1
00:00:00,160 --> 00:00:03,679
Welcome back to Data Driven, the podcast that peeks into the

2
00:00:03,679 --> 00:00:07,220
rapidly evolving worlds of data science, artificial intelligence,

3
00:00:07,520 --> 00:00:11,200
and the underlying magic of data engineering. Today's guest

4
00:00:11,200 --> 00:00:14,980
is someone who's redefining the rules of the game in AI and data,

5
00:00:15,215 --> 00:00:18,975
Ina Tokarev Saale. She's the CEO and founder of

6
00:00:18,975 --> 00:00:22,515
Illumix, a company pioneering the use of generative semantic

7
00:00:22,575 --> 00:00:26,335
fabric to make organizations AI ready. We'll dig into how

8
00:00:26,335 --> 00:00:29,855
Ina's background as a frustrated data user sparked her innovative

9
00:00:29,855 --> 00:00:33,330
journey, why 80% of enterprise decisions still aren't data

10
00:00:33,330 --> 00:00:36,790
driven, and her bold vision for a future with app free workspaces

11
00:00:36,930 --> 00:00:40,690
where AI copilots handle the heavy lifting. Oh, and we're

12
00:00:40,690 --> 00:00:44,385
tackling the ultimate question. If the future is already here,

13
00:00:44,385 --> 00:00:47,985
why does it still feel so delightfully chaotic? Sit

14
00:00:47,985 --> 00:00:51,585
back, grab your favorite coffee mug, or a Maryland state flag

15
00:00:51,585 --> 00:00:54,565
one if you're feeling fancy, and let's dive in.

16
00:00:57,105 --> 00:01:00,910
Alright. Hello, and welcome back to Data Driven, the podcast where we explore the emergent

17
00:01:00,910 --> 00:01:04,750
fields of data science, artificial intelligence, and, of course, it's all made

18
00:01:04,750 --> 00:01:08,369
possible by data engineering. And with me today is my most favoritest

19
00:01:08,590 --> 00:01:12,375
data engineer in the world, Andy Leonard. How's it going, Andy? It's going well,

20
00:01:12,375 --> 00:01:15,975
Frank. It always warms my heart when you introduce me like that. Well, you are

21
00:01:15,975 --> 00:01:19,575
my most favorite data engineer. Well, that's cool. You're well, you're my

22
00:01:19,575 --> 00:01:23,095
most favoritest. I like, there's so many things. Right? Data

23
00:01:23,095 --> 00:01:26,560
scientist, developer, evangelist.

24
00:01:27,260 --> 00:01:30,240
I mean, there's all sorts of cool things that you do. Super,

25
00:01:30,780 --> 00:01:34,580
certified person. What are you up to in certifies in certification?

26
00:01:34,940 --> 00:01:38,700
12. Wow. Yeah. I'm in I'm in the

27
00:01:38,700 --> 00:01:42,125
New York City area code now. So that's good. Next

28
00:01:42,125 --> 00:01:45,885
up, the Bronx area code 718. So Wow. That's a

29
00:01:45,885 --> 00:01:49,085
big jump. Yeah. Yeah. We're we're working on we're working on it, and I'm at

30
00:01:49,085 --> 00:01:52,925
760 some odd consecutive days. I'm at the point now

31
00:01:52,925 --> 00:01:56,350
where when I post anything on about Pluralsight

32
00:01:56,729 --> 00:02:00,170
or, my number, the search or the number of

33
00:02:00,170 --> 00:02:04,009
days, Pluralsight always sends me a congratulations, Frank. Keep

34
00:02:04,009 --> 00:02:07,450
going. So, like, I'm on their radar now. So which is really

35
00:02:07,450 --> 00:02:10,894
nice. I don't know. It's super cool. Yeah. It is super cool, which reminds me

36
00:02:10,894 --> 00:02:14,375
I still have to do 2 days. But in the

37
00:02:14,375 --> 00:02:17,974
virtual green room, we were talking about coffee mugs. We

38
00:02:17,974 --> 00:02:21,735
were. And, we're we're I don't have a coffee mug with

39
00:02:21,735 --> 00:02:25,350
me today, but, there's an

40
00:02:25,350 --> 00:02:28,470
interesting anecdote from a previous show, which I think the show is live now, about

41
00:02:28,470 --> 00:02:32,230
the Maryland state flag coffee mug, which is, pretty funny.

42
00:02:32,230 --> 00:02:35,370
So today we have with us a very special guest,

43
00:02:35,990 --> 00:02:39,655
Ina Tokarav Sala. She's the CEO and founder

44
00:02:39,655 --> 00:02:43,435
of Illumix, and a pioneer

45
00:02:43,655 --> 00:02:47,415
of generative semantic fabric, which I wanna know more about that, but it

46
00:02:47,415 --> 00:02:51,195
empowers organizations with AI readiness throughout her career

47
00:02:51,255 --> 00:02:54,840
leading data products, monetization, and as a data

48
00:02:54,840 --> 00:02:58,280
stakeholder. Ina recognized the oxymoron of our

49
00:02:58,280 --> 00:03:01,580
domain. Despite huge investments in data and analytics,

50
00:03:02,120 --> 00:03:05,725
most business decisions are still not based on these data or

51
00:03:05,725 --> 00:03:08,705
insights. And when I read that, I felt that one.

52
00:03:11,485 --> 00:03:15,025
So she, she works she founded this company,

53
00:03:15,565 --> 00:03:18,945
Lumix, which is, the the byline says, get your organization

54
00:03:19,245 --> 00:03:23,000
data generative AI ready. So So welcome to the show, Ina.

55
00:03:23,140 --> 00:03:26,980
And, tell us about this. Like, because I think this is a big problem

56
00:03:26,980 --> 00:03:30,420
with generative AI. Well, first off, let's tackle the big

57
00:03:30,420 --> 00:03:34,260
one, which is the idea that despite all this money that's been

58
00:03:34,260 --> 00:03:37,965
thrown at data and analytics for at least 2 decades, probably

59
00:03:37,965 --> 00:03:41,105
longer, a lot of decisions are not data driven.

60
00:03:44,285 --> 00:03:48,045
Yeah. Fine. Can you hear me? Because

61
00:03:48,045 --> 00:03:51,105
I see a little bit Yeah. We can hear you. Okay.

62
00:03:51,849 --> 00:03:55,610
So yeah. Thank you. You're totally right. The the benchmark says

63
00:03:55,610 --> 00:03:59,230
only 20% of decision making in enterprise is based on data.

64
00:03:59,370 --> 00:04:02,970
And to me, I I have been around for a

65
00:04:02,970 --> 00:04:06,625
while. So 25 years in data analytics, and it was

66
00:04:06,625 --> 00:04:10,385
always about cloud, big data. But

67
00:04:10,385 --> 00:04:14,165
what it actually boils down to? Are you able to

68
00:04:14,305 --> 00:04:18,144
pull out whatever analysis of data you need when you have, like, question on

69
00:04:18,144 --> 00:04:21,899
hand? Not really. And this is a situation in majority

70
00:04:22,280 --> 00:04:25,960
of enterprises, right? Even if those huge data

71
00:04:25,960 --> 00:04:29,580
teams and huge investments in infrastructure and all of that.

72
00:04:29,960 --> 00:04:33,544
And to me, the biggest promise

73
00:04:33,764 --> 00:04:37,305
of of LLMs in enterprise setting is to

74
00:04:37,604 --> 00:04:41,225
to bring the contextual and relevant data

75
00:04:41,365 --> 00:04:44,505
to the stakeholders in need.

76
00:04:44,725 --> 00:04:48,085
Right? In this experience which is impromptu which

77
00:04:48,085 --> 00:04:51,520
means it's improvised, it's governed and hallucination free, it's

78
00:04:52,460 --> 00:04:56,220
transparent. So I I would totally love have to

79
00:04:56,220 --> 00:04:59,840
have this experience where I'm in my Slack or Teams, right, and

80
00:05:00,060 --> 00:05:03,520
I've been able to to chat with my data copilot

81
00:05:03,995 --> 00:05:07,694
and ask a question and get the answer I can base decision happen.

82
00:05:07,835 --> 00:05:11,615
Right? Not just an answer. I should be reverse engineering

83
00:05:11,754 --> 00:05:13,615
with, you know, bunch of people.

84
00:05:15,354 --> 00:05:19,100
Interesting. Interesting. But I don't think that I think that

85
00:05:19,100 --> 00:05:22,860
the companies, they

86
00:05:22,860 --> 00:05:26,060
they they they throw a lot of data. They store a lot of data. They

87
00:05:26,060 --> 00:05:28,970
analyze a lot of data. But a lot of at the end of the day,

88
00:05:29,164 --> 00:05:32,865
not all decisions, but a lot of decisions are not based on just the direct

89
00:05:32,925 --> 00:05:36,764
decision of the data. They're based on quite frankly a lot

90
00:05:36,764 --> 00:05:40,384
of it's particularly the higher the, higher the

91
00:05:40,444 --> 00:05:43,830
level. Sometimes it's based on what's good for the person, not

92
00:05:43,990 --> 00:05:47,129
necessarily the organization or the business, let alone the customer.

93
00:05:48,069 --> 00:05:51,909
Do you think what are your thoughts on that? I'm familiar with the saying,

94
00:05:51,909 --> 00:05:55,449
if you touch your data long enough, it will confess. That's

95
00:05:55,909 --> 00:05:59,555
right. It goes exactly to the domain.

96
00:05:59,555 --> 00:06:02,935
So I guess you can you can massage the results

97
00:06:03,075 --> 00:06:06,755
right? But, secondhandly, when an

98
00:06:06,755 --> 00:06:09,955
employee comes to me with suggestion with a business plan with,

99
00:06:10,729 --> 00:06:14,570
you know some project I always ask like what's the ROI like what's

100
00:06:14,570 --> 00:06:18,169
it going to be to spend and what's the impact on on you know

101
00:06:18,169 --> 00:06:21,870
other activities and and what it's going to be on expense of

102
00:06:22,009 --> 00:06:25,625
so having numbers having data to you

103
00:06:25,625 --> 00:06:29,065
know to the basic decision or to bring to your boss is

104
00:06:29,065 --> 00:06:32,825
always has been a struggle and it's still struggle today so I

105
00:06:32,825 --> 00:06:36,590
think it overweights maybe some you know,

106
00:06:36,950 --> 00:06:40,550
reluctance to have open data for all just for the

107
00:06:40,550 --> 00:06:44,330
sake of of being able to to have specific context on it.

108
00:06:45,830 --> 00:06:49,655
Interesting. That that is very interesting. And, you know, that I

109
00:06:49,655 --> 00:06:53,414
think that's been the the purpose of a lot of

110
00:06:53,414 --> 00:06:57,035
data driven activities in in corporations globally

111
00:06:57,895 --> 00:07:01,115
is, you know, and for a very long time is how do you convert

112
00:07:01,735 --> 00:07:05,000
data in its raw natural form into

113
00:07:05,380 --> 00:07:08,680
information? Mhmm. And, you know, and and

114
00:07:09,380 --> 00:07:13,060
defining information as, something I

115
00:07:13,060 --> 00:07:16,515
can glance at and know, you know,

116
00:07:16,515 --> 00:07:19,975
almost instantly how my enterprise is performing.

