going back and there's doing like a best of right For like micro content.

145
0:15:27,325 --> 0:15:34,291
Mark Haney: You know, we've been doing this content stuff for a long time and I've a couple of times I've said some really cool stuff and I wanna go back and find it.

146
0:15:34,605 --> 0:15:35,007
Mark Haney: What about?

147
0:15:35,268 --> 0:15:41,492
Mark Haney: You know, I wanna go back and find something I said that was inspirational or maybe somebody else said was inspirational.

148
0:15:41,845 --> 0:15:45,112
Mark Haney: You know, take all the episodes and find the most inspirational stuff.

149
0:15:45,205 --> 0:15:52,574
Mark Haney: Or every time I talked about winning or something like that, or competitive advantage or the backyard advantage, give me the best of that.

150
0:15:52,705 --> 0:15:57,850
Mark Haney: And then I would wanna use that for social media, micro you know, micro content.

151
0:15:58,513 --> 0:15:58,773
Ken Miller: Right.

152
0:15:59,927 --> 0:16:06,549
Ken Miller: So what's funny about this, mark, is you can actually go use the Fathom Player for that functionality.

153
0:16:06,589 --> 0:16:11,768
Ken Miller: Now, like I can go into you know the Marcanie podcast and I can.

154
0:16:11,889 --> 0:16:15,233
Ken Miller: You can literally ask it, like, what is the backyard advantage?

155
0:16:15,785 --> 0:16:22,491
Ken Miller: And it's gonna pull up clips across your entire library, right, but that's from a listener perspective.

156
0:16:22,511 --> 0:16:23,314
Ken Miller: Wow.

157
0:16:23,485 --> 0:16:27,511
Ken Miller: And what we found out is that, search-wise maybe we were a little bit ahead of our time.

158
0:16:28,045 --> 0:16:42,570
Ken Miller: Sometimes you engineer and you invent something and the mark is just not ready for it, right, and when it comes to podcast search, it's something that people do, but we find that they maybe do it once every two weeks, once a month, not enough.

159
0:16:43,345 --> 0:16:49,033
Ken Miller: But for the podcaster themselves, the content creator, that kind of search is super powerful.

160
0:16:49,705 --> 0:17:05,577
Ken Miller: So we are currently working on basically incorporating all of that functionality into Podium as a product so that you can search against your entire content library using AI and also just generate clips.

161
0:17:06,045 --> 0:17:08,053
Ken Miller: We have a new transcript editor that's coming out.

162
0:17:09,586 --> 0:17:13,872
Ken Miller: That's actually going to be almost kind of like a simplified version of Descript.

163
0:17:15,745 --> 0:17:24,913
Ken Miller: We will likely have actual like audio editing in the future where, much like Descript, you'll be able to edit the audio by editing the transcript.

164
0:17:25,195 --> 0:17:36,830
Ken Miller: Oh interesting, Right so, but the ability to search across your entire content library and then repurpose clips and stuff like that into videos and audiograms, that's all coming to Podium.

165
0:17:36,905 --> 0:17:37,930
Ken Miller: We're working on it currently.

166
0:17:38,285 --> 0:17:54,814
Mark Haney: So in terms of video, so like we have micro content now that we it's hard to pull the audio off of that to you know, is there anything that allows us to put like a video clip into you know, small video clip into the?

167
0:17:54,834 --> 0:17:55,766
Ken Miller: system Right.

168
0:17:55,907 --> 0:18:05,815
Ken Miller: So when we first launched Podium it's kind of funny it was really a market test that ended up blowing up, you know, and I feel like we've been playing catch up ever since.

169
0:18:07,185 --> 0:18:12,093
Ken Miller: But we really focused a lot of our engineering effort just on the AI and it shows.

170
0:18:14,269 --> 0:18:21,046
Ken Miller: But now we're actually working a lot more on the UI and, to your point, yeah, audiograms is what they're typically called.

171
0:18:21,347 --> 0:18:28,191
Ken Miller: I think, scott, you know, uses some other services, like Headliner right, in order to generate those audiograms.

172
0:18:28,471 --> 0:18:46,433
Ken Miller: Currently, what we do provide you are like start and stop points for highlights, but what we're gonna be doing is taking those start and stop points that are found by the AI for interesting highlights and then, within the Podium UI, you're actually gonna be able to generate videos directly from there.

173
0:18:46,814 --> 0:18:47,395
Mark Haney: Oh, okay.

174
0:18:47,645 --> 0:18:52,609
Ken Miller: Right, so you know, I'd say that's you know, less than 90 days out.

175
0:18:53,165 --> 0:18:54,650
Mark Haney: Wow, you build the stuff fast.

176
0:18:55,874 --> 0:18:59,715
Mark Haney: Yeah, you're using AI to help drive this too right.

177
0:19:00,125 --> 0:19:02,033
Mark Haney: Some of the changes and the fixes you make.

178
0:19:02,426 --> 0:19:04,132
Mark Haney: I mean, what goes into that?

179
0:19:04,465 --> 0:19:06,011
Ken Miller: Right, well, absolutely now.

180
0:19:06,112 --> 0:19:21,708
Ken Miller: You know I almost use GPT-4 as, like you know, my junior coder, so I'll be oftentimes, you know, writing code and as I'm writing code, gpt-4 is writing some other piece of code I know I'm gonna need in five minutes.

181
0:19:22,425 --> 0:19:27,610
Ken Miller: So it's really, yeah, changed my workflow, you know, fairly substantially.

182
0:19:28,545 --> 0:19:29,570
Ken Miller: You know it's not perfect.

183
0:19:30,207 --> 0:19:31,773
Ken Miller: It still has a way to go, in my opinion.

184
0:19:31,905 --> 0:19:34,813
Ken Miller: I don't think it's gonna be replacing engineers GPT-4?

185
0:19:34,833 --> 0:19:35,215
Mark Haney: Yeah.

186
0:19:35,765 --> 0:19:41,272
Ken Miller: Engineers any time soon, but it definitely as a professional who understands what they're doing.

187
0:19:41,825 --> 0:19:44,915
Ken Miller: You know it's a and this has been said before.

188
0:19:44,985 --> 0:19:46,268
Ken Miller: You know it's not really gonna be.

189
0:19:46,769 --> 0:19:48,352
Ken Miller: Ai is replacing your job.

190
0:19:48,505 --> 0:19:53,966
Ken Miller: It's going to be somebody who's using AI to sort of 10x their productivity.

191
0:19:54,535 --> 0:20:01,842
Ken Miller: That's who's gonna be replacing your job if you don't kind of get on board and learn how to utilize these tools to accelerate your output.

192
0:20:02,075 --> 0:20:06,924
Mark Haney: Yeah, I'm gonna circle back around to that and maybe some general AI questions.

193
0:20:06,955 --> 0:20:10,706
Mark Haney: But on your company, what's the longer term vision?

194
0:20:11,196 --> 0:20:13,564
Mark Haney: Where do you see this thing in, say, three to five years?

