1
00:00:00,040 --> 00:00:02,520
Imagine taking a piece of 
marketing literature and 

2
00:00:02,520 --> 00:00:05,160
redacting any word that mentions
AI. 

3
00:00:05,160 --> 00:00:07,440
Why are you buying this software
solution? 

4
00:00:07,440 --> 00:00:10,680
The AI is a tool. 
The AI is a component in the 

5
00:00:10,680 --> 00:00:12,480
car. 
It's part of the engine. 

6
00:00:12,600 --> 00:00:16,320
AI is amazing, but if you don't 
have a great service offering or

7
00:00:16,320 --> 00:00:19,840
workflow or solution, it doesn't
matter what kind of magic 

8
00:00:19,840 --> 00:00:22,280
sprinkles you put on top. 
Of it, I'm Ryan Heath, and 

9
00:00:22,280 --> 00:00:26,120
welcome to the AI Leadership Lab
where I interview C-Suite 

10
00:00:26,120 --> 00:00:29,720
leaders about their AI journeys.
We cover what's working, where 

11
00:00:29,720 --> 00:00:32,600
they're getting stuck, how they 
pivot, and the lessons they take

12
00:00:32,600 --> 00:00:35,640
from the losses. 
Joining us now is Ryan Stilberg,

13
00:00:35,640 --> 00:00:39,080
CEO of Veritone. 
Ryan, welcome to the podcast, 

14
00:00:39,080 --> 00:00:40,640
first of all. 
Ryan, great name. 

15
00:00:40,640 --> 00:00:42,240
Thank you for hosting me on the 
show. 

16
00:00:42,840 --> 00:00:45,920
We're already winning now. 
Veritone is not an old company 

17
00:00:45,920 --> 00:00:51,400
by any means, but you have been 
around before the advent or the 

18
00:00:51,440 --> 00:00:55,400
deluge of generative AI over the
last three years, and you've got

19
00:00:55,720 --> 00:00:58,080
a deep history in the ad tech 
world. 

20
00:00:58,280 --> 00:01:01,280
Can you tell us a little bit 
about your own AI journey and 

21
00:01:01,520 --> 00:01:04,200
sort of how much you've had to 
retool over the last three 

22
00:01:04,200 --> 00:01:05,720
years? 
You know, it, it's, it has been 

23
00:01:05,720 --> 00:01:09,080
an exciting journey and, and 
obviously the speed in an 

24
00:01:09,080 --> 00:01:12,480
acceleration advancement of AI, 
right, as I think impressing and

25
00:01:12,480 --> 00:01:15,320
and dazzling you and I, right, 
who've sort of been in the space

26
00:01:15,320 --> 00:01:17,680
now for a while. 
But but you know, our journey 

27
00:01:17,680 --> 00:01:21,960
with AI really started from an 
ad tech problem, trying to 

28
00:01:21,960 --> 00:01:25,440
understand how I can target an 
ad against a piece of audio and 

29
00:01:25,440 --> 00:01:27,720
video. 
Primarily when we started to 

30
00:01:27,720 --> 00:01:30,680
sort of see the writing on the 
wall back in the day that you 

31
00:01:30,680 --> 00:01:32,960
weren't just going to see static
websites, but you're going to 

32
00:01:32,960 --> 00:01:36,280
see a very organic audio and 
video ecosystem on the web. 

33
00:01:36,480 --> 00:01:40,600
We needed a way to start to 
understand and process audio and

34
00:01:40,600 --> 00:01:44,640
video at scale, transcription, 
object detection, machine 

35
00:01:44,640 --> 00:01:47,400
vision. 
Those became critical components

36
00:01:47,400 --> 00:01:51,280
for us to start to create 
structure around audio and 

37
00:01:51,280 --> 00:01:54,560
video, right. 
And so I think because of our 

38
00:01:54,560 --> 00:01:59,560
background in ad tech, our entry
point into this ecosystem of AI,

39
00:01:59,960 --> 00:02:04,280
again, was really about bringing
structure to unstructured, messy

40
00:02:04,280 --> 00:02:05,120
audio and video. 
Yeah. 

41
00:02:05,240 --> 00:02:07,800
And honestly, that's so 
important that point about 

42
00:02:07,880 --> 00:02:10,759
bringing structure to all of it,
because what I found is the 

43
00:02:10,759 --> 00:02:14,920
reality is most organizations 
data is just a total mess. 

44
00:02:14,920 --> 00:02:17,680
And then one of the hurdles that
they sort of perceived 

45
00:02:17,800 --> 00:02:20,720
internally about their ability 
to use AI is like, well, our 

46
00:02:20,720 --> 00:02:23,680
data isn't organized. 
So we then can't apply it to any

47
00:02:23,680 --> 00:02:25,400
of this new AI. 
And it's, it's really 

48
00:02:25,400 --> 00:02:27,920
interesting to hear you talk 
about how that doesn't 

49
00:02:28,000 --> 00:02:30,800
necessarily need to be true. 
They've, they've been people 

50
00:02:30,800 --> 00:02:32,880
like yourselves have been out 
there thinking about ways to 

51
00:02:32,880 --> 00:02:36,240
bring structure to all of that. 
And that can help people get out

52
00:02:36,240 --> 00:02:39,120
of that kind of like first 
hurdle that they sometimes feel 

53
00:02:39,120 --> 00:02:41,560
stuck with when it comes to 
their own AI journey. 

54
00:02:42,280 --> 00:02:44,960
You know, I think companies, 
again, we started in meaning 

55
00:02:44,960 --> 00:02:48,360
entertainment and I like to say 
is, you know, if you're ambling 

56
00:02:48,360 --> 00:02:51,920
in its studios or NBC Universal,
a couple of our clients, you 

57
00:02:51,920 --> 00:02:55,680
know, they're experts in their 
primary data asset, which is the

58
00:02:55,680 --> 00:02:59,640
audio and video itself, right? 
For most other companies, I 

59
00:02:59,640 --> 00:03:03,440
would say what they're trying to
understand or apply AI to is a 

60
00:03:03,440 --> 00:03:06,880
secondary or derivative set of 
data silos, right? 