117
00:07:20,595 --> 00:07:24,195
And that was kind of my opening line 20 years ago when I

118
00:07:24,195 --> 00:07:27,590
started in data warehousing is to go talk

119
00:07:27,650 --> 00:07:30,550
to a decision maker, CIO, CEO,

120
00:07:31,410 --> 00:07:35,110
and, you know, try and do a very small, project,

121
00:07:35,250 --> 00:07:39,030
a phase 0. And just ask them that, how do you know?

122
00:07:39,345 --> 00:07:43,045
And the surprising answer, yeah, even then it was surprising,

123
00:07:43,905 --> 00:07:47,045
was, you know, something along the lines of, well,

124
00:07:47,665 --> 00:07:51,125
people email, information to

125
00:07:51,880 --> 00:07:55,580
a lady out front or a secretary assistant guy out front,

126
00:07:55,880 --> 00:07:59,720
and he or she compiles it and puts it into this summary,

127
00:07:59,720 --> 00:08:02,700
and then they tell me. And so, you know, 1 PM

128
00:08:03,400 --> 00:08:06,985
every day or, you know, Monday on 1 PM. I know how we

129
00:08:06,985 --> 00:08:10,825
did last week. Something like that. It's very

130
00:08:10,825 --> 00:08:14,665
manual processes. So does

131
00:08:14,665 --> 00:08:18,400
does Illumix, address that? The

132
00:08:18,400 --> 00:08:22,240
manual part? Yeah. Yeah. Totally. So

133
00:08:22,240 --> 00:08:25,840
I don't think reports will go anywhere, but I think we'll

134
00:08:25,840 --> 00:08:29,360
have, you know, at least 3 types of

135
00:08:29,360 --> 00:08:33,205
experience with data. So I do I do believe in

136
00:08:33,205 --> 00:08:37,044
application free future where you have a

137
00:08:37,044 --> 00:08:40,804
question or a task and then you have a launcher and you

138
00:08:40,804 --> 00:08:43,945
just, you know, articulate whatever request you have.

139
00:08:44,300 --> 00:08:48,139
And in the background whatever applications, workloads, and data have

140
00:08:48,139 --> 00:08:51,660
been engaged with each other to to basically come up with the

141
00:08:51,660 --> 00:08:55,500
results. Right? So I do believe in this future. Right? So this is

142
00:08:55,500 --> 00:08:59,055
the ultimate. Right? But I think we will have this intermediate

143
00:08:59,755 --> 00:09:03,535
stage where we'll have a lot of copilots or

144
00:09:03,995 --> 00:09:07,835
assisted insights in, in the context of

145
00:09:07,835 --> 00:09:11,529
applications you're already using. So using your CRM systems, you will have

146
00:09:11,529 --> 00:09:14,990
all kind of insights, suggestions, you know, data driven,

147
00:09:15,850 --> 00:09:19,290
actions which which might come up with the system in your

148
00:09:19,290 --> 00:09:22,990
workflow inside your context. Right? And you might have to have

149
00:09:23,315 --> 00:09:27,154
this pure experience when you do go to analytic systems like BI

150
00:09:27,154 --> 00:09:30,935
or something else where you do have your static dashboards,

151
00:09:31,795 --> 00:09:35,475
day after day, same way that I go to, you know, to to my

152
00:09:35,475 --> 00:09:39,269
CRM dashboards and see how pipeline is going and all of that. So I do

153
00:09:39,269 --> 00:09:42,389
not them need to them to change. Right? I don't want to go to some

154
00:09:42,389 --> 00:09:45,829
chatbot and and ask again and again the same question, like, what's the pipeline

155
00:09:45,829 --> 00:09:49,670
conversion today? Right? I do want to have those static dashboards where I just,

156
00:09:49,670 --> 00:09:53,295
you know, sneak peek and see if everything in line and

157
00:09:53,295 --> 00:09:57,055
we we in the benchmark. So those three types of experiences, I

158
00:09:57,055 --> 00:10:00,895
do not think they're going to to evaporate in

159
00:10:00,895 --> 00:10:04,735
the future. Right now, we are mostly bound to the last type of

160
00:10:04,735 --> 00:10:08,430
experience of being in the closed garden of our BI tools,

161
00:10:08,510 --> 00:10:12,190
like this 3 modeled analytic experience and then we'll have this

162
00:10:12,190 --> 00:10:15,790
phase where we do have embedded experience. Majority of the companies are

163
00:10:15,790 --> 00:10:19,390
already suggesting some kind of improvements in the

164
00:10:19,390 --> 00:10:22,855
space, some better, some halfway, let's

165
00:10:22,855 --> 00:10:26,214
say. And and the ultimate goal is to to have this

166
00:10:26,214 --> 00:10:29,755
launcher when for for majority of ad hoc

167
00:10:29,975 --> 00:10:33,115
task of questions, you will have this improvised experience.

168
00:10:33,800 --> 00:10:36,860
So a follow-up on that. You mentioned Copilot, and,

169
00:10:38,199 --> 00:10:41,720
Microsoft has been the company that I've heard using that term most

170
00:10:41,720 --> 00:10:45,240
often for some sort of digital assistance. It

171
00:10:45,480 --> 00:10:48,920
to me, outsider looking in, although I I use the

172
00:10:48,920 --> 00:10:52,435
tools, it it seems to have been a quantum leap,

173
00:10:53,135 --> 00:10:56,815
this year in that technology. It just seems like last year, they were

174
00:10:56,815 --> 00:11:00,495
talking about things that it might help with, and I've seen

175
00:11:00,495 --> 00:11:04,260
all sorts of examples of this. But have you seen that? Has that been

176
00:11:04,260 --> 00:11:07,700
your experience that in the last 12 months, these type of

177
00:11:07,700 --> 00:11:11,240
assistants have just, you know, taken a giant step forward?

178
00:11:11,860 --> 00:11:15,540
Mhmm. I will address this question together with the previous one, like, how

179
00:11:15,540 --> 00:11:19,285
Illumax is is positioned in in this context. So I

180
00:11:19,285 --> 00:11:22,985
do see many projects in the companies

181
00:11:23,365 --> 00:11:26,665
which, and mainly, they're providing

182
00:11:26,884 --> 00:11:30,665
copilots, for call centers or support centers

183
00:11:31,350 --> 00:11:34,410
and mainly based on document summarization.

184
00:11:35,269 --> 00:11:38,329
Right? So document summary is more,

185
00:11:39,670 --> 00:11:43,350
lightweight and and risk averse use

186
00:11:43,350 --> 00:11:46,765
of LLM technology where I can actually go and check the document

187
00:11:46,765 --> 00:11:50,605
itself based on the resource. Right? So it's kind of and documents

188
00:11:50,605 --> 00:11:54,205
are already articulated with lots of context in

189
00:11:54,205 --> 00:11:58,045
business language. So it's kind of low hanging fruit and majority

190
00:11:58,045 --> 00:12:01,505
of the companies go to the direction including, Microsoft.

191
00:12:02,079 --> 00:12:05,380
Where Elamax goes Elamax actually,

192
00:12:05,760 --> 00:12:08,420
tackles the market which is less,

193
00:12:09,600 --> 00:12:13,120
less digested, the market of structured data. So you mentioned you

194
00:12:13,120 --> 00:12:16,685
started your career in warehouse and, so warehouses,

195
00:12:16,905 --> 00:12:20,585
databases, data lakes, business applications such as supply

196
00:12:20,585 --> 00:12:24,425
chain, ARP, CRM, and all of that. All of that

197
00:12:24,425 --> 00:12:27,970
con defined as structured data space. And despite the

198
00:12:27,970 --> 00:12:31,810
name, it couldn't be less structured than it is at the

199
00:12:31,810 --> 00:12:35,569
moment. Right? So you have If it is structured, it's not structured

200
00:12:35,569 --> 00:12:37,514
the way you need it. Yeah. Exactly. So the nay namings are not meaningful, like

201
00:12:37,514 --> 00:12:40,975
abbreviations, frank table, or for like abbreviations,

202
00:12:41,074 --> 00:12:44,574
the, frank table or and this

203
00:12:45,035 --> 00:12:48,555
transformation or alias. Right? So all those weird names especially under

204
00:12:48,555 --> 00:12:52,154
SAP systems. I love that and and no

205
00:12:52,154 --> 00:12:55,830
single source of truth. Right? In documents, you might have versions, but you do

206
00:12:55,830 --> 00:12:59,670
still have some alignment to single source of truth. In data, you

207
00:12:59,670 --> 00:13:03,350
can have many definitions even in the same

208
00:13:03,350 --> 00:13:06,950
data source. And the thing is, if you put semantic

209
00:13:06,950 --> 00:13:10,755
models like semantic search on top of them and it works by proximity,

210
00:13:11,375 --> 00:13:15,215
you might have hallucinations and random answers every time you engage

211
00:13:15,215 --> 00:13:18,654
with the tool. So this this is where we chose with

212
00:13:18,654 --> 00:13:21,890
Illumix to to tackle the problem as,

213
00:13:22,430 --> 00:13:25,490
basically, defining as a 3 step approach.

214
00:13:25,870 --> 00:13:29,390
Right? The first step is getting data AI

215
00:13:29,390 --> 00:13:33,230
ready. So there is no yeah. There is

216
00:13:33,230 --> 00:13:36,885
no way of using generative I or AI analytics in general

217
00:13:36,885 --> 00:13:40,645
if you do not have other data. But for analytics, which is

218
00:13:40,645 --> 00:13:44,325
served to you as BI dashboard, it's actually feasible to do

219
00:13:44,325 --> 00:13:47,945
manual data massaging. Right? Well, fun. Yeah.

220
00:13:48,325 --> 00:13:51,960
Yeah. That's fun. That's near and dear to my heart as a as a data

221
00:13:51,960 --> 00:13:55,660
engineer, data quality. Because

222
00:13:56,040 --> 00:13:59,720
you can have the, you know, the fastest, best presentation, the

223
00:13:59,720 --> 00:14:03,400
slickest graphics, and it could be totally lying to

224
00:14:03,400 --> 00:14:07,115
you. And back, you know, even from the days of of

225
00:14:07,115 --> 00:14:10,715
data warehousing all the way through today's semantic models and

226
00:14:10,715 --> 00:14:14,175
dashboards, it's a the the quality

227
00:14:14,235 --> 00:14:17,215
of the data store you're reporting against,

228
00:14:17,750 --> 00:14:21,510
That that data quality, if you were to measure it, you know, there's a number

229
00:14:21,510 --> 00:14:25,270
of ways to do it. But it's well north of

230
00:14:25,270 --> 00:14:29,029
99% of that. And people see that, and they go, wow.

231
00:14:29,029 --> 00:14:32,615
That that's super good. And it's like, no. No. It didn't. You can't do

232
00:14:32,615 --> 00:14:35,675
predictive analytics off of something that's 99%

233
00:14:36,615 --> 00:14:40,375
because that that 1% of bad data or

234
00:14:40,375 --> 00:14:43,995
incorrect data or duplicate data will skew your results.

235
00:14:44,454 --> 00:14:48,170
And what often, you know, the the layperson doesn't understand

236
00:14:48,950 --> 00:14:52,570
is that if it lies to you and tells you you're gonna make a $1,000,000,000,

237
00:14:53,510 --> 00:14:57,030
that's just as bad as it telling you you're only gonna make a

238
00:14:57,030 --> 00:15:00,625
$1,000,000 if the if the truth is you're gonna you're at about 25,000,000.