195
0:20:14,537 --> 0:20:14,637
Ken Miller: Right.

196
0:20:14,738 --> 0:20:18,140
Ken Miller: I mean, I think it's our positioning is really interesting, right?

197
0:20:18,275 --> 0:20:26,722
Ken Miller: So if we look at the value chain for podcasting right, it starts right here, starts with this mic that's gonna be the first major purchase.

198
0:20:26,955 --> 0:20:32,582
Ken Miller: Anybody who becomes serious in this space, it's gonna be their first major purchase.

199
0:20:32,615 --> 0:20:50,800
Ken Miller: So you go from the mic and then you go into your recording software right, Whatever software you're using to record it From there, it's gonna go into, potentially, an editor like Descript, or if you're kind of cool with the way that it is, you won't put it into an editor like Descript, you're just gonna take the audio.

200
0:20:50,895 --> 0:20:58,983
Ken Miller: Now, what's interesting about Podium is it's actually a wedged node in the value chain that's just recently emerged right.

201
0:20:59,695 --> 0:21:05,823
Ken Miller: Which is before once it's out of Descript or if you're not gonna edit it, you take it right into your host, right?

202
0:21:05,895 --> 0:21:07,621
Ken Miller: You're gonna upload it to your RSS host.

203
0:21:08,435 --> 0:21:10,943
Ken Miller: Now, you upload it to Podium first, right?

204
0:21:11,715 --> 0:21:20,620
Ken Miller: So it's like this node that, because of the advent of large language models, has suddenly appeared within the value chain.

205
0:21:21,435 --> 0:21:27,706
Ken Miller: Now, as a company, we're in a really good position because we have a very, very strong foothold in that node.

206
0:21:27,975 --> 0:21:43,222
Ken Miller: We're really kind of the number one name now, especially after our whole bus sprout integration right Over on the player side, which is a much more difficult market to penetrate, because there we're competing with Apple, we're competing with Spotify, competing with everybody.

207
0:21:43,795 --> 0:21:46,164
Ken Miller: But we have a decent foothold over there too, right?

208
0:21:46,295 --> 0:21:58,381
Ken Miller: So, as a company, for the longer term vision, the question becomes okay, well, do we leverage, kind of the two footholds that we have one at the very end of the value chain?

209
0:21:59,043 --> 0:22:06,258
Ken Miller: Right, cause it goes mic recording, editing, hosting, player, listener right Into my ear.

210
0:22:07,115 --> 0:22:16,443
Ken Miller: So we're at the end of the value chain there software-wise, and now we're in this wedged, so we can compress and squeeze right, you could take other pieces out of that chain.

211
0:22:16,635 --> 0:22:16,916
Ken Miller: Right.

212
0:22:16,956 --> 0:22:26,803
Ken Miller: So the natural progression would be well, you're giving us your audio before you're even giving your RSS host your audio, so we become an RSS host.

213
0:22:26,915 --> 0:22:27,096
Mark Haney: Wow.

214
0:22:27,116 --> 0:22:28,822
Mark Haney: So the opportunity just gets enormous.

215
0:22:29,235 --> 0:22:31,764
Ken Miller: Well, I theorize to a certain extent.

216
0:22:31,855 --> 0:22:47,144
Ken Miller: That's why BuzzFront made a really smart move in integrating our technology to kind of like strategically navigate the fact that there's now this new node and potentially we could also go upstream, right.

217
0:22:48,095 --> 0:23:23,505
Ken Miller: But I think that the real possibility is, in sort of like, squeezing and compressing the value chain, and what that may end up looking like is a merger with another company and or the merger of several companies together, cause you could take a Zencaster or a StreamYard, right Like recording software, merge it with Podium, merge it with like a Buzzsprout, and then we have the Fathom player and now we have a completely vertically integrated solution, end to end, that begins to compete with the likes of Spotify.

218
0:23:23,915 --> 0:23:28,661
Ken Miller: If you look at Spotify's acquisitions over the last three years, it's exactly the move that they're making.

219
0:23:29,015 --> 0:23:32,025
Mark Haney: So it's right now just like a race to get market share with it.

220
0:23:32,556 --> 0:23:35,445
Mark Haney: Just get your thing out there, get Podium out there.

221
0:23:36,057 --> 0:23:36,901
Ken Miller: Get it integrated.

222
0:23:37,155 --> 0:23:50,640
Ken Miller: Get it as many podcasters as possible using Podium to accelerate the time from when they hit the stop button to end the recording to actually them being able to publish their episode.

223
0:23:50,895 --> 0:23:52,121
Mark Haney: You must have no churn either.

224
0:23:52,215 --> 0:23:54,183
Mark Haney: I mean, once we come in, why would you stop using it?

225
0:23:54,375 --> 0:23:57,704
Mark Haney: It's like I'm in, unless I quit my podcast.

226
0:23:57,804 --> 0:24:01,024
Ken Miller: I mean yeah yeah, no, the churn's been really good.

227
0:24:01,095 --> 0:24:17,100
Ken Miller: I mean, you're gonna see, there's actually quite a bit of churn in the podcast space, but what's interesting about it is that every year, the number of podcasts that completely disappear and drop off because somebody thinks, oh, I'm gonna start a podcast.

228
0:24:17,240 --> 0:24:19,847
Mark Haney: Yeah, I think you know a lot of people have had it.

229
0:24:19,868 --> 0:24:20,855
Mark Haney: It does take commitment to keep your podcast.

230
0:24:20,936 --> 0:24:22,634
Mark Haney: It does take commitment to keep it going.

231
0:24:22,695 --> 0:24:27,054
Mark Haney: It doesn't pay the bills my friends, for me anyway, no, you know but it leads you to other opportunities.

232
0:24:27,095 --> 0:24:32,626
Ken Miller: It leads you to other opportunities and for a lot of people you know it's a labor of love.

233
0:24:32,715 --> 0:24:37,619
Ken Miller: It's them getting their voice out into the world and I think that that's a beautiful thing and an important thing.

234
0:24:38,155 --> 0:24:44,546
Ken Miller: But for those podcasts that do kind of drop off, they're replaced by even more people starting a podcast.

235
0:24:44,635 --> 0:24:54,225
Ken Miller: So we fully expect to see some amount of churn, and that will really play out, I think, as the company matures.

236
0:24:55,355 --> 0:25:00,922
Ken Miller: But that's going to be replaced by an equal or greater number of new podcasters.

237
0:25:01,035 --> 0:25:10,367
Ken Miller: So when you're faced with that kind of business situation, then marketing and branding become incredibly important.

238
0:25:11,216 --> 0:25:13,209
Mark Haney: So right now, it's just about getting market share.

239
0:25:13,772 --> 0:25:14,175
Mark Haney: Right now it's.

240
0:25:14,175 --> 0:25:15,019
Mark Haney: Is that the biggest obstacle?

241
0:25:15,675 --> 0:25:18,184
Mark Haney: Because I mean, your product is great.

242
0:25:18,255 --> 0:25:18,917
Mark Haney: I mean we have a.

