Episode 3: Generative AI over the Years: Security, Job Creation, and Fighting Writer’s Block

1
00:00:01.150 –> 00:00:05.030
[email protected]: All right. We are now recording. So I’m going to get started in a couple of seconds.

2
00:00:08.500 –> 00:00:09.570
[email protected]: All right. Here we go.

3
00:00:10.500 –> 00:00:19.100
[email protected]: Welcome back to motivation. AI matters. Today I am really excited to be joined by Sarah Luger.

5
00:00:33.850 –> 00:00:36.530
[email protected]: Sarah, thank you for joining us today.

6
00:00:38.190 –> 00:00:53.090
sarah: Thank you so much, Alex. It’s a real pleasure, as you know. I’ve been working in the space for some time, but also excited about Prisado’s role in this space. So thank you again for having me. My name is Sarah Luger.

7
00:00:53.210 –> 00:01:00.069
I got my Phd. At University of Edinburgh many years ago in artificial intelligence, and I would be.

8
00:01:00.480 –> 00:01:18.790
sarah: I’m being quite honest to say that I’m surprised and excited about the developments in this space. In the past I’ve worked at startups. I’ve worked at Ibm building a precursor to the Ibm. Watson jeopardy challenge robot.

9
00:01:18.820 –> 00:01:28.430
sarah: I’ve also been at Orange Silicon Valley for 5 years. We’ve worked on numerous topics, including voice by metrics. Chat bots call center technology.

10
00:01:28.480 –> 00:01:30.540
and of course.

11
00:01:31.100 –> 00:01:35.130

sarah: everything that has to do with large language models and generative. AI,

12
00:01:35.320 –> 00:01:39.469

I’m really excited, especially because, as you may or may not know.

13
00:01:39.690 –> 00:01:56.930

sarah: Orange is a large company with almost a hundred 40,000 employees, and we’re in 27 countries. Many of those are in North and West Africa, and many of our customers speak languages that are low resource, which means they don’t have a lot of training data.

14
00:01:57.030 –> 00:02:12.949

sarah: They don’t have an online presence that supports the kind of data and that’s used in conventional translation systems to create high quality translations. So I’ve been working a lot in that space. And I do see both potential

15
00:02:13.080 –> 00:02:20.550

sarah: for large language models to support our customers, but also some peril. And I’m looking forward to chatting more about that today. Thank you.

16
00:02:20.720 –> 00:02:26.429
[email protected]: I think that’s fantastic. And you’ve got a great background. I know you’ve

17
00:02:26.580 –> 00:02:38.099
[email protected]: you’ve seen this industry evolve through through numerous iterations. It’s most recent what is now being referred to as generative. AI.

18
00:02:38.370 –> 00:02:50.530
[email protected]: I know you. You just touched on a lot of really interesting topics. We’ll dive into over the course of this episode. But to give the the listeners a good baseline

19
00:02:50.730 –> 00:03:03.009
[email protected]: in your own words, could you define what this new term generative AI means, and then talk to me a little bit about who generative AI is important for

20
00:03:04.990 –> 00:03:06.020
sarah: okay?

21
00:03:06.390 –> 00:03:09.710
sarah: So I think

22
00:03:09.820 –> 00:03:15.070
sarah: for the the average person out there. General, today I

23
00:03:15.360 –> 00:03:25.520
sarah: is AI, where the output resembles human content. It resembles a language that is either

24
00:03:25.540 –> 00:03:28.680
that seems like it’s constructed by a human

25
00:03:29.000 –> 00:03:54.640
sarah: or technically generative AI systems are based on algorithms that learn from the a vast amount of input data. And the most recent cases that we’ll dig into. That would be all of the digital data that’s on the web as well as some knowledge bases knowledge bases being things like Wikipedia that give structure and associate terms. And

26
00:03:54.700 –> 00:04:02.799
sarah: some apparent meaning to this, this vast sea of of language data. And so

27
00:04:02.820 –> 00:04:05.690
sarah: what’s going on under the hood is that there is

28
00:04:05.840 –> 00:04:14.769
sarah: this vast amount of data is being used to learn the patterns of how we as humans speak

29
00:04:14.950 –> 00:04:26.620
sarah: and how we write, and with innovations, both from Google’s 217 transformer paper incredible compute innovations

30
00:04:26.700 –> 00:04:31.960
as well as just ongoing neural networks developments.