239
00:15:01,005 --> 00:15:04,845
That's your real projection if you were to follow that line out and do the

240
00:15:04,845 --> 00:15:08,524
extrapolation, you know, properly. And you can make

241
00:15:08,524 --> 00:15:12,144
bad decisions with an overestimation just as easily,

242
00:15:12,430 --> 00:15:16,110
maybe more so than if it's an underestimation. Yeah.

243
00:15:16,110 --> 00:15:19,810
Exactly. So this goes to to, to the ground truth of

244
00:15:20,190 --> 00:15:23,790
your results as good as your data is. And you cannot

245
00:15:23,790 --> 00:15:27,204
trust, simple semantic search

246
00:15:27,264 --> 00:15:30,964
to solve all these problems for you. And

247
00:15:31,264 --> 00:15:35,024
so for us, the baseline, the first use

248
00:15:35,024 --> 00:15:38,870
case is to get data AI ready or generative AI ready And we

249
00:15:38,870 --> 00:15:42,630
do use generative AI for that from day 1. We actually generated company

250
00:15:42,630 --> 00:15:46,390
from 2021. Yeah. It's funny to say now. It it was very hard

251
00:15:46,390 --> 00:15:50,089
to explain to our investors back then what it actually means.

252
00:15:51,195 --> 00:15:54,155
Yeah. You know, I I get it. I mean, if you build on a crooked

253
00:15:54,155 --> 00:15:57,915
foundation, you you can't get anything straight, you know,

254
00:15:57,915 --> 00:16:01,515
out of that. So that makes perfect sense to me. And it and,

255
00:16:01,755 --> 00:16:05,455
please correct me if I'm mischaracterizing, the work that Illumix

256
00:16:05,595 --> 00:16:08,310
does. But is it automated,

257
00:16:09,730 --> 00:16:13,090
AI automated, data quality? Is that really what you're

258
00:16:13,090 --> 00:16:16,610
after? So, basically, we automated full

259
00:16:16,610 --> 00:16:20,450
stack of LLM deployment for structured data, and it takes the

260
00:16:20,450 --> 00:16:24,015
AI readiness part. AI readiness, which means we have automated

261
00:16:24,075 --> 00:16:27,755
reconciliation, labeling, sensitivity tagging Okay.

262
00:16:27,995 --> 00:16:31,535
Like lots of lots of data preparation which is automated.

263
00:16:32,154 --> 00:16:35,730
Gartner actually named us as a call vendor for that lately. We have

264
00:16:35,730 --> 00:16:39,570
this layer of a context automation. Right? So so any

265
00:16:39,570 --> 00:16:43,250
LLM, any semantic model needs context and this context and reasoning

266
00:16:43,250 --> 00:16:46,390
usually rebuild by data scientists. To me, it's controversial

267
00:16:46,610 --> 00:16:50,005
because, you know we had data modelers which didn't

268
00:16:50,005 --> 00:16:53,785
understand business logic and now we have data scientists who do not necessarily

269
00:16:54,245 --> 00:16:57,845
fully understand business logic and the model into black

270
00:16:57,845 --> 00:17:01,190
box experience of context. Right? So ElamX

271
00:17:01,649 --> 00:17:05,410
reverses process. We actually automate context and we wrap it

272
00:17:05,410 --> 00:17:08,849
up in augmented governance workflow so business people or

273
00:17:08,849 --> 00:17:12,630
governance folks can actually certify it. So it's auto generated

274
00:17:12,690 --> 00:17:16,495
context for LLMs but certifiable by humans. We do

275
00:17:16,495 --> 00:17:19,935
believe that we need to bring human to the loop, right, to to certify

276
00:17:19,935 --> 00:17:23,295
it. Yeah. And the last I love I'm sorry. I have

277
00:17:23,295 --> 00:17:26,975
interrupted you, like, 3 times now, and I apologize. I haven't met 2. I

278
00:17:26,975 --> 00:17:30,670
thought you paused. So finish please finish your thought.

279
00:17:31,130 --> 00:17:34,810
No. No. I'm saying, like, 3 parts. So you already did data governance and the

280
00:17:34,810 --> 00:17:38,510
actual alarm deployment because you need to interact with the whole thing, and the interaction

281
00:17:38,570 --> 00:17:42,250
to have to has to be explainable and transparent. You need to understand

282
00:17:42,250 --> 00:17:45,404
how, especially on structured data, you need to understand how

283
00:17:46,024 --> 00:17:49,865
the question was calculated based, sorry, how answer was

284
00:17:49,865 --> 00:17:53,705
calculated based on questions and how, data was

285
00:17:53,705 --> 00:17:57,470
actually sourced, what's the lineage, what is the governance and access

286
00:17:57,470 --> 00:18:01,309
control through search your clients. So all of that should be on the interaction layer.

287
00:18:01,309 --> 00:18:05,149
So AI readiness, governance, and the interaction layer explainability to

288
00:18:05,149 --> 00:18:08,255
the end user. Absolutely. Okay.

289
00:18:09,515 --> 00:18:13,115
Thanks. And I do apologize again for the

290
00:18:13,115 --> 00:18:16,955
interruption. So my my characterization of it as something that's just

291
00:18:16,955 --> 00:18:20,715
data quality is is way low. There's a little bit of overlap between

292
00:18:20,715 --> 00:18:24,480
data quality and what you're describing. You're talking about taking this into

293
00:18:24,480 --> 00:18:28,180
that next level that is specific to, generative

294
00:18:28,240 --> 00:18:31,940
AI and perhaps other, you know, AI related,

295
00:18:32,000 --> 00:18:35,760
AI adjacent technologies, machine learning leaps to mind and stuff like

296
00:18:35,760 --> 00:18:39,195
that. But your the tagging, the categorizing,

297
00:18:39,815 --> 00:18:43,515
and all of the things you're describing there, that is next level.

298
00:18:43,895 --> 00:18:47,735
And it's very interesting to me that you're

299
00:18:47,735 --> 00:18:51,115
using AI to get data ready for AI.

300
00:18:51,430 --> 00:18:55,270
That's an interesting combination. Mhmm. It makes sense, though. Right?

301
00:18:55,270 --> 00:18:58,950
You can kinda scale out human capability with AI. I

302
00:18:58,950 --> 00:19:02,790
think that's you you kind of alluded that with Newman in the loop. Right? Like,

303
00:19:02,790 --> 00:19:06,445
I think I think where you were kinda going with that, again, don't wanna speak

304
00:19:06,445 --> 00:19:10,125
for you, but it's like the idea that AI isn't gonna replace

305
00:19:10,125 --> 00:19:13,805
humans. It's just gonna make humans more productive. Yeah.

306
00:19:13,805 --> 00:19:17,340
For sure. Augment us because frankly speaking, no one

307
00:19:17,340 --> 00:19:20,700
wants to to model data, you know, as their

308
00:19:20,700 --> 00:19:24,380
career. We want to solve problems. Right? And to solve

309
00:19:24,380 --> 00:19:28,155
problems, we we have to to understand what the problems are

310
00:19:28,475 --> 00:19:32,235
And letting AI to surface the problems as alerts and for us

311
00:19:32,235 --> 00:19:35,915
to to resolve them as conflicts takes, you

312
00:19:35,915 --> 00:19:39,535
know, 1% to 10% of the time that it should take,

313
00:19:40,155 --> 00:19:43,940
where we are busy, you know, wrangling data still. And, you know,

314
00:19:43,940 --> 00:19:47,559
it's sad to some extent because data is growing and we cannot keep up.

315
00:19:48,180 --> 00:19:51,300
No. That's a good point. Even if even if there are people out there and

316
00:19:51,300 --> 00:19:54,820
some of our listeners may really do like modeling data. Right? But, you

317
00:19:54,820 --> 00:19:58,515
know, Dow, they can model about 10 times the amount of data or maybe

318
00:19:58,515 --> 00:20:01,895
a 100 times more. Right? And then ultimately, the expectation of

319
00:20:02,075 --> 00:20:05,715
what a, you know, what a person

320
00:20:05,715 --> 00:20:09,235
can do in a set period of time is gonna go up just

321
00:20:09,235 --> 00:20:13,060
because I I I think I think you're on to something there. Plus,

322
00:20:13,060 --> 00:20:16,260
I also I would also, like, double click on the idea that you said earlier,

323
00:20:16,260 --> 00:20:19,320
which I think was very intriguing, was this notion of

324
00:20:20,180 --> 00:20:22,980
a lot of the apps that you use would kind of fade away. You just

325
00:20:22,980 --> 00:20:26,795
have this virtual assistant. You know, I I think back to

326
00:20:26,875 --> 00:20:30,635
there's a number of scenes in, you know, Star Trek The Next Generation where they

327
00:20:30,635 --> 00:20:33,995
have a conversation with the computer. Right? Mhmm. You know, you they

328
00:20:33,995 --> 00:20:37,435
don't they use the computer. They get stuff done. There's no

329
00:20:37,435 --> 00:20:40,830
Microsoft Word. There's no PowerPoint. Right? Like, there's no, like, it's

330
00:20:40,830 --> 00:20:44,429
just the the there is no application. The application is kind of invisible. It

331
00:20:44,429 --> 00:20:48,030
becomes the computer. And I think that's a very

332
00:20:48,030 --> 00:20:51,549
intriguing kind of way. And if you had told me that a year ago, I

333
00:20:51,549 --> 00:20:55,309
would have been very skeptical. Now I look at it, I'm like, I

334
00:20:55,309 --> 00:20:58,184
mean, it's it's it's almost inevitable.

335
00:20:58,965 --> 00:21:02,725
Yeah. Yeah. I agree with you. Futures here,

336
00:21:02,725 --> 00:21:06,085
it's not evenly distributed as people say. So I

337
00:21:06,085 --> 00:21:09,625
guess, you know, when you're attending conferences in Bay Area,

338
00:21:09,924 --> 00:21:12,980
it's already it's already here. It happens. Right

339
00:21:14,240 --> 00:21:18,080
and when you go to let's say Europe we

340
00:21:18,080 --> 00:21:21,860
even just say you know just say a EU act in

341
00:21:21,920 --> 00:21:25,360
Europe is is ramping up so it's all about

342
00:21:25,360 --> 00:21:28,965
controls and and this is great So I do not think that regulation and

343
00:21:28,965 --> 00:21:32,805
innovation, actually, jeopardize each other. I think

344
00:21:32,805 --> 00:21:36,485
they should go hand by hand and, that's where I see

345
00:21:36,485 --> 00:21:40,325
industry is going. So so East Coast approach, majority of our customers

346
00:21:40,325 --> 00:21:42,585
are coming from East Coast US, Pharma,

347
00:21:44,300 --> 00:21:48,060
financial services, insurance, highly regulated data

348
00:21:48,060 --> 00:21:51,120
intensive companies. They have now,

349
00:21:51,740 --> 00:21:55,340
sometimes even inventing standards for generative AI

350
00:21:55,340 --> 00:21:58,945
implementations because everything is so new but companies

351
00:21:58,945 --> 00:22:02,705
want to go fast. Right? So no one wants

352
00:22:02,705 --> 00:22:06,225
to to downplay risks on one hand. On the other

353
00:22:06,225 --> 00:22:10,065
hand, everyone want to, you know, to implement generative AI

354
00:22:10,065 --> 00:22:13,760
and see the productivity cuts. It's, you know, it's evident productivity

355
00:22:13,900 --> 00:22:17,120
cuts are already here with all those co pilots summarization,

356
00:22:18,380 --> 00:22:22,220
what have you and this is where we are today. So I

357
00:22:22,220 --> 00:22:25,515
think like again Bay Area running fast

358
00:22:26,555 --> 00:22:30,235
and east is coming up with regulation. We will meet somewhere

359
00:22:30,235 --> 00:22:33,995
in between. I believe in both. Well, if you kind of,

360
00:22:33,995 --> 00:22:37,515
like, look at, like, historically, you know, when .coms first

361
00:22:37,515 --> 00:22:41,080
started, right, there were a number of, hey. Look. You know, we're gonna sell pet

362
00:22:41,080 --> 00:22:44,840
food online. Right? Like, and then it was

363
00:22:44,840 --> 00:22:48,679
like, back in the dial up days, it didn't really make a lot of

364
00:22:48,679 --> 00:22:51,580
sense. So it would just be easier for me to go to the store.