243
0:25:19,178 --> 0:25:21,085
Mark Haney: We're going to give you a testimonial right here.

244
0:25:21,235 --> 0:25:22,160
Mark Haney: It is fantastic.

245
0:25:22,275 --> 0:25:23,881
Mark Haney: It's saving all kinds of time.

246
0:25:24,435 --> 0:25:24,938
Mark Haney: Ok, why?

247
0:25:25,675 --> 0:25:27,261
Mark Haney: It's like you found, it's like a discovery.

248
0:25:27,943 --> 0:25:31,841
Mark Haney: Right when I met you, it's like oh, my gosh, you're driving down the freeway, sell our sign.

249
0:25:32,755 --> 0:25:34,442
Mark Haney: My gosh, this guy's got a product we could use.

250
0:25:34,975 --> 0:25:41,965
Ken Miller: Well, it's super weird, because now I'm speaking into this mic and the algorithms are going to be processing this episode.

251
0:25:42,475 --> 0:25:45,538
Ken Miller: So that's a little meta, yeah, you know.

252
0:25:45,880 --> 0:25:48,963
Ken Miller: Yeah, market share, it's all about getting as much as possible.

253
0:25:48,975 --> 0:26:02,042
Ken Miller: There are competitors in the space, which is another reason why I really like Podbook, because you have to be crazy as an engineer to try to tackle the problem, incidentally, but there's nobody doing Podbook at all.

254
0:26:02,575 --> 0:26:05,578
Ken Miller: There are competitors in the like write my show notes for me space.

255
0:26:07,918 --> 0:26:20,345
Ken Miller: But again, we've put enough love into the AI to try to provide value to our customers to where we really have a significant advantage there and become the number one name in that space.

256
0:26:20,475 --> 0:26:30,384
Ken Miller: And that shows with the fact that Buzzsprout, which is by episode publishing volume, is the number two RSS host behind Spotify.

257
0:26:31,275 --> 0:26:52,124
Ken Miller: They have about 25% market share, and so the fact that they built their new co-host AI directly off of the Podium API so co-host AI from Buzzsprout it's all Podium behind the scenes Basically gave us distribution, direct distribution to 400,000 podcasters, and so we kind of got all the market share there.

258
0:26:52,795 --> 0:27:00,643
Ken Miller: And then we're also talking with several of the other RSS hosts that kind of occupy the long tail of the space.

259
0:27:01,455 --> 0:27:02,178
Mark Haney: Exciting stuff.

260
0:27:02,775 --> 0:27:13,523
Mark Haney: And I met you because you were driving down Interstate 80 and you saw a venture capital site on the side of Freeway and you're like this guy's a crazy dude putting venture capital on the side of the Freeway.

261
0:27:13,755 --> 0:27:30,967
Ken Miller: I thought it was the most interesting thing because when you're in the SF venture capital space everything is very cappuccino and kind of coying and I was like somebody just straight put venture capital on their side.

262
0:27:31,195 --> 0:27:33,057
Mark Haney: Poor Tesla venture capital, yeah, well, what?

263
0:27:33,077 --> 0:27:42,041
Ken Miller: I really appreciated, mark, about your particular approach to marketing is I think you're very, very good at that kind of remarketing thing, even within this building.

264
0:27:43,115 --> 0:27:45,011
Ken Miller: It's like I can't get your face out of my head.

265
0:27:45,836 --> 0:27:50,823
Ken Miller: So you really are good at occupying mind share and I think that shows with your approach there.

266
0:27:51,476 --> 0:27:52,961
Ken Miller: And hey, it worked.

267
0:27:53,041 --> 0:27:54,265
Ken Miller: Right, I'm here.

268
0:27:54,335 --> 0:28:07,604
Ken Miller: But yeah, I drove past it several times and once I finally got tired of I think we all were working in the house during COVID, but with my three kids it was getting increasingly difficult and I'm like, oh, I've got to get out of here and get an office.

269
0:28:07,635 --> 0:28:08,297
Ken Miller: So I'm like you know what?

270
0:28:09,120 --> 0:28:10,303
Ken Miller: I'm going to go get an office there.

271
0:28:10,404 --> 0:28:11,046
Mark Haney: That's cool.

272
0:28:11,135 --> 0:28:15,046
Mark Haney: Well, you have added a lot to the facility.

273
0:28:15,135 --> 0:28:25,661
Mark Haney: We have a co-working space here in Rockland with executive suites, and so Ken has one of those suites, and so I know people are coming up and asking you questions and bugging you and stuff like that.

274
0:28:25,735 --> 0:28:31,025
Mark Haney: So sorry about that, but we appreciate you having resident genius in the house.

275
0:28:32,635 --> 0:28:39,749
Mark Haney: But let's maybe talk about AI as a subject is on the tip of everybody's tongue.

276
0:28:40,096 --> 0:28:47,379
Mark Haney: You can't flip through the news Like CNBC I watch all the time with the financial show, ai is discussed.

277
0:28:48,755 --> 0:28:49,899
Mark Haney: It's about half the stories.

278
0:28:49,979 --> 0:28:52,258
Mark Haney: Or have AI mixed in it into it somewhere?

279
0:28:53,695 --> 0:29:10,507
Mark Haney: Maybe talk to us about the pros and cons of AI generally, because some of that we know it's a productivity tool, but some people have concerns about privacy or there's other things that robots taking over the world stuff like that.

280
0:29:10,975 --> 0:29:14,704
Mark Haney: Let me just give us a high level education.

281
0:29:15,887 --> 0:29:16,168
Ken Miller: Right.

282
0:29:18,219 --> 0:29:21,085
Ken Miller: Well, I'm definitely more on the optimist side.

283
0:29:22,655 --> 0:29:38,606
Ken Miller: I kind of look at the history of humanity and I think just the general fact that we haven't nuked ourselves yet is a testament to the maturity of the overall collective of the human organism, civilization-wise.

284
0:29:39,355 --> 0:29:40,260
Ken Miller: So I'm an optimist.

285
0:29:40,656 --> 0:29:42,440
Ken Miller: I think that it's not going to.

286
0:29:43,002 --> 0:29:43,543
Ken Miller: It's very.

287
0:29:43,683 --> 0:29:44,926
Ken Miller: People love doom.

288
0:29:47,120 --> 0:29:48,143
Ken Miller: We love zombie shows.

289
0:29:49,235 --> 0:29:59,666
Ken Miller: It's a strange fascination that people have with apocalyptic events and I think what ends up happening is that's what gets airtime.

290
0:30:01,196 --> 0:30:04,250
Ken Miller: It's like when we look at the news, right, it's a lot of negative stories.

291
0:30:04,616 --> 0:30:05,714
Ken Miller: It's a lot of negative stories.

292
0:30:05,815 --> 0:30:09,622
Ken Miller: So it's like you get one positive story for every nine doom and gloom stories.

293
0:30:10,819 --> 0:30:28,375
Ken Miller: And then what you get on the AI researcher side is you'll get people in the space who are legitimate AI researchers, but maybe in their career they haven't experienced any amount of significant popularity and they start talking about AI doom.