31
00:04:33.070 –> 00:04:41.310
sarah: There’s the possibility, as many of us have have now tried since November thirtieth, 2,022, when Chat Gpt was launched

32
00:04:41.340 –> 00:05:01.849
sarah: to engage with an generative AI system in a way that most people had not engaged with an AI system, you know. Perhaps in the past you had AI, a secondary characters in a video game, you know, or there’d maybe been some predictive analytics in an enterprise app

33
00:05:02.070 –> 00:05:15.019
sarah: application you were using. But the core of generative AI is using these patterns of words at at a vast scale. That then for us makes it seem like

34
00:05:15.220 –> 00:05:17.369
sarah: this computer, is

35
00:05:17.590 –> 00:05:25.859
sarah: it? Almost? It’s almost a a human like content that’s being output. And it’s really a powerful difference between

36
00:05:25.870 –> 00:05:36.760
sarah: systems from even gosh, even 6 months ago, right? We’ve had 7 months ago. We’ve had a seat change. And your second question is, who is it

37
00:05:37.180 –> 00:05:39.400
sarah: most?

38
00:05:39.790 –> 00:05:43.560
sarah: Who is it most important? Who is? Who is it most important? For

39
00:05:44.540 –> 00:05:49.109
sarah: who is it most important? For well, right now the

40
00:05:49.190 –> 00:05:57.200
sarah: we’re in. We’re in the the hype cycle, and it’s and it’s a little bit of it’s important for everyone.

41
00:05:57.240 –> 00:05:59.539
This is great for everything.

42
00:05:59.570 –> 00:06:11.029
sarah: And I respect the hype as someone who’s in Silicon Valley, because I understand the role that it plays and the duality of of

43

00:06:11.410 –> 00:06:22.590
sarah: of how we get an investment and how we build prop products and how we have to compete with other hype cycles, be them metaverse. most recently in blockchain.

44
00:06:22.670 –> 00:06:26.100
sarah: But I think that this is really important for

45

00:06:26.200 –> 00:06:29.799
sarah: creating customer-centric tools

46
00:06:29.970 –> 00:06:32.610

sarah: that support

47
00:06:32.830 –> 00:06:46.540

sarah: voice bots textual. You know, marketing text. I think that marketing and customer support are the first areas that are going to see innovations in these really
48

00:06:46.810 –> 00:06:49.360
sarah: human seeming

49
00:06:49.390 –> 00:06:54.050
sarah: engagements that can be created for their customers. So

50
00:06:54.560 –> 00:07:02.539
those are the 2 areas that the people I see that it will most of affect. But then I want to also flip that and say.

51
00:07:02.620 –> 00:07:06.619
sarah: I think the Holy Grail of enterprise.

52
00:07:06.790 –> 00:07:15.259
sarah: Innovation that isn’t as shiny and sparkly as some of the other. you know. Key terms I’ve just mentioned

53
00:07:15.280 –> 00:07:19.839

sarah: is enterprise intranet. Search.

54
00:07:20.020 –> 00:07:36.159
sarah: so the ability to search through a company’s resources, to answer questions for employees, or answer questions for employees that are then passed on to customers. I think that that is

55
00:07:36.430 –> 00:07:45.519
sarah: really key cause it will help you and I do our jobs better and reduce mundane tasks and reducing

56
00:07:45.820 –> 00:07:47.410
Monday and task.