365
00:22:52,040 --> 00:22:55,795
Whereas now, I mean, if you think about ecommerce, obviously,

366
00:22:55,795 --> 00:22:59,495
Amazon is the £2,000,000,000 gorilla in the

367
00:22:59,875 --> 00:23:03,395
room. I like, do I really

368
00:23:03,395 --> 00:23:06,755
wanna think about, you know, dealing particularly as we get into the holiday season, do

369
00:23:06,755 --> 00:23:10,200
I really wanna deal with the traffic at the mall or the store when I

370
00:23:10,200 --> 00:23:13,820
can just click on something, either have, you know, groceries delivered

371
00:23:13,960 --> 00:23:17,640
or, you know, I'm I'm okay waiting 2 days for

372
00:23:17,640 --> 00:23:20,220
something to come up if I don't have to deal with them all.

373
00:23:21,080 --> 00:23:24,825
Yeah. Totally. What's what's the difference between Black Friday

374
00:23:24,825 --> 00:23:28,505
and Cyber Monday? No. It's not. Right? Like not really. Yeah.

375
00:23:28,505 --> 00:23:32,264
Yeah. So it's like Not anymore. I remember Yeah. You

376
00:23:32,264 --> 00:23:35,884
know? So we're recording this just before Black Friday. And,

377
00:23:36,860 --> 00:23:40,700
you know, this whole idea of, you know, going to the store, get

378
00:23:40,700 --> 00:23:44,300
the best deals, it's like, do I really wanna deal with the

379
00:23:44,300 --> 00:23:47,980
crowd? No. Yeah. Although ironically, the name for the

380
00:23:47,980 --> 00:23:51,535
podcast came on a Black Friday, while I was

381
00:23:51,675 --> 00:23:55,515
at a Dunkin' Donuts, drinking coffee, waiting waiting

382
00:23:55,515 --> 00:23:59,275
in line actually to get so there's a I'm a Krispy Kreme

383
00:23:59,275 --> 00:24:03,115
person. So I'm Ah, okay. Yeah. So With you and

384
00:24:03,115 --> 00:24:06,960
I, right, definitely. Right here. This is before we had a Krispy Kreme

385
00:24:06,960 --> 00:24:10,799
near us. So it's I I have split sides, but yeah. Yeah.

386
00:24:10,799 --> 00:24:14,399
Jeff's JT. He's a mess. From up north. So they are

387
00:24:14,559 --> 00:24:17,919
they're Dunkin' Donuts. I've noticed this. They're Dunkin' Donuts, like, north of

388
00:24:17,919 --> 00:24:21,475
Virginia. And he's in Maryland. I'm in Virginia. Then down

389
00:24:21,475 --> 00:24:25,075
south, you rarely see a Dunkin' Donuts. I see more Dunkin' Donuts down

390
00:24:25,075 --> 00:24:28,755
south than Krispy Kreme's up north, though, for sure. Yeah. But

391
00:24:28,755 --> 00:24:32,435
I They're they're from Boston. That's why. Yeah. Oh, that's why. And then So at

392
00:24:32,435 --> 00:24:35,930
Krispy Kreme's from Atlanta. And plus, it's funny. Right? Like, so I live in

393
00:24:35,930 --> 00:24:39,770
Maryland Mhmm. Which depending on who whom you ask is either

394
00:24:39,770 --> 00:24:43,610
north or south. So that's right. That's true.

395
00:24:43,610 --> 00:24:46,970
Interesting. Interesting. We're a quarter state for sure. Yeah. That that's

396
00:24:47,370 --> 00:24:51,165
that goes safe for Virginia. But I wanted to follow-up on, you know, you've

397
00:24:51,165 --> 00:24:54,465
been we've been talking about all the cool stuff. I'm

398
00:24:55,005 --> 00:24:58,445
gonna try and say this correctly. Illumix. Is that correct? Am I getting it

399
00:24:58,445 --> 00:25:01,265
right? So Illumix name

400
00:25:01,885 --> 00:25:05,710
from Illuminating the Dark Side of Organizational Data.

401
00:25:05,710 --> 00:25:09,550
Illuminate like illuminate. Illuminate. I like that. And x x

402
00:25:09,550 --> 00:25:13,070
for the x factor. Excellent. X for the x

403
00:25:13,070 --> 00:25:16,805
factor. Yeah. What? And I'm not asking you to I'll

404
00:25:16,805 --> 00:25:20,585
just ask a question. What are the risks in in what you're doing?

405
00:25:21,125 --> 00:25:24,505
And, you know, what are the risks you're aware of and how are you addressing

406
00:25:24,645 --> 00:25:27,780
those? Yeah.

407
00:25:28,160 --> 00:25:31,140
So I think the biggest risk of 2025

408
00:25:31,920 --> 00:25:35,760
is going to be, a TCO, total cost of

409
00:25:35,760 --> 00:25:38,820
ownership. So already today,

410
00:25:39,200 --> 00:25:42,675
it's, it's very hard for organizations to to

411
00:25:42,675 --> 00:25:46,455
monitor where the generative AI tokens are spent.

412
00:25:47,075 --> 00:25:50,055
And the benchmark say that 80%

413
00:25:50,675 --> 00:25:54,055
of LLM tokens actually spend on customization

414
00:25:55,130 --> 00:25:58,730
of off the shelf models. And that's not a good news because

415
00:25:58,810 --> 00:26:02,190
which means ROI is is pretty low on on the actual

416
00:26:02,250 --> 00:26:05,310
production use of generative AI in in enterprise.

417
00:26:05,930 --> 00:26:09,605
And I think it doesn't get any better because the

418
00:26:09,605 --> 00:26:13,145
customizations techniques which are used today gains a black box

419
00:26:13,365 --> 00:26:17,205
performed by super expensive data scientists and

420
00:26:17,205 --> 00:26:20,965
they're not very scalable for data that you don't want to, you know,

421
00:26:20,965 --> 00:26:24,700
to schmooze around. I think it's cost prohibitive actually to bring data

422
00:26:24,700 --> 00:26:28,160
to AI. You need to bring AI to data. So so putting

423
00:26:28,380 --> 00:26:32,220
data in some graph structures for graph, frog, and all of that, it's to me,

424
00:26:32,220 --> 00:26:36,045
it's cost prohibitive. So this is why I think that, the Telumex

425
00:26:36,105 --> 00:26:39,705
position for 2025 is actually favorable because we bring this

426
00:26:39,705 --> 00:26:43,245
transparency. We do create this, a virtual,

427
00:26:43,625 --> 00:26:46,985
a semantic knowledge graph, which is transparent to certify, which is

428
00:26:46,985 --> 00:26:50,799
created for business people. Based on business

429
00:26:50,799 --> 00:26:54,580
logic. We do use extensively industry ontologies and so on so forth.

430
00:26:54,799 --> 00:26:58,580
And I think the the most interesting part about generative AI is

431
00:26:58,720 --> 00:27:02,425
we do not necessarily going to mimic processes that

432
00:27:02,425 --> 00:27:06,045
the humans performed. Mhmm. We're going to invent

433
00:27:06,105 --> 00:27:09,865
those processes. Right? So new new processes and new workflows. So

434
00:27:09,865 --> 00:27:13,645
right now, a generative AI is deployed like like

435
00:27:13,785 --> 00:27:17,450
analytics is deployed, which means you you have to

436
00:27:17,450 --> 00:27:21,289
label your data, check the quality, usually manually, and then

437
00:27:21,289 --> 00:27:24,809
you have to to prepare the test set which is fed

438
00:27:24,809 --> 00:27:28,649
into customization of the model and then you actually provide the

439
00:27:28,649 --> 00:27:32,195
context to on every question. So this is

440
00:27:32,195 --> 00:27:35,955
very old fashioned or, you know, 40 years old

441
00:27:35,955 --> 00:27:39,654
machine learning technique to to actually train generative

442
00:27:39,715 --> 00:27:43,554
vi. So this is why why I'm saying that, many companies are

443
00:27:43,554 --> 00:27:47,080
probably going to to mimic what Equinox does in the sense

444
00:27:47,080 --> 00:27:50,840
that you have to you have to be focused on domain

445
00:27:50,840 --> 00:27:54,200
specific knowledge, reason, ontologies, and knowledge graphs. You have

446
00:27:54,200 --> 00:27:57,560
to onboard your customers automatically via metadata because

447
00:27:57,560 --> 00:28:01,045
metadata has the factor all

448
00:28:01,045 --> 00:28:04,645
activities in organization documented for us. We're

449
00:28:04,645 --> 00:28:08,325
just under utilizing them, right? And then you bring your

450
00:28:08,325 --> 00:28:11,845
business people, your domain experts, your governance teams to the

451
00:28:11,845 --> 00:28:15,620
loop because you can simply cannot bring this business acumen,

452
00:28:16,320 --> 00:28:19,780
to, you know, to data. You have to bring data to to those people.

453
00:28:20,480 --> 00:28:24,080
That's an interesting thing because I've seen the the particularly is this this this

454
00:28:24,080 --> 00:28:27,825
statistic around 80% of the tokens are being used to

455
00:28:27,825 --> 00:28:31,585
manipulate the data. I have a microcosm example of that

456
00:28:31,585 --> 00:28:35,425
where I use AI to augment my blog post, my blog

457
00:28:35,425 --> 00:28:38,725
that I create, and I finally took

458
00:28:40,110 --> 00:28:43,870
a closer look at this because I was spending a lot more on

459
00:28:43,870 --> 00:28:47,630
the OpenAI API than I really wanted to. And I'm like, well,

460
00:28:47,630 --> 00:28:50,450
what exactly am I I'm using a product called Fabric.