294
0:30:29,301 --> 0:30:31,255
Ken Miller: And then suddenly they're invited on podcasts.

295
0:30:31,396 --> 0:30:33,615
Ken Miller: Everybody wants to hear what they have to say.

296
0:30:33,736 --> 0:30:34,913
Ken Miller: And then what are you going to do?

297
0:30:36,226 --> 0:30:37,255
Ken Miller: Are you going to back off that doomsaing position?

298
0:30:39,204 --> 0:30:41,775
Ken Miller: No, because you're getting all of your flowers from it, right.

299
0:30:44,668 --> 0:30:46,115
Ken Miller: So I think there's a little bit of that going on.

300
0:30:48,085 --> 0:30:50,502
Ken Miller: Conversely, on the other hand, don't get these things guns.

301
0:30:54,451 --> 0:30:55,055
Ken Miller: That's a bad idea.

302
0:30:56,979 --> 0:31:03,955
Ken Miller: I think that there are some minor threats in the current AI space.

303
0:31:03,995 --> 0:31:08,155
Ken Miller: So, when we look at something like chatGPT, what are the threats that we face there?

304
0:31:09,540 --> 0:31:28,965
Ken Miller: Well, certainly kind of like an explosion of garbage content, or not garbage content, but content that's been AI generated specifically for business purposes, that don't serve any unique or distinct human value, but which successfully game, say, SEO algorithms right for Google.

305
0:31:28,985 --> 0:31:29,147
Ken Miller: So what?

306
0:31:29,975 --> 0:31:33,687
Ken Miller: You fast forward five years and now it's already bad enough.

307
0:31:33,949 --> 0:31:48,366
Ken Miller: On the internet there's enough garbage content, but now that we can generate garbage content automatically, that kind of exponentially grows and ends up compounding and existing.

308
0:31:49,359 --> 0:31:50,307
Mark Haney: It's more noisy out there yeah.

309
0:31:50,347 --> 0:31:51,616
Ken Miller: Just more noise Now.

310
0:31:52,459 --> 0:31:55,942
Ken Miller: But again, in this universe things tend to balance themselves out.

311
0:31:56,115 --> 0:32:07,226
Ken Miller: So while you all, on one hand, you have potentially more AI generated content noise, on the other hand AI can filter and sort that out for you, right?

312
0:32:08,899 --> 0:32:10,704
Ken Miller: So things tend to balance themselves out.

313
0:32:11,255 --> 0:32:25,965
Ken Miller: I definitely think that there's some issues around training, right, and the kinds of biases that a neural network may pick up on during the training processes.

314
0:32:27,477 --> 0:32:37,403
Ken Miller: So that's a definite issue, and then you can kind of try to work or mitigate the problem by working those biases out of the system.

315
0:32:37,535 --> 0:32:42,341
Ken Miller: But then maybe you go too far, at least from someone's opinion, right.

316
0:32:43,255 --> 0:33:14,645
Ken Miller: And so what you're going to get is, with a company like OpenAI, you're going to get people who have a very particular perspective on the world, very particular perspective on politics, a very particular perspective on ethics and epistemology, and those people are all in San Francisco and they're the ones who are training the neural network, and so you can absolutely expect unsurprisingly, that the neural network is going to be biased to their worldview.

317
0:33:16,096 --> 0:33:20,307
Ken Miller: So those kinds of biases are also like a potential problem.

318
0:33:20,375 --> 0:33:30,225
Ken Miller: But again, it's kind of like I don't discount human agency and people's ability to think for themselves and to judge and discriminate.

319
0:33:30,415 --> 0:33:31,525
Mark Haney: And we solve problems.

320
0:33:31,635 --> 0:33:33,041
Mark Haney: If we create problems, we solve them.

321
0:33:34,159 --> 0:33:35,163
Ken Miller: We tend to solve them.

322
0:33:36,958 --> 0:33:54,508
Ken Miller: So I think there's a lot of people out there in the AI space who are looking at regulation of these things, but potentially, I think that's dangerous too, because you can absolutely count on this.

323
0:33:54,695 --> 0:33:57,819
Mark Haney: This type of creativity and innovation Slow it down.

324
0:33:58,175 --> 0:34:04,605
Ken Miller: What you're going to get is you're going to get cronyism and lock-in in Washington.

325
0:34:04,826 --> 0:34:05,387
Ken Miller: I think from that.

326
0:34:06,055 --> 0:34:20,200
Mark Haney: And maybe, just maybe, from an educational standpoint, what have we been using over the last several years that is, as consumers or as business people that really has AI baked into it, because right now, chat GPT comes out, everybody starts using it.

327
0:34:20,315 --> 0:34:22,303
Mark Haney: It's a fun for some people, it's a fun toy.

328
0:34:22,435 --> 0:34:45,018
Mark Haney: For other people, it's a productivity tool, and then a lot of startups are happening because AI but it's really new on the scene for a lot of us, but we've been using AI in different products for a while, and so maybe walk us through just some examples of what that is and then maybe a little bit of how you envision it changing certain industries.

329
0:34:46,102 --> 0:34:46,383
Ken Miller: Right.

330
0:34:46,535 --> 0:34:57,439
Ken Miller: So well, I don't know if I'd say that we've been using AI and actually like the entire term AI, artificial intelligence.

331
0:34:58,161 --> 0:35:00,828
Ken Miller: What does the word intelligence even mean?

332
0:35:00,955 --> 0:35:04,097
Ken Miller: I think people sometimes maybe don't pay close enough attention to words.

333
0:35:05,275 --> 0:35:10,401
Ken Miller: In-telling, inner telling, right, and who are you even telling inside?

334
0:35:11,175 --> 0:35:12,460
Ken Miller: I think is an interesting question.

335
0:35:13,155 --> 0:35:22,398
Ken Miller: I like the term artificial cognition, right, and it really is more machine-like and cog-like in a sense.

336
0:35:23,815 --> 0:35:27,619
Ken Miller: But generally when we talk about AI there's kind of two categories.

337
0:35:27,679 --> 0:35:36,383
Ken Miller: It falls into weak AI and strong AI, and then a third category that's a major leap, which is AGI, or artificial general intelligence.

338
0:35:37,495 --> 0:35:49,297
Ken Miller: Weak AI is AI, specifically neural networks that are trained to do one specific task really really well.

339
0:35:50,155 --> 0:35:51,702
Ken Miller: That task could be anything.

340
0:35:51,855 --> 0:35:58,463
Ken Miller: It could be predicting when you're going to leave your house every day, right, and through learning it might learn.

341
0:35:58,623 --> 0:36:01,121
Ken Miller: Ok, mark gets up at 5 AM.

342
0:36:01,575 --> 0:36:11,379
Ken Miller: He's generally out the door at 6.15 for a run or whatever, but except for on Sundays then he's got a different routine and so it can pick a weak.

343
0:36:11,499 --> 0:36:13,425
Ken Miller: AI can pick up on patterns and predict things.