61
00:03:06,960 --> 00:03:10,600
And, and you're right, you know,
most of these groups are trying 

62
00:03:10,600 --> 00:03:13,200
to figure out their journey just
to move stuff to the cloud, 

63
00:03:13,320 --> 00:03:15,360
right? 
And, and, and we all know now 

64
00:03:15,360 --> 00:03:18,680
fundamentally is if you don't 
have a great handle on your 

65
00:03:18,680 --> 00:03:23,400
data, right, there is no concept
of, of doing something upstate 

66
00:03:23,680 --> 00:03:28,160
relatively or disproportionately
unique, leveraging AI, right? 

67
00:03:28,560 --> 00:03:32,240
You really got to understand and
embrace cloud at times, right? 

68
00:03:32,240 --> 00:03:36,880
And embracing the data journey 
before you can, I'd say, 

69
00:03:36,880 --> 00:03:39,800
uniquely apply AI to your 
business model. 

70
00:03:39,920 --> 00:03:42,080
Now we're making it, we're 
trying to make it easy and 

71
00:03:42,080 --> 00:03:44,720
showing up with, frankly, 
applications, right, and 

72
00:03:44,720 --> 00:03:47,960
workflows that are natively, you
know, embedded with AI. 

73
00:03:47,960 --> 00:03:49,200
So you don't have to worry about
that. 

74
00:03:49,360 --> 00:03:52,280
Lots changed over the last 
decade, but I still think that, 

75
00:03:52,320 --> 00:03:55,920
you know, we're, we're still in 
the understanding managing the 

76
00:03:55,920 --> 00:03:59,720
data side of the AI equation and
not just we're not quite there 

77
00:03:59,720 --> 00:04:02,400
ready to just go Sprint and 
apply this to any enterprise. 

78
00:04:02,480 --> 00:04:05,280
Yeah, yeah. 
Now that point about not being 

79
00:04:05,280 --> 00:04:08,560
able to skip a step for some 
enterprises, they also have to 

80
00:04:08,560 --> 00:04:11,360
add a step, I would argue. 
And what I'm thinking about 

81
00:04:11,360 --> 00:04:14,320
there is sort of those that are 
in the regulated industries and 

82
00:04:14,320 --> 00:04:16,480
you work with a lot of those 
sort of companies. 

83
00:04:16,480 --> 00:04:19,920
What, if anything, is special 
about deploying AI in those 

84
00:04:19,920 --> 00:04:23,040
heavily regulated fields? 
Well, for us, let's talk about 

85
00:04:23,320 --> 00:04:26,320
what we're doing with the public
sector really the you know, 

86
00:04:26,320 --> 00:04:30,320
let's say the Department of 
Defense, you know, a lot of our 

87
00:04:30,320 --> 00:04:34,000
original architecture of AI Ware
was readily deployed the full 

88
00:04:34,000 --> 00:04:37,920
stack in commercial cloud, 
right, deploy ready and 

89
00:04:37,920 --> 00:04:41,320
deployable on, you know, Azure, 
AWS and Google cloud. 

90
00:04:41,480 --> 00:04:46,120
When you transition into network
isolated environments or secure 

91
00:04:46,120 --> 00:04:49,640
environments, some areas are, 
are completely air gap airtight,

92
00:04:49,840 --> 00:04:51,680
right? 
There is no outbound calls, 

93
00:04:51,680 --> 00:04:54,400
nothing. 
So you you don't have the luxury

94
00:04:54,440 --> 00:04:59,080
of invoking a third party 
cognitive model or a large 

95
00:04:59,080 --> 00:05:00,600
language model or multi modality
model. 

96
00:05:00,760 --> 00:05:04,360
Everything has to be in effect, 
containerized, right, deployable

97
00:05:04,440 --> 00:05:07,120
and accessible within that 
secure environment. 

98
00:05:07,160 --> 00:05:10,400
So you don't, you don't always 
have the ability to go use the 

99
00:05:10,400 --> 00:05:14,200
latest AI model from AWS, right?
You don't have the luxury of 

100
00:05:14,200 --> 00:05:17,680
using the latest greatest open 
AI model available, right? 

101
00:05:17,680 --> 00:05:19,840
Let's think about your product 
road map for a second. 

102
00:05:19,840 --> 00:05:22,280
And then you don't have to give 
her any special secrets or 

103
00:05:22,280 --> 00:05:25,000
anything. 
But how do you keep that sort of

104
00:05:25,000 --> 00:05:28,200
evolving and moving quickly? 
Like how do you go out and 

105
00:05:28,200 --> 00:05:32,680
identify a particular process or
part of a workflow that's really

106
00:05:32,680 --> 00:05:36,080
ripe for automation that you can
kind of really go to the 

107
00:05:36,080 --> 00:05:38,520
customer, go like, hey, we're 
actually addressing this pain 

108
00:05:38,520 --> 00:05:40,400
point now. 
It's a great question. 

109
00:05:40,440 --> 00:05:44,160
I think one luxury we have is 
most of our usage of quote UN 

110
00:05:44,160 --> 00:05:49,320
quote, our AI solutions is still
traditional applications, right?

111
00:05:49,360 --> 00:05:50,680
Primarily web-based 
applications. 

112
00:05:50,680 --> 00:05:53,360
So meaning tactile. 
Ryan, you, you are using your 

113
00:05:53,360 --> 00:05:56,280
keyboard at times you're using 
your mouse and other things and 

114
00:05:56,280 --> 00:05:59,360
you're using our AI solution. 
I'll use an example. 