57
00:07:48.060 –> 00:07:55.780
sarah: It’s something that AI in general is, you know, aims to do any technology. We try to elevate our

58
00:07:55.790 –> 00:07:58.139
sarah: our work tasks up

59
00:07:58.220 –> 00:08:10.129
the the difficulty chain. So we want, as humans to not do the same thing every day, but to understand patterns

60
00:08:10.180 –> 00:08:13.969
sarah: and reduce repetition

61
00:08:14.020 –> 00:08:29.960
sarah: and do more and more challenging tasks, and those more challenging tasks are very hard for computers, so don’t fret. Many of us will still have jobs. On the other hand, those lower level tasks are.

62
00:08:30.000 –> 00:08:39.289
sarah: really great opportunities for computers to come in generative AI systems to come in and support us.

63
00:08:41.820 –> 00:08:51.909
[email protected]: Yeah, you you’ve raised, you raise some really interesting points, some of which, you know we’ve discussed in in prior conversations. But I’ll I’ll summarize those quickly for the listeners.

64
00:08:52.080 –> 00:09:08.989
[email protected]: You know we we’ve we’ve talked about. And and I I love the way that you put it. You know, some hype cycles around the metaverse around blockchain. I think we’re we’re definitely seeing a surge of a hype cycle around generative AI,

65
00:09:09.050 –> 00:09:19.740
[email protected]: you know, to another one of your points. The barrier to entry for consuming this type of technology in in my observation has been lowered.

66
00:09:19.860 –> 00:09:23.610
[email protected]: as of November of last year.

67
00:09:23.630 –> 00:09:26.399
[email protected]: you know, chat

68
00:09:26.660 –> 00:09:36.730
[email protected]: so readily available for the average consumer. you know, broaching the prosumer pro prosumer market.

69
00:09:36.860 –> 00:09:37.900
[email protected]: And

70
00:09:37.910 –> 00:09:44.730
[email protected]: you know, as you’re alluding to breaking into the enterprise. Space as well.

71

00:09:44.910 –> 00:09:47.030

[email protected]: I think, is going to have

72

00:09:47.100 –> 00:09:59.210

[email protected]: widespread benefit both in terms of the way consumers interact with this type of technology, but also the way large enterprises similar to orange

73

00:09:59.340 –> 00:10:12.729

[email protected]: derive value from this type of technology. But I do want to touch on another point that you’ve made, which is unique to the other conversations that I’ve had in the last couple of months

74

00:10:12.840 –> 00:10:19.369

[email protected]: which are the implications around education and career development.

75

00:10:19.430 –> 00:10:27.680

[email protected]: I love what you said because it goes. It goes against the grain, in my opinion

76

00:10:27.870 –> 00:10:32.050

[email protected]: of what we have seen in the news around job displacement

77

00:10:32.230 –> 00:10:50.360

[email protected]: and and the automation of tasks. I think you delineate very well, and artificial intelligence will help eliminate some mundane tasks. But you put a great perspective on the state of the market by saying it will free

78

00:10:50.380 –> 00:10:59.559

[email protected]: humans up to do more challenging tasks, and I find that to be a very pragmatic and and and uplifting perspective.

79

00:11:00.400 –> 00:11:04.460

[email protected]: if you could expand on that a bit, Sarah.

80

00:11:05.040 –> 00:11:06.130

[email protected]: What

81

00:11:06.240 –> 00:11:09.140

[email protected]: would you? What would you say

82

00:11:09.430 –> 00:11:16.780

[email protected]: some interesting applications of humans and AI working together

83

00:11:16.920 –> 00:11:21.819

[email protected]: might be that we haven’t necessarily considered yet.

84

00:11:24.690 –> 00:11:25.560

sarah: Great?

85

00:11:26.330 –> 00:11:30.300

yeah. And I I appreciate you.

86

00:11:30.720 –> 00:11:42.030

sarah: your perspective as well, because I think that we’re at a a point where. because of the hype, there are some

87

00:11:42.930 –> 00:11:48.020

sarah: main primary narratives that seem to be

88

00:11:48.380 –> 00:11:57.640

they. They seem to be extremes. They’re either fear, based or or wildly at outlandish. And you know the the reality is somewhere in between.