344
0:36:13,495 --> 0:36:18,367
Ken Miller: This is really good for things like recommendation of products and things like that.

345
0:36:18,675 --> 0:36:19,298
Mark Haney: Turning on and off.

346
0:36:19,318 --> 0:36:25,082
Mark Haney: We sold a company that turns on and off your air conditioner based upon some of those kind of habits there you go.

347
0:36:25,655 --> 0:36:29,961
Ken Miller: It's cold, weak AI, but it's still very powerful and it's very limited space.

348
0:36:30,735 --> 0:36:37,580
Ken Miller: What we're starting to see here with the advent of chat, gpt and, in particular, large language models, is what you would call strong AI.

349
0:36:38,475 --> 0:36:49,401
Ken Miller: Strong AI has more general abilities, right, you can ask chat GPT about explain to me the fundamental nature of quantum physics.

350
0:36:49,955 --> 0:36:56,722
Ken Miller: You can also ask it for a series of recipes that involve cayenne powder and garlic powder.

351
0:36:57,135 --> 0:37:03,828
Ken Miller: I mean, really the potential is quite vast there.

352
0:37:04,475 --> 0:37:25,123
Ken Miller: But as far as artificial intelligence goes, as it pertains to neural networks as the underlying learning mechanism, this has only really been in play in a business productivity way, like, so like within the engineering department, since maybe 2015.

353
0:37:26,150 --> 0:37:40,985
Ken Miller: We've actually been using machine learning for a very long time, since the early 2000s, right, but there were a different series of learning algorithms for machines that we would use because it took too much compute to actually model a neural network.

354
0:37:41,046 --> 0:38:02,135
Ken Miller: But what they came to discover, incidentally, is that just like perhaps nature and sort of like the evolution of organisms on this planet came to discover, is that neural networks as a fundamental mechanism for predicting the future is actually the most flexible and efficient way to go about it.

355
0:38:02,230 --> 0:38:05,884
Ken Miller: So, if you have enough compute now everything is neural networks right.

356
0:38:05,904 --> 0:38:23,899
Ken Miller: We don't use, or we tend not to use, some of those older machine learning algorithms, but the general idea of statistical machine learning has been in play, I mean since you know the early 2000s, really for product recommendations, amazon right, netflix and a variety of other use cases.

357
0:38:23,939 --> 0:38:52,204
Ken Miller: I think what really ended up happening from a like kind of industry insider standpoint is in 2017, there was this paper published called All you Need Is Attention, and that was literally the title of the paper, and the researchers who wrote it basically dug up because neural networks have been around since the 70s, but it just took too much compute in order to do anything really truly interesting with it.

358
0:38:53,147 --> 0:39:05,760
Ken Miller: This idea of baking attentional mechanisms into neural networks mathematically it's been around since like the mid 80s or early 90s, but again, not enough compute to do anything with it.

359
0:39:06,882 --> 0:39:13,772
Ken Miller: So some researchers dug up that old research and said, hey, what if we took these attentional mechanisms, made a few changes?

360
0:39:13,852 --> 0:39:17,467
Ken Miller: Now we have enough compute right, and they published this paper.

361
0:39:17,587 --> 0:39:20,495
Ken Miller: Attention is All you Need, and it was the key.

362
0:39:21,237 --> 0:39:33,324
Ken Miller: So the performance on various tasks, like the identification of objects and images, the ability to like, summarize text, the ability to extract an answer from text.

363
0:39:34,150 --> 0:39:39,082
Ken Miller: The performance scores went up through the roof as soon as you baked in these attentional mechanisms.

364
0:39:39,531 --> 0:39:50,450
Ken Miller: So it was really at that moment in time in late 2017, when that paper was published, that kind of sparked this whole revolution that now we're sitting in, six years later.

365
0:39:50,510 --> 0:39:55,573
Mark Haney: Yeah Well, so thinking about some of the industries, I'm in the security industry, so facial recognition is big.

366
0:39:55,593 --> 0:39:57,477
Mark Haney: I just saw self-driving cars.

367
0:39:57,517 --> 0:40:02,512
Mark Haney: Our cars are self-driving at some level.

368
0:40:02,552 --> 0:40:12,201
Mark Haney: Now we're invested into a number of companies that are using AI at some level and to even solve disease and things like that.

369
0:40:12,241 --> 0:40:13,866
Mark Haney: So maybe paint a picture.

370
0:40:13,906 --> 0:40:25,278
Mark Haney: And it's changing so fast now because AI, I mean it has come upon us faster than anything I've seen in my life, faster than mobile, faster than when computers hit the.

371
0:40:25,618 --> 0:40:32,022
Mark Haney: You know, when computers and mobile and social media and all these different trends are, excuse me, turning points.

372
0:40:33,211 --> 0:40:35,359
Mark Haney: They all happen pretty fast, but this has happened faster.

373
0:40:36,830 --> 0:40:45,930
Ken Miller: Intelligence is a great product, you know, in any way it goes, and it's well.

374
0:40:45,950 --> 0:40:48,837
Ken Miller: I think there's a lot of factors involved there, right?

375
0:40:48,937 --> 0:41:00,445
Ken Miller: So one of the interesting things about AI that's different than any other product is that AI helps make better AI.

376
0:41:01,387 --> 0:41:25,592
Ken Miller: So I think, to a certain extent, some of that exponential or compound growth that we're all feeling which can be a little bit anxiety inducing for people, you know, is part of that is that you know, once you have intelligence as a product, or cognition as a product, you know you're gonna, that's gonna assist you in making further advances.

377
0:41:26,374 --> 0:41:43,358
Ken Miller: So, whereas a car I mean in that that even, like you know, potentially applies to something like like a car, you know, I'm sure, once the first cars came out and we're rolling off the line, it's like, oh well, now we don't have to use horses, we can actually get parts to the factory fast.

378
0:41:43,378 --> 0:41:45,592
Ken Miller: We're gonna use cars to make cars right, yeah.

379
0:41:45,952 --> 0:41:58,448
Ken Miller: So all technology has a bit of that kind of you know potential for compound, you know advancement baked into it, but intelligence itself as a product.

380
0:41:58,529 --> 0:42:00,475
Ken Miller: It's kind of like you know that that's.

381
0:42:00,536 --> 0:42:02,663
Ken Miller: That's the mountain peak there.

382
0:42:03,104 --> 0:42:11,385
Ken Miller: And then beyond that, what's also happened is you have every math major on the planet now is doing AI research.

383
0:42:12,307 --> 0:42:22,980
Ken Miller: It really is just a lot of you know statistics and probability theory and all of the other various branches of math kind of all smashed together, and so that's really been contributing to.

384
0:42:23,040 --> 0:42:31,814
Ken Miller: Like if you look at the, the research papers that are published out of both private and public institutions, and you look at the number, it is an exponential curve.

385
0:42:32,996 --> 0:42:37,714
Ken Miller: So so that's really what's kind of like powering a lot of these advancements.