115
00:05:59,600 --> 00:06:03,040
One of our applications is 
called Track for Public Safety. 

116
00:06:03,040 --> 00:06:05,240
In effect, it helps them sift 
through, you know, 10s of 

117
00:06:05,240 --> 00:06:08,520
thousands of hours to find 
objects or people of interest 

118
00:06:08,680 --> 00:06:11,160
right across disparate video 
assets. 

119
00:06:11,320 --> 00:06:14,400
So there's an operator that's 
used using that software and 

120
00:06:14,400 --> 00:06:17,360
yes, the back end models, AI 
models are making it more 

121
00:06:17,360 --> 00:06:20,920
efficient, right, to identify 
those people of interest, 

122
00:06:20,920 --> 00:06:23,360
etcetera. 
But we're, but, but working with

123
00:06:23,360 --> 00:06:25,360
the customer, we're tracking 
everything. 

124
00:06:25,600 --> 00:06:29,200
And so now over time, we really 
could see what, where, where's 

125
00:06:29,200 --> 00:06:32,240
their bottlenecks, right? 
Where is Ryan running into 

126
00:06:32,440 --> 00:06:34,640
deficiencies in, in the 
application layer? 

127
00:06:34,880 --> 00:06:36,880
Is it fundamentally something 
wrong? 

128
00:06:36,880 --> 00:06:39,400
And here's a big one. 
Is it wrong with the AI model or

129
00:06:39,400 --> 00:06:42,720
is it wrong with the application
layer that's in pivoting, right,

130
00:06:42,720 --> 00:06:44,800
Your best mass utilization of of
the solution. 

131
00:06:45,000 --> 00:06:48,520
So tracking that and and what I 
love about, and I'll throw in 

132
00:06:48,520 --> 00:06:50,600
the, you know, the agentic 
opportunity of how we're 

133
00:06:50,600 --> 00:06:55,080
leveraging it is, is having that
not generic copilot, if you want

134
00:06:55,080 --> 00:06:58,960
to call it that, but that fully 
integrated agent, one to north 

135
00:06:58,960 --> 00:07:02,400
number of agents that is helping
not just do something radically 

136
00:07:02,400 --> 00:07:05,440
new, but helping them get the 
most value out of the 

137
00:07:05,440 --> 00:07:07,400
application they already have. 
Right? 

138
00:07:07,440 --> 00:07:09,840
And I saw it. 
So for us when we're looking at 

139
00:07:09,920 --> 00:07:12,000
is obviously we monitor 
everything. 

140
00:07:12,040 --> 00:07:13,920
How are you using it as your 
product? 

141
00:07:13,920 --> 00:07:16,240
Where are you running traps? 
Where's their gaps, delays, you 

142
00:07:16,240 --> 00:07:20,280
know, in use of the application,
but then having AI models learn 

143
00:07:20,280 --> 00:07:23,320
what you're doing and then be 
able to to the goal here is to 

144
00:07:23,320 --> 00:07:27,880
greatly extend and amplify your 
capabilities, Ryan, And there's 

145
00:07:27,880 --> 00:07:30,480
a lot of data right there. 
If hopefully you've, if you've 

146
00:07:30,480 --> 00:07:33,600
architected your software in our
mind correctly, you better be 

147
00:07:33,600 --> 00:07:37,200
tracking everything you do 
because then I can reinforce, 

148
00:07:37,320 --> 00:07:40,400
right? 
Either fix the application or 

149
00:07:40,440 --> 00:07:42,720
fix the AI models to improve the
solution. 

150
00:07:42,840 --> 00:07:46,040
Yeah, that sounds like a good 
example of where if someone 

151
00:07:46,040 --> 00:07:49,680
isn't willing to do that 
tracking, they're kind of, you 

152
00:07:49,680 --> 00:07:52,880
know, it sounds like it could be
a fairly common mistake and how 

153
00:07:53,120 --> 00:07:56,200
a given enterprise is 
integrating AI into their 

154
00:07:56,200 --> 00:07:59,200
workflows like 1 is. 
Do you agree with that? 

155
00:07:59,480 --> 00:08:02,000
And then two other other sort of
common mistakes that you come 

156
00:08:02,000 --> 00:08:04,720
across where you really want to,
will you try to tell them, hey, 

157
00:08:04,720 --> 00:08:07,360
you shouldn't do that or you 
wish you had the, the space and 

158
00:08:07,360 --> 00:08:10,480
the leverage to, to get them to 
avoid a certain mistake? 

159
00:08:10,640 --> 00:08:13,880
You know, I think we're, we're, 
we're jump we're envisioning a 

160
00:08:13,920 --> 00:08:18,160
kind of dynamic UI application 
experience right here in the 

161
00:08:18,160 --> 00:08:19,720
short term. 
So I'll give you an example. 

162
00:08:20,000 --> 00:08:22,480
You know, if I take one of our 
big media customers like an 

163
00:08:22,480 --> 00:08:26,240
iHeartMedia and they have, you 
know, about 5000 users that use 

164
00:08:26,240 --> 00:08:30,000
our software and it's, it's in 
it's the AI models help them 

165
00:08:30,080 --> 00:08:33,320
understand or bring structure to
a lot of the audio. 

166
00:08:33,520 --> 00:08:37,400
But still it's it's a person 
using I'll call legacy Boolean 

167
00:08:37,400 --> 00:08:42,440
search for example to say find 
me all audio mentions across all

168
00:08:42,440 --> 00:08:47,040
of you know I hearts corpus of 
of FM programming that talks 

169
00:08:47,040 --> 00:08:50,360
about Tiger Woods and Nike right
the relationship hypothetically.