89

00:11:57.730 –> 00:11:59.160

sarah: I’ve heard that

90

00:11:59.230 –> 00:12:03.270

with new innovations we grossly

91

00:12:03.350 –> 00:12:10.729

sarah: out, we overestimate the near term change, and then underestimate the long term, change

92

00:12:11.010 –> 00:12:27.699

sarah: and that’s not my quote, but I I strongly agree so to your point. What are, what are the ways that this hasn’t been using? And it hasn’t been clear and and with collaboration between humans and

93

00:12:27.790 –> 00:12:30.179

these tools. What will be

94

00:12:30.550 –> 00:12:39.399

sarah: more? what will become clear in the in the near term? So what is happening now, what do we see now?

95

00:12:39.910 –> 00:12:58.110

sarah: Right now, chat Gpt and Gpt. Based technologies are being used in a lot of marketing, writing, and text and ideation and creativity based tasks.

96

00:12:58.110 –> 00:13:21.119

sarah: So they’re really good for this blank page problem. Maybe call it writers. Block call it Also, from a prompt perspective. You know, we’ve all grown up to some extent feeling comfortable with Google searches. And so we we can figure out how to engineer prompts that give us an output that we’re looking for.

97

00:13:21.260 –> 00:13:32.460

But humans and computers working together has been There’s a long standing history of this. And in artificial intelligence we’ve developed agents

98

00:13:32.640 –> 00:13:39.060

sarah: which are some little little software

99

00:13:39.140 –> 00:13:57.239

sarah: products that help humans complete tasks. And there’s been research in this multimodal, you know, with sensors, what is there. There’s a lot of use in industry. But I think that the next wave of

100

00:13:57.590 –> 00:14:02.659

sarah: this technology will be using the

101

00:14:02.720 –> 00:14:15.749

sarah: trust which a lot of users have gained from this first iteration. And you know, I later, I want to maybe push back on that trust. But if we think of this as kind of a consumer life cycle.

102

00:14:15.910 –> 00:14:24.009

sarah: using the trust that the consumer has from the output of their large language model

103

00:14:24.030 –> 00:14:26.460

sarah: chat interface

104

00:14:26.520 –> 00:14:30.389

sarah: and turning that into a resource

105

00:14:31.230 –> 00:14:44.590

sarah: that maybe takes the place of a Google search. Right? So I think that chat interfaces are going to transform the way that humans work

106

00:14:44.690 –> 00:14:45.540

sarah: with

107

00:14:46.460 –> 00:15:01.670

sarah: computers in the fact that we may no longer have Google page rank be as an important in terms of Google being the font of all knowledge. Maybe that verb itself. I’m gonna Google. It will

108

00:15:01.780 –> 00:15:11.079

sarah: will cease to be as relevant. I think we’re going to have a split in our society where people who have become very

109

00:15:11.090 –> 00:15:23.469

sarah: adept at using chat technology may go to use this as their primary resource for new information, for for information discovery.

110

00:15:23.760 –> 00:15:35.740

I think that that could have profound implications on the Enterprise search and our ecosystems, you know, if we think of trying to make a decision about, say, is

111

00:15:35.970 –> 00:15:43.760

sarah: Siri, or Google Home or Apple home pod. What is what is the best smart device

112

00:15:43.840 –> 00:15:46.620

sarah: it’s hard to evaluate, which is

113

00:15:46.670 –> 00:15:51.900

sarah: in a you know, just a discrete test.

114

00:15:52.260 –> 00:16:03.200

sarah: But maybe, as as many consumers do, people make decisions based on their ecosystem. So I imagine in the future that

115

00:16:03.290 –> 00:16:06.279

sarah: open AI will leverage this

116

00:16:06.660 –> 00:16:16.670

sarah: collaboration with Microsoft to really pivot the enterprise solutions that the average person spends a lot of their time on.

117

00:16:16.830 –> 00:16:18.010

and by

118

00:16:18.240 –> 00:16:19.980

sarah: changing that habit.

119

00:16:20.620 –> 00:16:25.100

sarah: Anytime you use an ecosystem, you’re providing them with a lot of data.