386
0:42:37,895 --> 0:42:50,442
Ken Miller: But you never know when you're going to hit a wall and I actually think, like GPT four, until some other major major leap is made.

387
0:42:50,502 --> 0:42:52,565
Ken Miller: Gpt four has actually hit a wall.

388
0:42:54,391 --> 0:43:09,780
Ken Miller: In fact, it was kind of leaked just about a week ago because they've been very hush hush about what the underlying architecture is, despite their name open AI but it was leaked the fundamental architecture and they're using this thing called mix of experts.

389
0:43:10,481 --> 0:43:25,302
Ken Miller: So GPT four is actually I think it's five, yeah, five neural networks that are 220 billion parameters, just slightly, actually considerably larger than than GPT three.

390
0:43:25,382 --> 0:43:31,138
Ken Miller: But everybody thought, oh, this is going to be like a 500 trillion parameter models gonna be massive and massively forward.

391
0:43:31,158 --> 0:43:46,282
Ken Miller: Now it's really they have five different models, but each model is trained on a different subset of data and so they feed it through all five and then they use the output from the model that's most confident.

392
0:43:46,323 --> 0:43:47,672
Ken Miller: It's called mix of experts.

393
0:43:48,013 --> 0:43:49,937
Ken Miller: The thing is mix of experts.

394
0:43:50,338 --> 0:43:53,835
Ken Miller: That kind of model architecture is what you do when you're out of ideas.

395
0:43:54,598 --> 0:43:56,766
Mark Haney: So open AI is what we're talking about Now.

396
0:43:56,806 --> 0:43:57,790
Mark Haney: That's got competitors.

397
0:43:59,273 --> 0:44:00,636
Ken Miller: Kind of yeah, yeah.

398
0:44:01,217 --> 0:44:06,115
Mark Haney: And when you say, kind of what you're saying, it doesn't really have competitors, well, you know, they're, they're gonna.

399
0:44:06,136 --> 0:44:07,119
Mark Haney: Is there going to be one big winner?

400
0:44:09,071 --> 0:44:09,532
Mark Haney: It's it's.

401
0:44:09,993 --> 0:44:13,222
Ken Miller: It's really interesting mark, like the way that the space has shaped up.

402
0:44:15,250 --> 0:44:23,954
Ken Miller: There was this paper leaked again from Google where one of the researchers was basically saying we have no moat.

403
0:44:24,094 --> 0:44:31,658
Ken Miller: You know, the open source large language models are advancing so quick, you know, and open AI doesn't have any moat either.

404
0:44:31,778 --> 0:44:38,058
Ken Miller: Right, like the open source research community is going to surpass the ability of any private institution.

405
0:44:39,242 --> 0:45:00,156
Ken Miller: But what's interesting is that in the graphs where they were showing the performance of the open source models versus something like the original chat GPT, the performance analysis was coming from GPT for's opinion of how well the things are, so I'm like come on guys like are you?

406
0:45:00,176 --> 0:45:01,099
Ken Miller: just blind to this.

407
0:45:01,700 --> 0:45:08,311
Ken Miller: Clearly, they have a moat if you're using their highest end neural network to even judge the performance of everybody else's right.

408
0:45:09,052 --> 0:45:11,336
Ken Miller: So, um, yeah, you know.

409
0:45:11,476 --> 0:45:31,184
Ken Miller: I think that, uh, google has been a little bit late to the game, and I mean, for obvious reasons, they're already making a tremendous amount of money, and so there's not a whole lot of incentive for them to until they have some kind of Google search right until they have some kind of fundamentally existential threat.

410
0:45:31,284 --> 0:45:37,228
Ken Miller: And as soon as Microsoft came in and integrated chat GPT into being, that existential threat came to bear.

411
0:45:37,248 --> 0:45:47,496
Ken Miller: And so now you've got Palm, you've got some of these other things, but there really is nothing out there that competes with the competence of GPT.

412
0:45:47,556 --> 0:45:52,912
Ken Miller: For although Google deep mind um you know, amazon do they have?

413
0:45:52,992 --> 0:45:54,115
Ken Miller: something on the horizon.

414
0:45:54,155 --> 0:45:56,882
Ken Miller: Yeah, amazon's working on a new offering called Bedrock.

415
0:45:57,850 --> 0:46:08,676
Ken Miller: They're closely integrated with another company called Anthropic and uh, anthropics got some, some great models Again, not quite GPT four level, but some decent models for lower end tasks.

416
0:46:08,716 --> 0:46:09,358
Mark Haney: How about meta?

417
0:46:09,398 --> 0:46:10,321
Mark Haney: Are they in the game?

418
0:46:10,803 --> 0:46:22,335
Ken Miller: Yeah, so meta uh released, uh, a model called llama Um, and actually it was just a research model, but uh, it got leaked All these things, they end up getting leaked, right.

419
0:46:22,395 --> 0:46:23,758
Ken Miller: So it got leaked.

420
0:46:23,798 --> 0:46:31,763
Ken Miller: And actually what's crazy is is Meta's llama model actually sparked the entire open source LLM revolution.

421
0:46:31,803 --> 0:46:47,153
Ken Miller: When that thing was leaked, the kind of people in the open source space now had a massive what we call foundational model to begin working from, and so the open source community began applying a lot of different AI training techniques to it.

422
0:46:47,634 --> 0:46:52,655
Ken Miller: And next thing, you know, you have their model, which was kind of it was okay, but it wasn't great.

423
0:46:53,116 --> 0:47:05,757
Ken Miller: But after the open source community got ahold of it, now you have other, uh, you know, uh, models that are in a similar fashion, named after camel, like creatures, like vacuna and stuff like that.

424
0:47:06,519 --> 0:47:15,801
Ken Miller: Uh, you know everybody's funny right, uh, that are performing at GPT three point five levels, like, like, like standard GPT three levels.

425
0:47:15,841 --> 0:47:23,351
Ken Miller: So yeah, metas in the game, amazon's in the game, google is in the lion's share is going to GPT four right now.

426
0:47:23,391 --> 0:47:30,175
Mark Haney: That's where the guys like you, the players, the guys that know what they're doing in a big way they're mostly on GPT four.

427
0:47:30,215 --> 0:47:35,711
Ken Miller: Well, it's, it's, you know it's the best, it's just it's there there.

428
0:47:36,072 --> 0:47:48,323
Ken Miller: There are no other models that offer its level of cognitive capabilities, not merely as, like, say, a summarization engine, but also just as a reasoning engine, right.

429
0:47:48,443 --> 0:48:03,380
Ken Miller: So um, gpt four is able to reason out solutions to problems and in fact you know some of the other um growth factory portfolio companies, like a 4pm right that um has recently come out.

430
0:48:04,262 --> 0:48:23,441
Ken Miller: They're really leveraging those um uh reasoning engine capabilities of GPT four, where they're essentially taking a conversation and then reasoning which portions of those conversations map to fields in a CRM that takes reasoning capability.

431
0:48:23,902 --> 0:48:24,906
Mark Haney: Yeah, that's an interesting one.