170
00:08:51,000 --> 00:08:53,680
Yeah. 
That that application allows 

171
00:08:53,680 --> 00:08:56,280
you. 
It creates structure and even 

172
00:08:56,280 --> 00:08:58,560
though there's a lot more 
horsepower behind it, it creates

173
00:08:58,560 --> 00:09:01,880
structure so that user doesn't 
make a mistake and the and the 

174
00:09:01,880 --> 00:09:05,080
result that the output from the 
query is very structured. 

175
00:09:05,200 --> 00:09:08,040
Now obviously with natural 
language, obviously I can wrap 

176
00:09:08,040 --> 00:09:11,440
that and integrate a full like 
organic large language model 

177
00:09:11,440 --> 00:09:14,520
natural language around that. 
And the problem now is, are you 

178
00:09:14,520 --> 00:09:17,160
learning effectively? 
If, if it's such an organic 

179
00:09:17,160 --> 00:09:21,600
experience from every user has a
different input function using 

180
00:09:21,600 --> 00:09:25,640
natural language, I think the 
resolution on understanding how 

181
00:09:25,760 --> 00:09:28,520
you need to make modifications 
to the workflow over time is 

182
00:09:28,520 --> 00:09:32,120
going to be impaired unless 
you're really thinking about how

183
00:09:32,120 --> 00:09:35,240
you're going to collect data. 
If this shifts from structured, 

184
00:09:35,240 --> 00:09:38,800
albeit limited tactile interface
and interfacing with an 

185
00:09:38,800 --> 00:09:42,800
application, what we classically
think about it as an application

186
00:09:42,960 --> 00:09:46,400
to a completely dynamic UI using
natural language. 

187
00:09:46,600 --> 00:09:49,800
So I think there's going to be a
lot of companies are going to be

188
00:09:49,800 --> 00:09:52,560
sort of shell shocked of like, 
you know, I don't know, I don't 

189
00:09:52,560 --> 00:09:53,640
know where I don't know where 
it's going. 

190
00:09:53,640 --> 00:09:56,600
I don't know what, I don't know 
what the consumer or end user 

191
00:09:56,600 --> 00:09:59,120
habits truly are. 
Because now it's, it's so 

192
00:09:59,120 --> 00:10:02,400
organic what Ryan, how you're 
communicating to the application

193
00:10:02,400 --> 00:10:05,680
versus what I'm doing, it's 
going to be hard to to normalize

194
00:10:05,680 --> 00:10:06,760
that. 
Does that make sense? 

195
00:10:07,040 --> 00:10:09,160
It does. 
And then that brings me to sort 

196
00:10:09,160 --> 00:10:12,720
of a skills or upskilling sort 
of thought because you know, 

197
00:10:12,720 --> 00:10:14,800
like I've heard a lot of 
discussions about people 

198
00:10:14,800 --> 00:10:18,320
worrying, OK, but the the 
juniors or tomorrow's juniors 

199
00:10:18,320 --> 00:10:22,520
are not going to get the manual 
skills or the ability to compare

200
00:10:22,520 --> 00:10:25,920
between what was and what is now
possible and why that is all the

201
00:10:25,920 --> 00:10:27,680
case. 
Like I've absorbed all of that. 

202
00:10:27,880 --> 00:10:30,120
But then you're raising a sort 
of similar but slightly 

203
00:10:30,120 --> 00:10:34,360
different point, which is that 
if you don't like, if you, if 

204
00:10:34,480 --> 00:10:37,840
too much of your own thinking 
gets automated now, like your 

205
00:10:37,840 --> 00:10:40,800
ability to improve the product 
later might disappear. 

206
00:10:40,800 --> 00:10:43,560
Then we might not realize how 
lucky we are that we've got a 

207
00:10:43,560 --> 00:10:46,400
before and after data set in our
own minds. 

208
00:10:46,400 --> 00:10:49,880
And then all these next round of
people might be a bit stuck. 

209
00:10:49,960 --> 00:10:52,600
And so that raises some really 
important skill questions. 

210
00:10:52,600 --> 00:10:54,000
I don't know what the answer is,
but. 

211
00:10:54,520 --> 00:10:55,840
We can spend forever on this 
one. 

212
00:10:55,840 --> 00:10:57,800
I mean, we're living this change
right now. 

213
00:10:57,800 --> 00:10:59,560
I think it's it what, what is 
it? 

214
00:10:59,760 --> 00:11:03,600
What I'm seeing is leveraging 
both different types of people 

215
00:11:03,600 --> 00:11:05,160
coming into this equation, 
right? 

216
00:11:05,400 --> 00:11:09,720
There's you and I that say have 
been predominantly educated 

217
00:11:09,720 --> 00:11:12,760
through more structured 
applications, workflows. 

218
00:11:12,920 --> 00:11:15,240
And then you have people who 
have no concept there. 

219
00:11:15,600 --> 00:11:18,040
I'll say no inhibitions to walk 
in and saying this is 

220
00:11:18,040 --> 00:11:20,040
ridiculous. 
I should be able to openly 

221
00:11:20,040 --> 00:11:22,800
communicate with my data, right,
using natural language. 

222
00:11:23,000 --> 00:11:25,600
I think the magic is in the 
combination, right? 

223
00:11:25,840 --> 00:11:29,000
So I think, you know, case in 
point, you hire kids, you know, 

224
00:11:29,000 --> 00:11:31,520
over the summer that all went 
back to college, you know, and 

225
00:11:31,520 --> 00:11:33,600
they, and let's say they're 
fresh and new into the marketing

226
00:11:33,600 --> 00:11:36,640
department and they're walking 
in with just great expertise, 

227
00:11:36,640 --> 00:11:39,880
almost organic relationship with
a large language model. 