461
00:28:51,710 --> 00:28:55,054
And I'm like, wait, what exactly is the source of this prompt? And I look

462
00:28:55,054 --> 00:28:58,095
at it, and I'm like, I can't. It's a lot. It's a long prompt. And

463
00:28:58,095 --> 00:29:01,615
I'm like, I really don't need that. Right? So we are gonna do a deep

464
00:29:01,615 --> 00:29:05,294
dive in a show on Fabric at some point. Not not the Fabric Andy

465
00:29:05,294 --> 00:29:08,780
works with, but there's an open source thing called fabric. There's

466
00:29:08,780 --> 00:29:12,460
a I'm sure there are lawyers right now that are doing their

467
00:29:12,460 --> 00:29:15,900
holiday shopping based on how much money they're gonna make off of this

468
00:29:15,900 --> 00:29:19,660
dispute. But, the the short of it is, like,

469
00:29:19,660 --> 00:29:22,535
I realized, like, well, no wonder why I spent so much money. I'm sending all

470
00:29:22,535 --> 00:29:26,295
of this in my prompt plus the content. So I

471
00:29:26,295 --> 00:29:29,575
actually in the verse before you joined in, Andy and I were talking, and I

472
00:29:29,575 --> 00:29:33,115
was like, I actually got a really good result based on a more optimized

473
00:29:33,255 --> 00:29:36,950
prompt. You know? And, you know, strictly speaking, it's

474
00:29:36,950 --> 00:29:39,670
not I I like your approach of bringing the AI to the data rather than

475
00:29:39,670 --> 00:29:41,929
bringing the data to the AI because that is expensive.

476
00:29:43,750 --> 00:29:47,030
You know, I I think that bringing the AI to the data will be less

477
00:29:47,030 --> 00:29:50,825
expensive. How less, I think, remains to be seen. But I like that approach,

478
00:29:50,825 --> 00:29:54,585
right? Because that's typically what we've done, you know, and we've seen

479
00:29:54,585 --> 00:29:58,125
huge upsides to that, whether it's from Hadoop bringing the

480
00:29:58,345 --> 00:30:02,024
compute to the data rather than vice versa. I like that

481
00:30:02,024 --> 00:30:05,510
approach. And it's backed by historical precedent. Right? So it's not

482
00:30:05,510 --> 00:30:09,110
completely gonna be this crazy idea. It's just a very sensible

483
00:30:09,110 --> 00:30:12,790
idea. Yeah. Yeah. I believe the future was already

484
00:30:12,790 --> 00:30:16,310
invented. Right? So it's just the inclination of technologies we already have.

485
00:30:16,310 --> 00:30:19,965
It's been healthy about it. So, we had

486
00:30:19,965 --> 00:30:23,665
machine learning practices which are very healthy like feature

487
00:30:23,725 --> 00:30:27,245
exploration, feature definitions and then we had neural net brute

488
00:30:27,245 --> 00:30:30,945
force and then majority of companies used combination of both,

489
00:30:31,210 --> 00:30:34,730
right, to to to be optimized. This is what I think what's happening with

490
00:30:34,730 --> 00:30:38,570
generative AI. So this, you know, wild west of brute

491
00:30:38,570 --> 00:30:42,090
force or great spend is going to be replaced by methods

492
00:30:42,090 --> 00:30:45,934
which have, like, this automated context filtering or pre

493
00:30:45,934 --> 00:30:49,695
processing and then use like fraction of your budget to to actually

494
00:30:49,695 --> 00:30:53,455
run the query. Yeah. I remember hearing about a lot

495
00:30:53,455 --> 00:30:57,230
of this in the late nineties. And, I worked for a company who

496
00:30:57,230 --> 00:31:01,070
was a big SAP shop. I see you have a history with SAP. Yeah. And

497
00:31:01,070 --> 00:31:04,350
this lady and and and so we were an we were the IT department. So

498
00:31:04,350 --> 00:31:08,110
we were in the basement, but the analytics team back then was in

499
00:31:08,110 --> 00:31:11,895
a closed in space inside the basement. So it was

500
00:31:11,895 --> 00:31:15,655
like even more like, you know, I was the web developer, so I didn't

501
00:31:15,655 --> 00:31:19,195
have a window, but I could see the window about 50 feet away.

502
00:31:19,415 --> 00:31:23,255
But, like, when you when when you went

503
00:31:23,255 --> 00:31:26,840
into this, like, you know, further enclosed space deeper into

504
00:31:26,840 --> 00:31:30,620
the the the the the depths of the IT department,

505
00:31:31,080 --> 00:31:34,600
there was the database team. And and and and in the back of that area

506
00:31:34,600 --> 00:31:37,900
was the analytics group. And I remember this lady telling me

507
00:31:40,674 --> 00:31:44,355
that she was working with these things called OLAP cubes. Oh, wow.

508
00:31:44,355 --> 00:31:47,794
Yeah. And I was like, what is that? And then she went on this thing

509
00:31:47,794 --> 00:31:51,510
and, you know, I'm remembering a conversation, oh my god,

510
00:31:51,510 --> 00:31:55,270
almost 30 years ago. But I just remember walking away with,

511
00:31:55,270 --> 00:31:59,110
like, that sounds either crazy because she's talking about,

512
00:31:59,110 --> 00:32:02,890
like, you know, figuring out patterns. Right? So, you know, will

513
00:32:03,765 --> 00:32:07,385
rainfall patterns in Australia affect not just the agricultural

514
00:32:07,525 --> 00:32:10,825
side of the chemical business, but also the plastics purchasing

515
00:32:10,885 --> 00:32:14,485
versus rainfall in the Amazon versus this and all of

516
00:32:14,485 --> 00:32:18,325
that? And I just remember walking away from that conversation as I as I

517
00:32:18,325 --> 00:32:22,130
as I as I leave the depths of the IT department back to my normal

518
00:32:22,130 --> 00:32:25,890
kinda, basement. Back to the regular basement from

519
00:32:25,890 --> 00:32:29,570
the sub basement. I remember thinking that is either the craziest thing I

520
00:32:29,570 --> 00:32:33,105
ever heard or the most profound thing I ever heard, which

521
00:32:33,345 --> 00:32:36,705
now with the, hindsight of time, it turns out it was the most profound

522
00:32:36,705 --> 00:32:40,325
thing. Yeah. You you can think about it as

523
00:32:40,465 --> 00:32:43,924
semantic layers of, you know, that era. Right?

524
00:32:44,225 --> 00:32:48,070
Mhmm. Right. And I think You know go ahead.

525
00:32:48,070 --> 00:32:51,590
I'm sorry. Sorry. I think it's delayed between the

526
00:32:51,830 --> 00:32:55,430
between the connection. So I think around the same time I was

527
00:32:55,430 --> 00:32:59,130
doing my bachelor and my project was about multi dimensional

528
00:32:59,350 --> 00:33:02,385
theory. So multi dimensional geometry,

529
00:33:03,565 --> 00:33:07,165
of these neural nets. So basically, you model neural nets as multi

530
00:33:07,165 --> 00:33:10,785
dimensional graph and it does operational research calculations.

531
00:33:11,405 --> 00:33:15,005
So it's exactly the same. You you model your universe in a

532
00:33:15,005 --> 00:33:18,640
graph. Back then it wasn't MATLAB. We didn't have any, you

533
00:33:18,640 --> 00:33:22,400
know, neural nets Right. Structures or graph structures and so you're

534
00:33:22,400 --> 00:33:26,179
modeling in MATLAB in this weird language,

535
00:33:26,400 --> 00:33:30,044
a graph which has a neural nets on there. And

536
00:33:30,044 --> 00:33:33,485
this is exactly like modeling all of cubes. Right? A

537
00:33:33,485 --> 00:33:36,924
multidimensional representation of your reality. Now,

538
00:33:36,924 --> 00:33:40,284
unfortunately, we have a new technologies which,

539
00:33:40,924 --> 00:33:44,700
which are semantic and context. Right? Large language

540
00:33:44,700 --> 00:33:48,460
models and graphs, which do the same thing but much

541
00:33:48,460 --> 00:33:52,300
more efficiently. Yeah. So this is amazing. Like, I

542
00:33:52,300 --> 00:33:55,915
think it goes back to what you said. You know, The future's already here. It's

543
00:33:55,915 --> 00:33:59,455
just not widely distributed yet, which I think is a William Gibson

544
00:33:59,595 --> 00:34:03,295
quote, or is it a Esther Dyson quote? I forgot.

545
00:34:04,075 --> 00:34:07,539
But it's one of those 2 kinda luminaries. Yep.

546
00:34:07,840 --> 00:34:11,440
You you said what I was going to say, you know, and it

547
00:34:11,440 --> 00:34:15,119
was, you know, more of what off of what Frank

548
00:34:15,119 --> 00:34:18,659
said is it turns out that we're just

549
00:34:18,800 --> 00:34:22,525
doing more nodal analysis and vector

550
00:34:22,585 --> 00:34:26,265
geometry as a result of that. That's it did all start

551
00:34:26,265 --> 00:34:30,025
with multidimensional and and grow from there. And

552
00:34:30,025 --> 00:34:33,085
that's where these algorithms, like nearest neighbor

553
00:34:33,705 --> 00:34:37,489
originated, was in that math. So

554
00:34:38,350 --> 00:34:41,630
Yeah. Yeah. Great minds. Exactly. Exactly.

555
00:34:41,949 --> 00:34:45,630
Alike. Exactly. Now you're

556
00:34:45,630 --> 00:34:49,469
complimenting me. Thank you. I I feel I feel better

557
00:34:49,469 --> 00:34:51,975
when smart people in the room agree with me.

558
00:34:53,095 --> 00:34:56,455
No. I'm on the right path. You know, I employ

559
00:34:56,455 --> 00:34:59,595
millennials. So so having people with experience in multidimensional

560
00:34:59,895 --> 00:35:03,335
geometry and all of cubes, it's just a miracle to me to to start

561
00:35:03,335 --> 00:35:06,960
with. You know? People now like Python, neural

562
00:35:06,960 --> 00:35:10,800
nets, we do actually, the average age in in in

563
00:35:10,800 --> 00:35:14,640
Lumex is around 35, 37, something like that. So we do

564
00:35:14,640 --> 00:35:18,420
have like also pretty experienced folks, you know, but new talent,

565
00:35:18,640 --> 00:35:22,455
they, they they're not familiar with all all of that.

566
00:35:22,535 --> 00:35:25,655
And I think it's actually a disadvantage because,

567
00:35:26,135 --> 00:35:29,775
when when you do know different patterns in architecture Yeah.

568
00:35:29,895 --> 00:35:33,335
You can model them with new technology. Right? Make them more

569
00:35:33,335 --> 00:35:36,935
efficient, but you already know what works and what doesn't, and it

570
00:35:36,935 --> 00:35:40,390
helps. That yeah. That's a great point. The old

571
00:35:40,390 --> 00:35:43,610
experience, you know, the experience that we have from doing this for

572
00:35:43,750 --> 00:35:47,190
decades is that we see the patterns that have

573
00:35:47,190 --> 00:35:50,970
repeated over time, architectural patterns and design patterns. And,

574
00:35:51,110 --> 00:35:54,535
you know, and we know that they've

575
00:35:55,095 --> 00:35:58,295
I I love that how you said that. The, you know, the future's already been

576
00:35:58,295 --> 00:36:01,735
invented. We we realize that if we reapply some of these

577
00:36:01,735 --> 00:36:05,575
patterns, that there are use cases for them, not just now, but

578
00:36:05,575 --> 00:36:08,820
also in the future. So totally get you.

579
00:36:09,260 --> 00:36:10,520
Too, you know, like,

580
00:36:12,980 --> 00:36:16,260
you know, it it is painful to think that, you know, we've been in this

581
00:36:16,260 --> 00:36:20,100
industry for decades. Right? It is a little hurts a little bit. But,

582
00:36:20,100 --> 00:36:23,805
like, also, if you're listening to this, you've not been in the industry for

583
00:36:23,805 --> 00:36:27,405
decades, and you're thinking like, woah. You know, what are these what are these

584
00:36:27,405 --> 00:36:31,105
old geezers now? I would point out when I was

585
00:36:31,244 --> 00:36:34,704
a young kid in the industry and, you know,

586
00:36:35,165 --> 00:36:38,380
client server was like the new hotness. Right?