432
0:48:24,946 --> 0:48:36,533
Mark Haney: We'll probably talk more about that one, but yeah, it's basically bolting on or working with a you know a sales force or you know a hub spot or something like that, and making it making the job of the sales rep, uh, a lot easier.

433
0:48:36,614 --> 0:48:37,335
Ken Miller: And you know it's.

434
0:48:37,456 --> 0:48:38,157
Mark Haney: So that's a, that's a.

435
0:48:38,959 --> 0:48:43,020
Mark Haney: It's interesting, just a little way, not little, that's a big way that could transform a sales force.

436
0:48:43,924 --> 0:48:48,101
Ken Miller: I like to say every single piece of software at this point is going to be rewritten.

437
0:48:48,402 --> 0:48:49,205
Ken Miller: Yeah, Interesting.

438
0:48:49,487 --> 0:49:02,525
Mark Haney: Okay, so I'm going to circle back around to something you touched on, uh, at the beginning of the show, and that was, uh, job, jobs, right, job displacement, but yet job, job creation.

439
0:49:02,545 --> 0:49:17,954
Mark Haney: You got all these math wizards that are going to have new types of jobs, and there are other jobs getting created out of uh, out of AI, or this, uh, this technology, and there are going to be some people that you know, some jobs that just aren't needed as much.

440
0:49:18,014 --> 0:49:19,498
Mark Haney: They're kind of the some of those redundant jobs.

441
0:49:19,779 --> 0:49:21,231
Mark Haney: How do you see that shaking out?

442
0:49:22,013 --> 0:49:26,551
Ken Miller: You know, I, I don't think it's any different mark than any other technological advancement.

443
0:49:26,692 --> 0:49:30,339
Ken Miller: Okay, uh, you don't see too many blacksmiths walking around.

444
0:49:31,221 --> 0:49:39,380
Ken Miller: Um, you know, and I'm sure that if you go back in the history books there were probably blacksmiths saying you can't trust the strength of the metal coming out of the.

445
0:49:39,902 --> 0:49:43,334
Ken Miller: You know the machines that are machining them, the metal, you can't trust it.

446
0:49:43,676 --> 0:49:45,421
Ken Miller: It hasn't been hammered on by me, you know.

447
0:49:46,024 --> 0:49:54,307
Ken Miller: So, uh, but did advancements in you know, manufacturing and milling and what not did?

448
0:49:54,387 --> 0:49:56,975
Ken Miller: Did those create less jobs or more jobs?

449
0:49:57,245 --> 0:50:00,167
Ken Miller: Ultimately created more jobs, even, you know.

450
0:50:00,227 --> 0:50:03,296
Ken Miller: But the jobs shift, they change.

451
0:50:03,845 --> 0:50:06,374
Ken Miller: Humans are incredibly adaptable, beautiful organisms.

452
0:50:06,414 --> 0:50:11,834
Ken Miller: You know there's lots of different things that we can do and we will always find creative ways to occupy our time.

453
0:50:12,865 --> 0:50:14,271
Ken Miller: Is it gonna hurt a little bit?

454
0:50:14,365 --> 0:50:18,993
Ken Miller: Yeah, and I think that we need to be, you know, compassionate in that way.

455
0:50:19,426 --> 0:50:32,272
Ken Miller: I think that, you know, I'm even less worried about some of the white color impacts that you know GPT or any of the large language models really potentially represent.

456
0:50:32,345 --> 0:50:36,530
Ken Miller: I'm really more interested in, like you know, big rig truck drivers.

457
0:50:37,665 --> 0:50:50,615
Ken Miller: You know that's like one of the number one jobs, highest paying jobs, you know, a job where you can really support your family but doesn't require, you know, a tremendous amount of formal education, a lot of hard work And-.

458
0:50:51,605 --> 0:50:54,113
Mark Haney: You don't need that guy because the truck drives itself.

459
0:50:55,327 --> 0:50:55,728
Ken Miller: The truck.

460
0:50:56,009 --> 0:51:03,651
Ken Miller: It will drive itself, trust me, before 2030, the truck, I mean, we already have the tech, it just needs a bit more refinement.

461
0:51:03,765 --> 0:51:05,070
Ken Miller: It will absolutely get there.

462
0:51:05,685 --> 0:51:12,854
Ken Miller: I'd say it's even behind where I thought it was gonna be by about five years, but yeah, it's gonna get there.

463
0:51:13,005 --> 0:51:17,571
Ken Miller: And then you're going, you will have, it is cheaper, so you will have that.

464
0:51:18,294 --> 0:51:29,953
Ken Miller: You know, labor force displaced, essentially, and that needs to be thought through, you know, as a society, but again, you're not going to stop it.

465
0:51:31,506 --> 0:51:32,429
Ken Miller: You can't stop it.

466
0:51:32,570 --> 0:51:38,492
Mark Haney: So Well, I think it allows, it's gonna allow, maybe, the truck driver to do something that uses a little bit more imagination, right?

467
0:51:38,585 --> 0:51:44,567
Mark Haney: So how could he help the trucking industry in a different way, or the transportation industry in a different way?

468
0:51:44,988 --> 0:51:52,770
Mark Haney: Instead of, you know, sitting behind the wheel of a vehicle, he could do something more productive and interesting for himself or herself.

469
0:51:53,065 --> 0:51:59,133
Ken Miller: Right, and you know, I mean I think that there's lots of people who have put that viewpoint out there.

470
0:51:59,406 --> 0:52:07,232
Ken Miller: But I think, if we're really being honest with ourselves, you have a 43 year old, 45 year old truck driver, maybe even a 50 year old truck driver.

471
0:52:07,305 --> 0:52:10,166
Ken Miller: Been driving truck for 15, 20 years, whatever.

472
0:52:10,206 --> 0:52:11,310
Ken Miller: Maybe they changed careers.

473
0:52:12,686 --> 0:52:21,731
Ken Miller: It pays well, right, hard work breaks the body and then you take that away because now the trucks drive themselves.

474
0:52:22,765 --> 0:52:26,972
Ken Miller: I don't think it's gonna be like oh yeah, you know, go write poetry now, you know.

475
0:52:27,426 --> 0:52:31,791
Ken Miller: It's like that person's life is going to be fundamentally disrupted.

476
0:52:31,845 --> 0:52:44,113
Ken Miller: Now, the next generation is not gonna have that problem, cause they're never gonna get into truck driving, but you're going to have 20 years of It'll be a little bit of pain or adjustment, adjustment, people suffering, you know.

477
0:52:44,285 --> 0:52:52,875
Ken Miller: And so I think if you wanna look at, like, what the real danger for AI is, it's that kind of job displacement.

478
0:52:53,105 --> 0:53:08,004
Ken Miller: I think it's a lot easier for people in the white collar space to make that kind of, you know, creative shift than it is for people in the manual labor space.

479
0:53:08,105 --> 0:53:08,446
Mark Haney: It's interesting.

480
0:53:08,486 --> 0:53:14,156
Mark Haney: So I own a concrete company and it's not gonna pour the concrete for us right.