228
00:11:39,960 --> 00:11:42,280
And you're seeing like these 
clashes, you know, where with 

229
00:11:42,280 --> 00:11:45,000
people who are experts in 
copywriting and, and so, and, 

230
00:11:45,120 --> 00:11:47,640
and where the magic happens is 
when they come together, right. 

231
00:11:47,640 --> 00:11:51,280
So to your point is, I think 
you're going to see quickly. 

232
00:11:51,280 --> 00:11:54,200
Well, I think we all, I mean, 
everybody needs to be educated 

233
00:11:54,200 --> 00:11:55,680
on, on these technologies, 
right? 

234
00:11:55,680 --> 00:11:58,280
If you're, if you're showing up 
with great expertise on writing,

235
00:11:58,280 --> 00:12:02,680
marketing, copyright, you better
accelerate and educate yourself 

236
00:12:02,680 --> 00:12:06,080
quickly because you're going to 
be, you are going to run into 

237
00:12:06,400 --> 00:12:09,280
that brilliant student who 
organically was born into the 

238
00:12:09,280 --> 00:12:11,280
space, right? 
And it, and it's doing it. 

239
00:12:11,280 --> 00:12:14,320
So I think that it's going to be
a combination of both that that 

240
00:12:14,320 --> 00:12:16,720
you're going to see the magic 
happen here that I don't think 

241
00:12:16,720 --> 00:12:18,000
it's just going to be a 0 sum 
game. 

242
00:12:18,000 --> 00:12:21,680
I do think it's going to be a 
force multiplier for everybody, 

243
00:12:21,680 --> 00:12:24,040
but education is critical across
the board. 

244
00:12:24,080 --> 00:12:25,840
Is it a prerequisite no matter 
what? 

245
00:12:25,920 --> 00:12:29,560
Yeah, We touched a bit on ROI 
earlier in the conversation, you

246
00:12:29,560 --> 00:12:31,480
know, and maybe the metrics are 
going to be different for 

247
00:12:31,480 --> 00:12:33,280
everyone. 
You know, I've got like 

248
00:12:33,280 --> 00:12:34,920
obviously there are efficiencies
for me. 

249
00:12:34,920 --> 00:12:37,200
The real dividends come when you
get to the space where you can 

250
00:12:37,400 --> 00:12:39,600
innovate and just build 
completely new things. 

251
00:12:39,680 --> 00:12:43,000
Like the efficiencies are going 
to become the table stakes is my

252
00:12:43,000 --> 00:12:45,960
personal point. 
Do you have any views on what 

253
00:12:46,320 --> 00:12:50,680
some useful new metrics for ROI 
around using your products or 

254
00:12:50,680 --> 00:12:53,320
other AI tools that you think 
people listening need to think 

255
00:12:53,320 --> 00:12:53,880
about? 
Yep. 

256
00:12:53,960 --> 00:12:56,640
I mean, I think those are very 
custom, right. 

257
00:12:56,840 --> 00:13:00,040
And I think you got to be very, 
they're very subjective relative

258
00:13:00,040 --> 00:13:03,280
to what your use cases are. 
And I think that's where you're 

259
00:13:03,280 --> 00:13:06,880
going to see major 
differentiations or companies 

260
00:13:07,480 --> 00:13:11,000
who, who are that level of under
of right of understanding and 

261
00:13:11,000 --> 00:13:14,160
invoking and leveraging AI are 
going to see the huge dividends 

262
00:13:14,200 --> 00:13:16,760
for us, for example. 
And I, I, again, I have to speak

263
00:13:16,760 --> 00:13:20,680
in context to the type of 
solution that we're offering is 

264
00:13:20,680 --> 00:13:24,040
if I say that the US Tennis 
Association, right, what we're 

265
00:13:24,040 --> 00:13:26,320
doing for them is, you know, 
we're, we're both in the cloud, 

266
00:13:26,320 --> 00:13:28,920
but on site. 
So we actually have a local AI 

267
00:13:28,920 --> 00:13:31,840
where where we're actually 
processing everything that 

268
00:13:31,840 --> 00:13:34,800
they're capturing, not just what
you and I see that's broadcast 

269
00:13:34,800 --> 00:13:38,720
over the air, but every raw feed
of every camera, every voice, 

270
00:13:38,720 --> 00:13:42,200
it's every mic is all coming in 
and being indexed in real time. 

271
00:13:42,560 --> 00:13:46,920
And so the ROI for them, Ryan is
speed to get this content out, 

272
00:13:46,920 --> 00:13:50,960
which translates into dollars 
more ad dollars, right? 

273
00:13:50,960 --> 00:13:54,120
And more obviously I'll say 
supporting their sponsors 

274
00:13:54,320 --> 00:13:57,840
quickly. 
We can now truly identify and 

275
00:13:57,840 --> 00:14:00,920
document and have hard metrics 
on performance in ROI. 

276
00:14:00,960 --> 00:14:03,480
If I, if I take that same 
technology stack and I go to 

277
00:14:03,480 --> 00:14:07,520
CNBC, another customer, their 
ROI metrics are radically 

278
00:14:07,520 --> 00:14:09,680
different, right? 
Same, same technology. 

279
00:14:09,680 --> 00:14:12,120
That's, that's, that's creating 
that opportunity. 

280
00:14:12,120 --> 00:14:15,080
That's how you apply that to 
your unique business in your our

281
00:14:15,080 --> 00:14:17,640
unique business processes where 
I think you're going to see 

282
00:14:17,640 --> 00:14:22,120
those X type of returns for new 
revenue generation or ROI 

283
00:14:22,120 --> 00:14:23,720
improvement. 
Yeah, I love that. 