587
00:36:39,080 --> 00:36:42,620
And, you know, the whole notion of going back to,

588
00:36:43,000 --> 00:36:46,540
you know, cloud and and and and and, you know, terminal

589
00:36:47,160 --> 00:36:50,780
and an old mainframe geezer basically said to me, like, this is just

590
00:36:51,025 --> 00:36:54,705
this industry has a cycles. Right? It's like the fashion industry. This goes in

591
00:36:54,705 --> 00:36:58,465
style. This goes out style. And it was like, I had that moment

592
00:36:58,465 --> 00:37:02,225
of, like, wait. I think he's on to something, but he's just an old geezer,

593
00:37:02,225 --> 00:37:06,040
so I won't listen. So, you know, so so

594
00:37:06,040 --> 00:37:08,700
if you are a young buck, like, or,

595
00:37:11,480 --> 00:37:14,920
buck is a male deer, right? What would be a Yes. A doe. A young

596
00:37:14,920 --> 00:37:18,675
doe. So if you're a young buck or a young doe, I grew up

597
00:37:18,675 --> 00:37:22,355
in New York City. So all of this wildlife thing is brand new. I'm here

598
00:37:22,355 --> 00:37:26,195
for you. I'm here for you, Frank. So, you

599
00:37:26,195 --> 00:37:29,955
know, listen to, like, some of the things that these, you know, more

600
00:37:29,955 --> 00:37:32,935
experienced colleagues will say. Yeah. You know,

601
00:37:34,000 --> 00:37:36,720
if you don't believe it right away, just put it on the shelf in your

602
00:37:36,720 --> 00:37:40,420
mind because you're gonna need it later. It'll come up at some point.

603
00:37:40,480 --> 00:37:44,080
And it's like, if you look at kind of, you know, everybody ran to the

604
00:37:44,080 --> 00:37:47,600
cloud. Right? And cloud is effectively like a

605
00:37:47,600 --> 00:37:51,065
mainframe effectively. Right? The same philosophy. Right? Centralized

606
00:37:51,205 --> 00:37:54,965
computing somewhere else. Right? And then your browsers become

607
00:37:54,965 --> 00:37:58,185
the terminals, terminals with fancy graphics, but terminals nonetheless.

608
00:37:58,965 --> 00:38:02,770
Now I think you're gonna start seeing it kind of we're about due for a

609
00:38:02,770 --> 00:38:06,549
seismic shift backwards, right, as people kinda move

610
00:38:06,690 --> 00:38:10,369
repatriate data and things like that. Particularly, I think driven by AI

611
00:38:10,369 --> 00:38:14,150
because of the cost of some of this. You know, I had this debate,

612
00:38:14,210 --> 00:38:18,015
you know, the other day. It was like, you know, if if one of these

613
00:38:18,015 --> 00:38:21,855
super clusters with, you know, a 100, 8 100,

614
00:38:21,855 --> 00:38:24,915
all of this, if it costs, say, $500,000,

615
00:38:26,494 --> 00:38:30,335
right, I could probably do the math, and that probably means

616
00:38:30,335 --> 00:38:34,099
about, you know, there's a certain break even point,

617
00:38:34,099 --> 00:38:37,859
and it's probably after about 7 or 8 fine tunings or full

618
00:38:37,859 --> 00:38:41,400
on trainings where it's just cheaper to have it. Just buy it.

619
00:38:41,460 --> 00:38:45,059
Yeah. Yeah. Yeah. Totally on that. And also, you

620
00:38:45,059 --> 00:38:48,795
know, salary skills are the most expensive part. So you

621
00:38:48,795 --> 00:38:52,635
want to spend it on your business specific problems and

622
00:38:52,635 --> 00:38:56,395
not generic problems you can solve with software. Right? So

623
00:38:56,395 --> 00:38:59,900
it's always like that. Yeah. Yeah. So,

624
00:39:00,280 --> 00:39:04,059
I do think that, basically capacity to process data

625
00:39:04,200 --> 00:39:08,040
is is going to be a challenge. Right? And this is why we

626
00:39:08,040 --> 00:39:10,940
see that, that majority of,

627
00:39:11,559 --> 00:39:14,984
of I would even say countries not

628
00:39:14,984 --> 00:39:18,345
only specific enterprises, kind of gear

629
00:39:18,345 --> 00:39:21,565
up with, with GPUs, FPGAs,

630
00:39:21,705 --> 00:39:25,224
whatever hardware you have. Right? So do you see it in

631
00:39:25,224 --> 00:39:28,780
middle east, in emirates? They they have national generative

632
00:39:28,840 --> 00:39:32,680
vi grid and they're building it for, you know, not only government companies

633
00:39:32,680 --> 00:39:35,660
but also private companies. We see the same in Europe

634
00:39:36,280 --> 00:39:40,125
and I would assume, you know, US based telcos

635
00:39:40,125 --> 00:39:43,485
are going to to provide those data centers with GPU soon

636
00:39:43,485 --> 00:39:47,005
enough, right, for, you know, for everyone to purchase as an

637
00:39:47,005 --> 00:39:50,845
alternative to the public cloud. Yes. And we'll

638
00:39:50,845 --> 00:39:54,490
see it. So this is for starters. And second one, the second part where

639
00:39:54,490 --> 00:39:58,030
you don't need, this, you know, heavy machinery,

640
00:39:58,570 --> 00:40:01,870
you might just have your variables processing

641
00:40:02,890 --> 00:40:06,570
parts of whatever generated AI on your end before sending to the cloud

642
00:40:06,570 --> 00:40:10,155
because you do not necessarily need to to process everything in a central

643
00:40:10,155 --> 00:40:13,915
manner. We basically have pretty powerful machines on

644
00:40:13,915 --> 00:40:17,435
our hands or in our hand, you know, as

645
00:40:17,435 --> 00:40:21,035
glasses as well. We can see that, and it's

646
00:40:21,035 --> 00:40:24,450
going to be part of the processing. So the processing is going to be distributed.

647
00:40:24,510 --> 00:40:28,190
You bring AI to your data, where your data is. You do

648
00:40:28,190 --> 00:40:32,030
not shift your data all the time. It's not, it's not

649
00:40:32,030 --> 00:40:35,635
cheap anymore. And we'll have this, as you mentioned,

650
00:40:35,635 --> 00:40:38,934
those central repositories of mass processing

651
00:40:39,394 --> 00:40:43,154
and those distributed powerhouses which are

652
00:40:43,154 --> 00:40:46,375
small enough to to process data on on edge.

653
00:40:47,330 --> 00:40:50,530
I think you're right. I think you're gonna see a set of data being processed

654
00:40:50,530 --> 00:40:54,210
in one place. I think it's gonna be everywhere. There's gonna be some

655
00:40:54,450 --> 00:40:58,210
and and I think that that introduces some interesting, consequences. Right?

656
00:40:58,210 --> 00:41:02,050
So my wife works in IT security, and I can immediately hear her voice in

657
00:41:02,050 --> 00:41:05,625
the back of my head. Contrary to what you think, ladies, we do

658
00:41:05,625 --> 00:41:09,465
listen. We just don't always pay attention. But

659
00:41:09,465 --> 00:41:12,045
I can hear her like, well, if compute's happening everywhere,

660
00:41:13,225 --> 00:41:16,700
gee, couldn't like that be poisoned anywhere.

661
00:41:16,700 --> 00:41:20,220
Right? I think I think that's going to be the next kind of thing. Right?

662
00:41:20,220 --> 00:41:23,119
It's and it's again, it's a pattern. Right? Advancement.

663
00:41:23,980 --> 00:41:27,500
Bad actors take advantage for that. Problem happens. And

664
00:41:27,500 --> 00:41:30,635
then then that's the new thing. Right? So it's almost like you're you're building like

665
00:41:30,635 --> 00:41:33,675
a, like a like a like a layer cake. Right? Like, you know, the cake

666
00:41:33,675 --> 00:41:37,035
goes down then the frosting. The cake is the innovation. The frosting is

667
00:41:37,035 --> 00:41:40,715
security, and then so on and so on. So Yeah. Yeah. Yeah.

668
00:41:40,715 --> 00:41:44,520
So it basically back to the semantics. What we started is

669
00:41:44,520 --> 00:41:48,220
semantic ontology as a baseline for generative AI.

670
00:41:48,840 --> 00:41:52,440
It has multiple benefits. Single source of truth, of course, has the

671
00:41:52,440 --> 00:41:56,200
benefits for accuracy. But also, if you're passing every

672
00:41:56,200 --> 00:41:59,365
question to this semantic ontology context,

673
00:41:59,985 --> 00:42:03,345
it's almost impossible to poison it because we're going to either

674
00:42:03,345 --> 00:42:07,185
match to part of your logic or Right. Right. We're going to

675
00:42:07,185 --> 00:42:10,705
miss. So it's it's another layer of security if you think about

676
00:42:10,705 --> 00:42:13,480
it. So, so yeah.

677
00:42:14,340 --> 00:42:18,020
That's an interesting point. All new. Yeah. All new ontology, all new

678
00:42:18,020 --> 00:42:21,640
semantics have governance meaning, it has

679
00:42:21,700 --> 00:42:25,320
accuracy meaning, it has also security meaning.

680
00:42:27,245 --> 00:42:30,205
And also if you want to have single source of truth you have to to

681
00:42:30,205 --> 00:42:33,965
have means to distribute it to those edge devices or

682
00:42:33,965 --> 00:42:37,565
to to bring it back to central location and without ontologies, without

683
00:42:37,565 --> 00:42:41,400
semantic layers, simply it's impossible to do that. I was gonna

684
00:42:41,400 --> 00:42:44,839
say, like, the the the infrastructure, not just the computer infrastructure, but the

685
00:42:44,839 --> 00:42:47,260
logical infrastructure to distribute this stuff,

686
00:42:48,599 --> 00:42:51,799
it's probably not a trivial problem. That's the first thing that popped in my mind.

687
00:42:51,799 --> 00:42:54,700
I was like, you know, like, oh, yeah. You're right about the distributed

688
00:42:55,955 --> 00:42:59,715
activity on this data, but, wow, what does that

689
00:42:59,715 --> 00:43:02,915
look like? What do updates look like? Like, the whole like, it's a it sounds

690
00:43:02,915 --> 00:43:04,535
like a growth industry to me.

691
00:43:07,315 --> 00:43:10,755
Definitely. Yeah. Yeah. I don't it's, it's

692
00:43:10,755 --> 00:43:14,530
what we call, engineering problem. Right? So

693
00:43:14,530 --> 00:43:17,970
creating ontology is data science or generative AI problem, but

694
00:43:17,970 --> 00:43:21,570
distributing it, maintaining it, thinking it's its engineering problem.

695
00:43:21,570 --> 00:43:25,395
Engineering problems tend to to have engineering solutions. Oh, Oh,

696
00:43:25,395 --> 00:43:27,715
that's a good point. That's a good way to look at it. I like that.