481
0:53:14,285 --> 0:53:17,254
Mark Haney: So we have trouble getting enough labor.

482
0:53:17,535 --> 0:53:26,311
Mark Haney: You know to do all the work right when there's more demand out there than we have the resources to finish the concrete and so on, because it's very.

483
0:53:26,392 --> 0:53:29,430
Mark Haney: You know we push the wheelbarrows around and you know it's manual labor.

484
0:53:31,566 --> 0:53:32,691
Mark Haney: How's it gonna affect the trades?

485
0:53:33,066 --> 0:53:34,670
Mark Haney: How is AI gonna affect?

486
0:53:34,811 --> 0:53:36,956
Mark Haney: Other than make us plan better?

487
0:53:37,145 --> 0:53:38,952
Mark Haney: And you know a lot of the behind the scenes?

488
0:53:38,972 --> 0:53:40,129
Mark Haney: Stuff will happen a lot faster.

489
0:53:40,545 --> 0:53:48,828
Mark Haney: But in terms of swinging a hammer or, you know, finishing concrete, do you see any way that it can affect that stuff?

490
0:53:49,832 --> 0:53:50,313
Ken Miller: Oh yeah, absolutely.

491
0:53:51,006 --> 0:53:52,332
Ken Miller: I didn't realize you own a concrete.

492
0:53:52,425 --> 0:53:53,932
Ken Miller: Concrete's such a fascinating sub.

493
0:53:54,125 --> 0:54:00,834
Ken Miller: You know it's the most produced manmade substance that every other manmade substance plastics everything.

494
0:54:01,565 --> 0:54:05,890
Ken Miller: If you add all of them up by way, we make more concrete.

495
0:54:06,345 --> 0:54:06,727
Ken Miller: It's crazy.

496
0:54:07,705 --> 0:54:09,813
Ken Miller: Yeah, so GPT is not gonna pour your concrete.

497
0:54:10,808 --> 0:54:18,733
Ken Miller: However, there will be a robot that will pour your concrete for you and you'll be able to give it orders and talk to it, because of GPT like technology.

498
0:54:18,905 --> 0:54:19,146
Ken Miller: Right.

499
0:54:19,286 --> 0:54:27,296
Ken Miller: So, yeah, I think, but to get there, I mean, we greatly underestimate this thing.

500
0:54:27,525 --> 0:54:47,249
Ken Miller: This human body is a masterwork and to get a robot to be anywhere near as energy efficient or dexterous as the human body, we are a long, yeah, we are a far cry from that, and but I think you will get.

501
0:54:47,390 --> 0:55:00,415
Ken Miller: In the next like 30 to 50 years, you will have robots that basically, yeah, replace a lot of manual labor.

502
0:55:01,889 --> 0:55:29,068
Ken Miller: But whether or not they're gonna, I mean, what's interesting is that in the past, when it's come to robotics, it's all been about almost like a Henry Ford-like assembly line type automation of a repetitive task that can be done with precision every single time, like that singular spot weld from a robot arm, it can go in the same place, and then you have error tolerances For some of this stuff.

503
0:55:29,389 --> 0:55:30,693
Ken Miller: That's more manual labor.

504
0:55:31,225 --> 0:55:40,711
Ken Miller: You do need you know when you're working out in nature and you need to have that human body or that kind of like dexterous thing.

505
0:55:40,785 --> 0:55:51,856
Ken Miller: So I think there's a question as to when robotics will get to that point to where, say like, an iron worker might be displaced by a robot.

506
0:55:52,366 --> 0:55:58,092
Ken Miller: I think we're probably about 20, 30, potentially even 50 years out from that.

507
0:55:58,445 --> 0:56:01,433
Mark Haney: But we are using robots, a couple of portfolio companies at the growth factory.

508
0:56:01,625 --> 0:56:10,051
Mark Haney: One inspects solar farms, another one is a drone that inspects airports and stuff like that.

509
0:56:10,988 --> 0:56:12,814
Mark Haney: So it's coming, we're moving in that direction.

510
0:56:13,265 --> 0:56:13,887
Ken Miller: It's coming.

511
0:56:14,168 --> 0:56:15,993
Ken Miller: It's when you have to grab something.

512
0:56:16,905 --> 0:56:18,993
Ken Miller: It's the grabbing of something you know.

513
0:56:19,445 --> 0:56:21,304
Ken Miller: The human hand is a incredible tool.

514
0:56:21,425 --> 0:56:23,854
Mark Haney: We're invested into a strawberry picker type company.

515
0:56:23,925 --> 0:56:26,232
Mark Haney: That's a robotic and it's like a hand.

516
0:56:27,865 --> 0:57:05,290
Ken Miller: Again, it's one of those things where, yes, it can pick strawberries, but can it pick watermelons Right, and so that's the thing is that you might get a robot where, because of the fundamental shape of the form and material reality, you're able to create a robot that handles that singular task, but can the fundamental mechanical mechanisms of that bot being then generalize to other things, Whereas you can take someone who picks strawberries and I guarantee you they're gonna pick apples almost equally as well.

517
0:57:05,845 --> 0:57:07,291
Mark Haney: Okay, let me give you the last word here.

518
0:57:07,465 --> 0:57:15,665
Mark Haney: We're about out of time, but I wanna make sure, if there's anything I didn't ask you or you wanna make sure that our audience has a perspective on, maybe.

519
0:57:15,685 --> 0:57:16,970
Mark Haney: I'll just give you the last word here.

520
0:57:17,765 --> 0:57:34,647
Ken Miller: Yeah, you know, I would say for everybody out there who's kind of maybe feels nervous about like the advent of AI or is looking at all of the doom, saying I'll go back to.

521
0:57:34,667 --> 0:57:52,257
Ken Miller: We have nothing to fear but fear itself, stay optimistic and really consider the ways in your life that you may be able to use these tools creatively to enhance your life and enhance your productivity.

522
0:57:56,033 --> 0:58:00,432
Ken Miller: And don't pay too much attention to everybody who thinks it's all gonna fall apart because of AI.

523
0:58:01,445 --> 0:58:03,331
Ken Miller: I mean, I don't know anything.

524
0:58:03,531 --> 0:58:29,450
Ken Miller: I know a thing or two, but I have a strong feeling that what we're gonna see in the next 20 to 50 years is actually a increased flourishing for humanity, new inventions, new ways to solve problems that previous generations have, you know, somewhat naïvely or unknowingly created through their use of technology.

525
0:58:29,605 --> 0:58:33,592
Ken Miller: So I'm an optimist about everything and I think there's some great things coming down the road.

526
0:58:34,005 --> 0:58:45,612
Mark Haney: Ken Miller, thank you for the education, appreciate you being a part of the group factory and what we're doing around here on Granite Drive, and I look forward to learning more from you as over the years to come.

527
0:58:45,885 --> 0:58:46,913
Ken Miller: For sure, thanks for having me, Mark.

528
0:58:47,045 --> 0:58:48,284
Ken Miller: Thanks for that.