284
00:14:23,720 --> 00:14:25,880
Like the one thing I was 
thinking of, as you said, that 

285
00:14:25,880 --> 00:14:29,560
is just allow so much deeper fan
engagement now, which is what 

286
00:14:29,800 --> 00:14:33,520
drives up your audience, allows 
them to bring in better, bigger 

287
00:14:33,520 --> 00:14:35,120
sponsorship deals next year and 
so on. 

288
00:14:35,120 --> 00:14:38,520
Because the fans have gone from 
having to watch the three games 

289
00:14:38,520 --> 00:14:40,920
on network TV 20 years ago and 
that's it. 

290
00:14:40,920 --> 00:14:44,360
Or you pay your 500 bucks to go 
out to the stadium versus now 

291
00:14:44,920 --> 00:14:47,560
anyone can experience every 
single match in multiple 

292
00:14:47,560 --> 00:14:51,080
different ways that was just 
never possible a generation ago.

293
00:14:51,400 --> 00:14:52,880
Here's something fascinating 
about that. 

294
00:14:53,160 --> 00:14:56,200
It's anecdotal, but back in the 
day when we should just shoot 

295
00:14:56,240 --> 00:14:59,880
everything on film, there was 
always a ratio of because it was

296
00:14:59,880 --> 00:15:04,040
so expensive, how much could you
shoot versus what was actually 

297
00:15:04,040 --> 00:15:05,520
edited down to the final 
product? 

298
00:15:06,360 --> 00:15:09,360
Let's say the coverage back in 
the day was 3 to 1. 

299
00:15:09,520 --> 00:15:12,320
Seriously, you shoot 3 hours, 
you edit it down. 

300
00:15:12,400 --> 00:15:14,360
They would dump that footage, 
they wouldn't use it. 

301
00:15:14,360 --> 00:15:17,400
They're like there's no value. 
And now if they want it because 

302
00:15:17,400 --> 00:15:20,280
they can, we can actually afford
to store this and we can mine 

303
00:15:20,280 --> 00:15:22,760
it. 
Leveraging AI, we like to say is

304
00:15:22,760 --> 00:15:25,640
don't ever throw away your ore. 
You don't know really where the 

305
00:15:25,640 --> 00:15:27,760
gold or diamonds are going to 
be, you know, sort of 

306
00:15:27,760 --> 00:15:29,080
materialized or processed 
through. 

307
00:15:29,600 --> 00:15:31,640
Yeah, I mean, it's just wild 
when you think about how many 

308
00:15:31,640 --> 00:15:35,440
silent films or early TV shows 
that, like, they just exist as 

309
00:15:35,440 --> 00:15:37,720
Wikipedia pages now. 
We don't even have the original 

310
00:15:37,720 --> 00:15:39,840
material for those reasons you 
described, right? 

311
00:15:40,280 --> 00:15:44,160
We never have to do that again. 
Now, maybe let me get, like, 

312
00:15:44,440 --> 00:15:46,400
positively selfish with you for 
a second. 

313
00:15:46,600 --> 00:15:50,240
Like, I love hearing about 
customer stories that, you know,

314
00:15:50,240 --> 00:15:52,560
for example, you were surprised 
by where a customer has figured 

315
00:15:52,560 --> 00:15:54,720
out some use case of your 
technology and you never 

316
00:15:54,720 --> 00:15:56,080
imagined they would do it that 
way. 

317
00:15:56,080 --> 00:15:57,400
Well. 
I think this was a little bit 

318
00:15:57,400 --> 00:15:59,880
older, but I just think, you 
know, this is the kind of, you 

319
00:15:59,880 --> 00:16:01,600
know, magic that can happen when
you can. 

320
00:16:01,680 --> 00:16:04,600
You know, you got two parties 
that just fully jump into it. 

321
00:16:04,920 --> 00:16:07,920
It's a little bit older of a 
deal we did, but stats perform 

322
00:16:09,000 --> 00:16:12,080
is, is a client and partner. 
When you think about sports, you

323
00:16:12,080 --> 00:16:14,160
know, I'm stuck in the sports 
vein because, you know, the US 

324
00:16:14,160 --> 00:16:16,760
Open's happening. 
But you know, as we know, every 

325
00:16:16,760 --> 00:16:19,680
time somebody swings a bat, 
throws a ball, catches 

326
00:16:19,680 --> 00:16:22,280
something, there's a there's a 
piece of metadata that's 

327
00:16:22,280 --> 00:16:24,600
created, right? 
How fast was that, you know, 

328
00:16:24,600 --> 00:16:26,960
pass thrown, how many yards, 
etcetera. 

329
00:16:27,080 --> 00:16:29,240
It runs every sport application 
we like. 

330
00:16:29,240 --> 00:16:31,320
We're obsessed with metrics and 
data, right? 

331
00:16:31,320 --> 00:16:34,600
And sports obviously is married 
to those metrics. 

332
00:16:34,920 --> 00:16:37,120
There's two companies that 
dominate that, right? 

333
00:16:37,120 --> 00:16:39,320
So they pull it together and 
they sell it and it runs every 

334
00:16:39,320 --> 00:16:42,160
sports betting app, Sports Radar
and Stats Perform. 

335
00:16:42,320 --> 00:16:45,200
OK, so stats perform. 
We we kind of went to them. 

336
00:16:45,360 --> 00:16:48,600
You know, I think we met him in 
a trade show years ago and said,

337
00:16:48,640 --> 00:16:50,520
you know, you're producing all 
this data, but you but you have 

338
00:16:50,520 --> 00:16:54,120
a limitation, obviously. 
Imagine if we could create a 

339
00:16:54,120 --> 00:16:57,400
generative AI based solution 
organically where I could take 

340
00:16:57,400 --> 00:17:02,360
your raw data feed and turn it 
that stat stat line into an 

341
00:17:02,360 --> 00:17:07,359
organic play by play in 50 
different languages and and in 

342
00:17:07,560 --> 00:17:10,000
and and in like in near real 
time. 