697
00:43:27,715 --> 00:43:31,475
I like that. So did you wanna do the, premade questions?

698
00:43:31,475 --> 00:43:34,355
Because we haven't we've gone a few shows without them. If you're okay with those,

699
00:43:34,355 --> 00:43:38,115
Ina, we can we can ask them. If not, that's fine

700
00:43:38,115 --> 00:43:41,800
too. Of course. Yeah. Sure. Mhmm. So they're not they're not complicated.

701
00:43:41,860 --> 00:43:45,080
They're more kinda just general questions. I pasted them in the chat.

702
00:43:46,660 --> 00:43:50,260
But the first question and and you've had a a pretty

703
00:43:50,260 --> 00:43:53,845
significant career with SAP and and before that. How'd you

704
00:43:53,845 --> 00:43:57,525
find your way into this space? Did you find data or did

705
00:43:57,525 --> 00:44:01,205
data find you? I

706
00:44:01,205 --> 00:44:04,425
found my way to data by being frustrated

707
00:44:04,645 --> 00:44:07,465
user. Right? So I started in engineering

708
00:44:08,085 --> 00:44:10,900
and it was evident to me that

709
00:44:12,000 --> 00:44:15,760
using data as engineer is not enough. You have to go to

710
00:44:15,760 --> 00:44:19,380
data management. You have to fix those things because otherwise

711
00:44:19,760 --> 00:44:22,974
I will I will going to be frustrated for the end of my life. Right?

712
00:44:22,974 --> 00:44:26,355
So I went to data management analytics to to solve the problem

713
00:44:26,895 --> 00:44:30,595
and I discovered that, as you mentioned, every experience

714
00:44:30,655 --> 00:44:33,954
has a footprint. So my experience with graphs and with

715
00:44:34,335 --> 00:44:38,049
operational research and multidimensional geometry and all of that is so

716
00:44:38,049 --> 00:44:41,349
useful for data management. And it was actually exhilarating.

717
00:44:42,770 --> 00:44:46,609
That's true. Like and I like that because, like, every experience does leave

718
00:44:46,609 --> 00:44:50,369
a footprint. Like, you know, that that's cool. I'm gonna I'm gonna pull that out

719
00:44:50,369 --> 00:44:54,165
as a special quote for the episode. That's a great quote. Yeah. So

720
00:44:54,165 --> 00:44:58,005
our next question why we do these? Yeah. Is what's your favorite part of your

721
00:44:58,005 --> 00:45:01,765
current gig? My favorite part of being a

722
00:45:01,765 --> 00:45:05,300
founder is is

723
00:45:05,300 --> 00:45:08,600
unlimited ability of experimentation,

724
00:45:09,860 --> 00:45:13,640
right? So majority of my day actually say no

725
00:45:13,780 --> 00:45:17,540
to things, not to experiment, which is which is hard, which is not fun part,

726
00:45:17,540 --> 00:45:21,085
right? But, still, we can

727
00:45:21,085 --> 00:45:24,224
make decisions and we can do

728
00:45:24,684 --> 00:45:28,305
new stuff every day. So as a founder,

729
00:45:28,365 --> 00:45:32,170
it's been very, very different than enterprise setting. And don't don't take

730
00:45:32,170 --> 00:45:35,790
me wrong. Like, SAP is a huge place of growth and had

731
00:45:36,330 --> 00:45:39,770
very, fulfilling career at SAP, you know, building

732
00:45:39,770 --> 00:45:43,070
stuff, founding p and l's, running big organizations,

733
00:45:43,370 --> 00:45:46,994
but but been able to to actually, you know,

734
00:45:46,994 --> 00:45:50,454
start anything new. And, like, right now, we have this customer

735
00:45:50,755 --> 00:45:54,434
and they want to to try Illumax on in

736
00:45:54,434 --> 00:45:58,275
parallel on the newest, you know, newest BI

737
00:45:58,275 --> 00:46:02,060
tool with semantic layer or and on the oldest

738
00:46:02,060 --> 00:46:05,520
warehouse on premise at once. I'm like, okay. Challenge accepted.

739
00:46:05,980 --> 00:46:09,740
Yeah. And next Wow. Yeah. And next day, you know, engineer

740
00:46:09,740 --> 00:46:13,280
comes with we have this academic data set and they have these benchmarks.

741
00:46:13,420 --> 00:46:16,915
Let's beat them. I'm like, yeah, let's do it. It could be cool stuff.

742
00:46:17,215 --> 00:46:20,975
Right? Lovely. So, you know, you know, it's to some extent,

743
00:46:20,975 --> 00:46:24,255
so we don't need to justify it, you know, business wise and but but in

744
00:46:24,255 --> 00:46:27,410
majority of cases, we can. Cool.

745
00:46:28,510 --> 00:46:31,950
We have a couple of complete the sentences. When I'm not working, I

746
00:46:31,950 --> 00:46:35,630
enjoy blank. I used to

747
00:46:35,630 --> 00:46:39,170
enjoy doing jogging and yoga when I'm not working.

748
00:46:39,765 --> 00:46:43,285
Right? So right now when I'm not working which means when I'm not

749
00:46:43,285 --> 00:46:47,125
traveling I just spend time with my family. Whatever

750
00:46:47,125 --> 00:46:50,585
is the plan for the weekend if it's just you know Netflixing,

751
00:46:51,365 --> 00:46:55,125
or cooking or hiking whatever is the plan I just

752
00:46:55,125 --> 00:46:58,700
join So sometimes just, you know, plan it. But spending time with my

753
00:46:58,700 --> 00:47:02,460
family has become, indulgence and I'm

754
00:47:02,460 --> 00:47:05,900
very focused on that. Cool. Very cool. Our

755
00:47:05,900 --> 00:47:09,420
next is I think the coolest thing in technology today

756
00:47:09,420 --> 00:47:13,184
is blank. I think the coolest tech is

757
00:47:13,424 --> 00:47:16,944
thing right now is not in tech. It's actually the

758
00:47:16,944 --> 00:47:20,085
pull from CEOs of companies

759
00:47:20,785 --> 00:47:24,500
for technology. This is something which didn't experience for decades.

760
00:47:24,500 --> 00:47:27,700
So we were pushing cloud and big data and machine learning and deep learning. We

761
00:47:27,700 --> 00:47:31,460
were explaining to business stakeholders why do they need that. Mhmm.

762
00:47:31,460 --> 00:47:35,140
And now, so you're all coming and saying, okay, I want to have

763
00:47:35,140 --> 00:47:38,820
chatbot experience for x y that, so just

764
00:47:38,820 --> 00:47:42,495
build it. This is actually I think this is the coolest

765
00:47:42,555 --> 00:47:46,395
part because it's kind of a removes majority of the friction that

766
00:47:46,395 --> 00:47:49,775
we had to to deploy technology in the past.

767
00:47:50,555 --> 00:47:54,340
Interesting. On our 3rd and final complete the sentence,

768
00:47:54,640 --> 00:47:58,340
I look forward to the day when I can use technology to blank.

769
00:48:00,320 --> 00:48:04,160
So many things. You know, travel has

770
00:48:04,160 --> 00:48:07,825
been so frustrating lately, and, I

771
00:48:07,825 --> 00:48:11,285
don't think what happened because it's like kind of technology goes

772
00:48:11,345 --> 00:48:15,045
forward but airline, you know, travel technology,

773
00:48:15,185 --> 00:48:18,545
hospitality technology in general, I don't feel it bridges a

774
00:48:18,545 --> 00:48:21,920
gap. So I really look forward to the

775
00:48:21,920 --> 00:48:25,460
future where I can just have this comment, this prompt

776
00:48:26,240 --> 00:48:29,540
of plan, this conference in Dallas on

777
00:48:29,599 --> 00:48:33,359
x and the system already knows all by preferences and

778
00:48:33,359 --> 00:48:36,214
just done. Oh, boy. It would be it would be fantastic.

779
00:48:38,035 --> 00:48:41,714
Yeah. That that the travel experience as I I've had to

780
00:48:41,714 --> 00:48:45,555
travel quite a bit, like, for the past, like,

781
00:48:45,555 --> 00:48:49,315
couple months, and it's just like, oh my god. Like, it never was

782
00:48:49,315 --> 00:48:52,900
great, but awful is not a word I remember. But it's post

783
00:48:52,900 --> 00:48:56,660
pandemic, I think it's gotten way worse. It's like there's just so many small things

784
00:48:56,660 --> 00:48:59,700
that you could be done a lot better. I'm I'm a 100% with you on

785
00:48:59,700 --> 00:49:02,865
that one. So true. So our our next

786
00:49:03,425 --> 00:49:06,725
question is to, ask you to share something

787
00:49:06,785 --> 00:49:08,244
different about yourself.

788
00:49:11,905 --> 00:49:15,055
Sharing something different about myself. I think I'm a controversial

789
00:49:15,750 --> 00:49:18,890
person in general. So, so some people,

790
00:49:20,870 --> 00:49:24,310
so some people agree with, you know, with the degree

791
00:49:24,310 --> 00:49:27,910
of, of living in the future. So I,

792
00:49:27,910 --> 00:49:31,305
I, you know take myself as person who is very much in the future so

793
00:49:31,305 --> 00:49:34,985
all this seed happening and I might be a little bit you know ahead because

794
00:49:34,985 --> 00:49:38,585
I see the technology being developed in my mind is already there, it's already

795
00:49:38,585 --> 00:49:42,410
used right? So and so where this is

796
00:49:42,410 --> 00:49:46,250
where I see myself controversial because you know in majority of the

797
00:49:46,250 --> 00:49:49,870
cases, then you sit over family dinner

798
00:49:50,250 --> 00:49:53,550
and say, you know, we're still paying our bills

799
00:49:53,785 --> 00:49:57,145
online when we have this notification. Right?

800
00:49:57,145 --> 00:50:00,505
So everyday technology has

801
00:50:00,505 --> 00:50:04,045
developed a lot. And when I'm speaking about this application

802
00:50:04,185 --> 00:50:07,570
free future and, you know,

803
00:50:08,350 --> 00:50:12,190
automated, x y zed. Sometimes or many

804
00:50:12,350 --> 00:50:16,030
oftentimes on everyday level, we are still not there and

805
00:50:16,030 --> 00:50:19,810
this is where people think that I'm too visionary or too

806
00:50:20,110 --> 00:50:23,625
too dreamer on that. Interesting.

807
00:50:23,925 --> 00:50:25,545
No. I'm with you on that one.

808
00:50:28,965 --> 00:50:32,085
Growing up, I was the technical person in the family. So

809
00:50:32,805 --> 00:50:36,005
Yeah. They don't they don't know what you're talking about. Right? I I I love

810
00:50:36,005 --> 00:50:39,490
how the, you know, or, you know, they all they

811
00:50:39,869 --> 00:50:42,430
all get confused until the printer breaks and then suddenly

812
00:50:43,549 --> 00:50:46,589
But you're the smartest people in the room. That's why you're the smartest person in

813
00:50:46,589 --> 00:50:50,430
the world. Alright. So where can people find out more about you and

814
00:50:50,430 --> 00:50:53,875
Illumix? I love socializing on

815
00:50:53,875 --> 00:50:57,234
LinkedIn. I don't know that many people think LinkedIn became a

816
00:50:57,234 --> 00:51:00,835
marketing tool. I still see tons of valuable

817
00:51:00,835 --> 00:51:04,440
discussions and I just absolutely love keeping in touch

818
00:51:04,440 --> 00:51:08,280
on LinkedIn and and see the latest and greatest and I also share quite a

819
00:51:08,280 --> 00:51:11,740
bit. So LinkedIn would be the the most

820
00:51:11,880 --> 00:51:15,020
straightforward way in Atokaropsala on LinkedIn.