343
00:17:10,000 --> 00:17:13,599
We were taking just static data 
and added intuition tone. 

344
00:17:13,839 --> 00:17:16,640
We just sucked in the data feed 
from like the weather and it it 

345
00:17:16,640 --> 00:17:18,560
blows your mind. 
You're like, I I can't believe 

346
00:17:18,560 --> 00:17:21,880
it, Like if we took five bits of
data and turn it into an 

347
00:17:21,880 --> 00:17:24,640
organic, multilingual right 
approach. 

348
00:17:24,800 --> 00:17:27,560
Those those things to me, like, 
you know, I can I'm actually 

349
00:17:27,560 --> 00:17:29,680
getting, you know, like kind of 
like tingly now thinking about 

350
00:17:29,680 --> 00:17:31,040
it, Did it generate tons of 
money? 

351
00:17:31,040 --> 00:17:34,760
No, but those are those are the 
kind of things that I I just 

352
00:17:34,760 --> 00:17:38,440
think that creative minds, when 
people fully embrace all the 

353
00:17:38,440 --> 00:17:41,480
latest tools here is this stuff 
that magic's going to happen and

354
00:17:41,480 --> 00:17:44,800
that you're going to see that 
whether it's in the military or 

355
00:17:44,800 --> 00:17:47,040
meeting entertainment or back 
office. 

356
00:17:47,600 --> 00:17:49,600
First thing I thought of when 
you were saying all that was LA 

357
00:17:49,600 --> 00:17:52,280
28. 
Like this is Olympics gold. 

358
00:17:52,280 --> 00:17:56,360
Like, never mind the 50 
languages in soccer, it's 200 

359
00:17:56,360 --> 00:17:58,680
languages and 200 countries and 
30 sports. 

360
00:17:58,680 --> 00:17:59,920
That's where we need to get 
stuff. 

361
00:18:00,160 --> 00:18:03,520
Now, I also love numbers, seeing
as we're on a statistics line of

362
00:18:03,520 --> 00:18:06,520
thought at the moment, and I 
recently read that you surpassed

363
00:18:06,880 --> 00:18:09,840
a trillion tokens through your 
VDR offering. 

364
00:18:10,000 --> 00:18:12,840
People often get like 
overwhelmed or confused by 

365
00:18:12,840 --> 00:18:15,920
numbers like that. 
What does that signify for the 

366
00:18:15,920 --> 00:18:18,400
company more broadly, or where 
the AI industry is? 

367
00:18:18,400 --> 00:18:22,160
Headed, you know, and you called
it, right, that's is a data 

368
00:18:22,160 --> 00:18:25,760
science, right marketing plant 
and drop that we put out there 

369
00:18:25,760 --> 00:18:28,560
on purpose, right. 
And So what that means is is 

370
00:18:28,560 --> 00:18:32,880
when we look at the amount of 
training data that was required 

371
00:18:33,080 --> 00:18:36,680
to, to build these initial 
unbelievably exciting 

372
00:18:36,680 --> 00:18:41,320
foundational transformer models,
GPT 2, etcetera, They're all 

373
00:18:41,320 --> 00:18:46,720
based upon the concept of 
parameters or a level of a level

374
00:18:46,720 --> 00:18:51,440
of training data, often from our
vernacular, defined as tokens to

375
00:18:51,440 --> 00:18:53,800
train these models. 
So think of it as, you know, how

376
00:18:53,800 --> 00:18:55,440
many books have you read you and
I, right? 

377
00:18:55,440 --> 00:18:58,480
How much is how much input is 
going into my brain before you 

378
00:18:58,480 --> 00:19:00,080
and I can sit down and have this
conversation. 

379
00:19:00,080 --> 00:19:03,640
So obviously Phase 1 real quick,
as we all know, a lot of these 

380
00:19:03,640 --> 00:19:06,480
models went out and sort of 
sucked down the the public web 

381
00:19:06,480 --> 00:19:08,320
data, right? 
Primarily text. 

382
00:19:08,400 --> 00:19:10,840
Obviously they're having to pay 
for it now there's IP laws and 

383
00:19:10,840 --> 00:19:14,120
all these things, but text was 
first, then images and now we're

384
00:19:14,120 --> 00:19:16,640
dealing with audio and video 
that are, you know, sort of the 

385
00:19:16,640 --> 00:19:18,800
latest state-of-the-art AI based
models. 

386
00:19:19,040 --> 00:19:24,240
To put it in context that that 
that 5 trillion amount of token 

387
00:19:24,520 --> 00:19:28,320
frankly was about the size of 
the data corpus that was choose 

388
00:19:28,320 --> 00:19:30,720
to create, I believe ChatGPT. 
Yeah. 

389
00:19:31,400 --> 00:19:33,760
And that was the big one that 
wowed everyone back in. 

390
00:19:33,840 --> 00:19:37,080
That was it. 
So remember the token size for 

391
00:19:37,080 --> 00:19:41,480
us is so big because when you 
talk about 1 frame or one minute

392
00:19:41,480 --> 00:19:45,600
of audio and video, it's rich, 
it's deep, it's thick, it's got 

393
00:19:45,880 --> 00:19:48,960
so much, so many dimensions, 
data dimensions and hence that's

394
00:19:48,960 --> 00:19:52,320
translates to a lot of tokens. 
So you're that we are now 

395
00:19:52,320 --> 00:19:57,320
working with the largest AI 
multimodal model companies, the 

396
00:19:57,320 --> 00:20:00,520
biggest hyper scaler names, you 
know, and we're now working with

397
00:20:00,520 --> 00:20:05,480
them creating homogeneous index 
training data at that scale, 

398
00:20:05,480 --> 00:20:07,600
helping them train these next 
generation models. 