821
00:51:15,800 --> 00:51:19,315
We do have blogs and I actually write many of

822
00:51:19,315 --> 00:51:21,335
them. So if you go to illumeg.ai/blocks,

823
00:51:23,954 --> 00:51:27,494
you will see lots of materials written on semantics,

824
00:51:27,795 --> 00:51:31,400
on ontologies, on generative AI governance. So those

825
00:51:31,400 --> 00:51:35,079
topics which are close to my heart, and we communicate quite

826
00:51:35,079 --> 00:51:38,859
frequently on that. Very cool. Very cool. Very cool. So

827
00:51:39,240 --> 00:51:42,825
so Audible is a sponsor. And if you

828
00:51:42,825 --> 00:51:46,425
would, like to take advantage of a free month of

829
00:51:46,425 --> 00:51:49,725
Audible on us, you can go to the datadrivenbook.com.

830
00:51:51,385 --> 00:51:55,225
I just tested the link. That's why I was looking over here for anyone watching

831
00:51:55,225 --> 00:51:58,750
the video. And it works. Sometimes it doesn't. And

832
00:51:59,050 --> 00:52:02,730
we ask, our guests, do you have, do first, do you

833
00:52:02,730 --> 00:52:06,490
listen to audio books? And if so, can you recommend 1? If

834
00:52:06,490 --> 00:52:09,550
you don't listen to audio books, just a a good book.

835
00:52:11,065 --> 00:52:14,665
I do listen to audiobooks. I also podcast, more

836
00:52:14,665 --> 00:52:18,105
frequently recently. I I'm not sure this book is

837
00:52:18,105 --> 00:52:21,945
already on Audible, but, if not, it's going to be

838
00:52:21,945 --> 00:52:25,385
in Audible soon enough. So it's Nexus by Yuval Noah

839
00:52:25,385 --> 00:52:28,799
Harari. It is audible. I have it in the library already.

840
00:52:28,799 --> 00:52:32,020
Yeah. Amazing. So it speaks about the truth

841
00:52:32,720 --> 00:52:36,559
in the age of generative AI. Right? Interesting.

842
00:52:36,559 --> 00:52:40,185
What's the truth? What's the ground truth? And I was

843
00:52:40,185 --> 00:52:43,885
actually in the lunch party in SoHo, New York, you know when Yuval

844
00:52:44,025 --> 00:52:47,165
was speaking about you know how how technology

845
00:52:47,785 --> 00:52:51,165
and what we see right now is not very different from what we experience

846
00:52:51,225 --> 00:52:54,920
in you know middle age like when when Gothenburg

847
00:52:55,060 --> 00:52:58,820
and printing was was a new thing and like what was

848
00:52:58,820 --> 00:53:01,960
printed actually was you know rumors

849
00:53:02,340 --> 00:53:06,035
and juicy stuff rather than scientific books and this

850
00:53:06,035 --> 00:53:09,875
is where what we see right now in, you know, in chatbots and internet, on

851
00:53:09,875 --> 00:53:13,555
social overall. So it's it's interesting parallels that he's

852
00:53:13,555 --> 00:53:17,160
taking about what's what truth is in generative

853
00:53:17,220 --> 00:53:20,740
AI age where what truth were was, like, 20 years

854
00:53:20,740 --> 00:53:24,500
ago or even, like, 500 years ago. Yeah.

855
00:53:24,500 --> 00:53:28,340
We're the we're the same species with the same problems and the same drama

856
00:53:28,340 --> 00:53:32,035
and the same drivers. Like, it's just our tools have changed, whether

857
00:53:32,035 --> 00:53:35,775
it's a printing press or, you

858
00:53:36,115 --> 00:53:39,875
know, celebrity gossip or whatever or fake news

859
00:53:39,875 --> 00:53:43,315
or anything like that. Plus, I also think the, you know, there's an old phrase

860
00:53:43,315 --> 00:53:46,720
like who watches the watchers. Right? Like Mhmm. Who decides what's

861
00:53:46,720 --> 00:53:50,480
misinformation and who decides what's true? I think. I think

862
00:53:50,480 --> 00:53:54,079
because misinformation could be, you know, there there's

863
00:53:54,079 --> 00:53:57,220
a image of me robbing a bank. Right? Like, you know?

864
00:53:57,885 --> 00:54:01,645
Mhmm. Mhmm. I thought, Frank, I thought when the US

865
00:54:01,645 --> 00:54:05,485
Marshals put you into the witness protection program, they said

866
00:54:05,485 --> 00:54:09,325
we couldn't bring up you robbing a bank any any longer.

867
00:54:09,485 --> 00:54:13,140
Misinformation. You gotta be careful because, like, one of the things I I wanted the

868
00:54:13,140 --> 00:54:15,859
flow was so good. I didn't wanna interrupt it. But, like, one of the things

869
00:54:15,859 --> 00:54:18,680
was I was experimenting with fine tuning an LLM locally.

870
00:54:19,380 --> 00:54:23,220
Mhmm. And I'm basically trained it on information about my blog. My blog's

871
00:54:23,220 --> 00:54:26,680
been around since 1995. Right? Or my site has been around since 1995.

872
00:54:28,434 --> 00:54:31,954
One of them hallucinated this really great origin story for my

873
00:54:31,954 --> 00:54:35,555
website. It was awesome. It was awesome. I'm like, I like that

874
00:54:35,555 --> 00:54:39,335
better. So basically, it said that Always. Always.

875
00:54:39,474 --> 00:54:42,835
It was really good. It was basically that Frank's World started as a

876
00:54:42,835 --> 00:54:46,560
show, a kids TV show in the nineties on

877
00:54:46,560 --> 00:54:50,319
the BBC or channel 4. I forget. Like one

878
00:54:50,319 --> 00:54:53,920
of the big British channels. And it was about a talking

879
00:54:53,920 --> 00:54:57,615
trash can named Frank that would teach kids about the importance

880
00:54:57,915 --> 00:55:01,535
of, recycling. That's my favorite part.

881
00:55:01,755 --> 00:55:04,715
And it was and it was the best part was that it was it was

882
00:55:04,715 --> 00:55:08,395
the first professional project of the guys who did Sean the sheep and Wallace and

883
00:55:08,395 --> 00:55:12,020
Gromit. Yeah. And I'm like so I

884
00:55:12,020 --> 00:55:14,840
I I pinged the guy I worked with. Has this ever been a show?

885
00:55:15,940 --> 00:55:18,740
Because no. Not that I ever heard of. And I looked over it. I couldn't

886
00:55:18,740 --> 00:55:22,500
find it. But and then what I did was as an experiment, I fed

887
00:55:22,500 --> 00:55:26,315
that that whole paragraph that it came up with into

888
00:55:26,315 --> 00:55:30,154
notebook l m. Mhmm. Notebook l m

889
00:55:30,154 --> 00:55:33,835
took that and ran with it. There's, like, a 20

890
00:55:33,835 --> 00:55:37,135
minute audio, and it is the funniest thing because it basically

891
00:55:38,200 --> 00:55:41,960
talks about the early environmental movement. They said it was the Britain's

892
00:55:41,960 --> 00:55:45,560
answer to, Captain Planet. Like, they made up all the

893
00:55:45,560 --> 00:55:49,320
stuff. And now it's documented. So now someone is going

894
00:55:49,320 --> 00:55:53,000
to pulling to pull some information. And if you have Right now it's out there.

895
00:55:53,160 --> 00:55:56,974
Right. And I guess to your point earlier about Lumix, like, if you start

896
00:55:56,974 --> 00:56:00,815
building a crooked foundation, right, like, that eventually as

897
00:56:00,815 --> 00:56:04,654
it moves on, it's gonna so, I mean, who knows, like, couple of years

898
00:56:04,654 --> 00:56:08,099
from now, like, Wikipedia may say, like, there might be a

899
00:56:08,099 --> 00:56:11,940
Wikipedia article about this TV show didn't exist. We're talking about it. We're feeding

900
00:56:11,940 --> 00:56:14,920
the machine. That's fascinating.

901
00:56:15,780 --> 00:56:18,900
Yeah. And it was a so a little bit on the books. I have to

902
00:56:18,900 --> 00:56:22,255
mention it, like, in a couple of sentences. So, in US

903
00:56:22,555 --> 00:56:26,315
a legal entity actually is a citizen. It

904
00:56:26,315 --> 00:56:30,015
has social number. Right. So, technically machines

905
00:56:30,155 --> 00:56:33,695
can create legal entities. They can vote, they can,

906
00:56:34,070 --> 00:56:37,830
you know, they can create information and this information is,

907
00:56:37,830 --> 00:56:41,590
you know, created with social number, with identifiers. So it's actually real

908
00:56:41,590 --> 00:56:44,730
information. It's not fake news. It's created by social number.

909
00:56:45,590 --> 00:56:49,365
And so this is how you create, like, this new truth. Right?

910
00:56:49,365 --> 00:56:53,125
And, and how do you control that? So it's an interesting aspect of what's,

911
00:56:53,605 --> 00:56:56,585
what even is defined as ground truth.

912
00:56:57,285 --> 00:57:00,325
That's true. Everybody needs to define it. I think that's gonna be the question of

913
00:57:00,325 --> 00:57:03,990
the 20 That's a big deal. Mhmm. Yeah. Well,

914
00:57:03,990 --> 00:57:06,310
awesome. It's been great. We wanna be respectful of your time. This has been an

915
00:57:06,310 --> 00:57:10,070
awesome show. Yeah. We'll let Bailey finish the show. And

916
00:57:10,070 --> 00:57:13,830
that's a wrap for today's episode of data driven. A massive

917
00:57:13,830 --> 00:57:17,234
thank you to Ina Tokarev Saleh for joining us and sharing her

918
00:57:17,234 --> 00:57:20,935
fascinating insights into the world of generative AI, semantic

919
00:57:20,994 --> 00:57:24,295
fabrics, and the ever evolving relationship between humans,

920
00:57:24,515 --> 00:57:27,890
data, and decision making. If you're as inspired as we

921
00:57:27,890 --> 00:57:31,730
are, be sure to check out IllumiX and follow INA on LinkedIn for

922
00:57:31,730 --> 00:57:35,570
more thought leadership in the AI space. As always, thank

923
00:57:35,570 --> 00:57:39,245
you, our brilliant listeners, for tuning in. Don't forget

924
00:57:39,245 --> 00:57:43,005
to subscribe, leave a review, and share this episode with your data

925
00:57:43,005 --> 00:57:46,365
loving friends or that one colleague who insists they don't trust

926
00:57:46,365 --> 00:57:49,885
AI. We'll convert them eventually. Until next

927
00:57:49,885 --> 00:57:53,425
time, stay curious, stay caffeinated, and remember,

928
00:57:53,720 --> 00:57:57,079
in a world driven by data there's no such thing as a trivial

929
00:57:57,079 --> 00:58:00,760
question, just fascinating answers waiting to be found. Catch

930
00:58:00,760 --> 00:58:02,700
you next time on Data Driven.