399
00:20:07,800 --> 00:20:10,760
And that's kind of what you read
is just is, is showing people 

400
00:20:11,000 --> 00:20:15,520
the level of scale now that we 
can help mine that ore to turn 

401
00:20:15,520 --> 00:20:19,640
it into not just ultimately rich
final solution, but very rich 

402
00:20:19,760 --> 00:20:22,720
reliable training data, right 
for these to continue to power 

403
00:20:22,720 --> 00:20:25,280
these models. 
Yeah, that makes a lot of sense.

404
00:20:25,280 --> 00:20:27,680
I mean, what you, when you're 
offering a specialist product, 

405
00:20:27,880 --> 00:20:30,840
you don't want to be training on
someone's mom's Facebook posts? 

406
00:20:30,840 --> 00:20:34,960
Like you need to have the 
specialist data to make it work.

407
00:20:34,960 --> 00:20:36,800
That's matured past, I mean, you
know, the space where a lot of 

408
00:20:36,800 --> 00:20:39,200
people think like they just wait
for the new car to come out 

409
00:20:39,320 --> 00:20:43,440
right, You know, chat GBD 5 and 
obviously what they don't see is

410
00:20:43,640 --> 00:20:47,040
there's hundreds and thousands 
of iterations of tuning, micro 

411
00:20:47,040 --> 00:20:49,840
tuning behind it. 
You know it is an arms race, as 

412
00:20:49,840 --> 00:20:51,960
you know. 
Maybe one final question now and

413
00:20:51,960 --> 00:20:55,040
it's sort of a practical one. 
It's about what what would you 

414
00:20:55,040 --> 00:20:58,600
put in peoples minds if they're 
out there thinking about 

415
00:20:58,760 --> 00:21:01,160
purchasing your product or 
another AI product? 

416
00:21:01,160 --> 00:21:04,440
What are sort of some of the 
sort of smart approaches or 

417
00:21:04,440 --> 00:21:07,640
questions that they have to 
build into their diligence to 

418
00:21:07,640 --> 00:21:10,480
make sure that they're getting 
something that is actually 

419
00:21:10,480 --> 00:21:13,760
useful for them and not just the
shiny object that that they read

420
00:21:13,760 --> 00:21:16,280
about last week? 
Yeah, imagine taking a piece of 

421
00:21:16,280 --> 00:21:19,960
marketing literature and 
redacting any word that mentions

422
00:21:20,040 --> 00:21:22,320
AI, right? 
Like, why are you buying this 

423
00:21:22,320 --> 00:21:25,120
software solution? 
Why are you buying a car, right?

424
00:21:25,400 --> 00:21:27,000
Does it satisfy your needs, 
right? 

425
00:21:27,280 --> 00:21:31,040
The AI is a tool, right? 
The AI is a component in the 

426
00:21:31,040 --> 00:21:33,360
car. 
It's part of the engine, right? 

427
00:21:33,480 --> 00:21:37,120
And so, so I think people like, 
let's get back to reality is, 

428
00:21:37,160 --> 00:21:39,640
you know, if you're going to 
have to spend $100,000 on a 

429
00:21:39,640 --> 00:21:42,520
piece of software as SAS 
solution for your company, 

430
00:21:42,720 --> 00:21:43,920
right? 
Do you have a very clear 

431
00:21:43,920 --> 00:21:46,400
understanding? 
Don't get, you know, you know, 

432
00:21:46,920 --> 00:21:51,320
don't be swayed that you need to
get off of your perfectly viable

433
00:21:51,320 --> 00:21:54,800
solution for ERP because you're 
being sold that this is next 

434
00:21:54,800 --> 00:21:56,680
generation. 
It's future proof because it's 

435
00:21:56,680 --> 00:21:58,680
got AI. 
It's not future proof if it 

436
00:21:58,680 --> 00:22:00,200
doesn't integrate with what 
you're already. 

437
00:22:00,200 --> 00:22:03,880
Doing I'm obviously all in on 
you know what that AI and 

438
00:22:03,880 --> 00:22:06,000
improve almost every application
out there. 

439
00:22:06,200 --> 00:22:08,120
You know, and that should be 
that should be part of your 

440
00:22:08,120 --> 00:22:11,600
questioning, right is in 
ultimately, is it satisfying 

441
00:22:11,600 --> 00:22:14,440
your core well defined needs? 
Right. 

442
00:22:14,440 --> 00:22:16,840
And what I just said there's 
probably is do you really even 

443
00:22:16,840 --> 00:22:19,480
have your well defined needs 
documented right before you're 

444
00:22:19,480 --> 00:22:21,600
getting sold something new, 
Ryan? 

445
00:22:21,600 --> 00:22:23,480
Stillberg, thank you so much for
joining us. 

446
00:22:23,560 --> 00:22:25,200
Thank you. 
Thanks for listening. 

447
00:22:25,480 --> 00:22:28,520
Two things before you go. 
First, if you'd like to appear 

448
00:22:28,520 --> 00:22:31,600
on the show or know someone I 
should have on, you can find me 

449
00:22:31,600 --> 00:22:33,600
through Ryan Heath 
consulting.com. 

450
00:22:34,080 --> 00:22:37,240
Finally, please tell a friend or
colleague about AI Leadership 

451
00:22:37,240 --> 00:22:39,680
Lab. 
Even better, leave a short five 

452
00:22:39,680 --> 00:22:42,560
star rating or review on 
whatever platform you found this

453
00:22:42,560 --> 00:22:44,560
podcast. 
It'll help us reach more of the 

454
00:22:44,560 --> 00:22:46,640
people who'd enjoy these tips 
and tricks. 

455
00:22:46,920 --> 00:22:48,240
Thanks and talk soon.