120

00:16:25.200 –> 00:16:38.960

sarah: And when you provide an ecosystem with a lot of data, it allows them to personalize and learn your your preferences. And this could change the way that we online

121

00:16:39.440 –> 00:16:45.780

sarah: receive, ask for and receive new information, answer questions. And

122

00:16:45.860 –> 00:17:00.449

sarah: could could that pivot could reshuffle the companies that we get many of our goods and services from, or at least our initial point of contact for those goods and services, because I do think that

123

00:17:00.540 –> 00:17:07.520

sarah: especially younger consumers, will feel more comfortable using these tools

124

00:17:08.140 –> 00:17:09.640

sarah: as

125

00:17:09.670 –> 00:17:13.109

sarah: as a co-author, as a co-pilot

126

00:17:15.890 –> 00:17:28.250

[email protected]: that that’s really interesting. And and, Sarah, you’re you’re touching on a a topic that actually comes up with a lot of clients that I speak to. You know, we’re in an interesting

127

00:17:28.680 –> 00:17:40.300

[email protected]: environment from a macro perspective. At the moment there have been macroeconomic headwinds that many.

128

00:17:40.650 –> 00:17:53.389

[email protected]: you know, large mid market startup businesses have to navigate in terms of personnel decisions, R&D investment decisions

129

00:17:53.660 –> 00:17:57.960

[email protected]: specifically within tech, there have been a lot of reductions in force.

130

00:17:58.480 –> 00:18:02.189

[email protected]: So let me just paint that as the

131

00:18:02.480 –> 00:18:09.760

[email protected]: personnel backdrop for humans interacting with machines simultaneously

132

00:18:09.870 –> 00:18:15.489

[email protected]: this new technology appears seemingly out of nowhere last year

133

00:18:16.820 –> 00:18:29.850

[email protected]: to the untrained I it can come across as this, all seeing oracle that can answer any question. So if I’m sitting from the perspective of

134

00:18:30.540 –> 00:18:34.430

[email protected]: someone in the workforce that spells a threat.

135

00:18:35.030 –> 00:18:44.960

[email protected]: I love the direction that you’re taking your point of view, because the way that you phrased it and and I will have to look up this quote. We

136

00:18:45.490 –> 00:18:51.040

[email protected]: overestimate the near term impact and underestimate the long term impact.

137

00:18:51.670 –> 00:18:53.399

[email protected]: And what I’m finding

138

00:18:54.200 –> 00:18:58.689

[email protected]: in my personal experience with our clients is

139

00:19:00.140 –> 00:19:02.220

[email protected]: these Llms

140

00:19:03.150 –> 00:19:08.840

[email protected]: content, assistance or generation tools. They help alleviate

141

00:19:08.990 –> 00:19:13.850

[email protected]: writers block. They help get the first notes on the page.

142

00:19:14.250 –> 00:19:20.570

[email protected]: But that human expertise of motivating someone to act.

143

00:19:20.790 –> 00:19:24.670

[email protected]: of understanding the nuance of

144

00:19:24.850 –> 00:19:26.930

[email protected]: what’s being discussed

145

00:19:27.380 –> 00:19:32.249

[email protected]: is it is from from what I’m gathering, from what you’re saying, not quite.

146

00:19:33.290 –> 00:19:41.750

[email protected]: And it maybe this is 2 points, not quite the capability and not quite the near term intent of this technology. And I I think

147

00:19:41.860 –> 00:19:43.210

[email protected]: that that is.

148

00:19:44.430 –> 00:19:52.050

[email protected]: I think that that is a very good dynamic to exist specifically in the environment that we find ourselves in. Now.

149

00:19:54.100 –> 00:19:58.589

sarah: Alex, can I? Let me ask a clarification? So

150

00:19:58.740 –> 00:19:59.970

sarah: are you.

151

00:20:01.450 –> 00:20:08.319

sarah: I agree, completely with what you said. But is your question.

152

00:20:09.860 –> 00:20:17.150

sarah: I I think. Let me just. I kind of started thinking about one word you said intent human intent.

153

00:20:17.480 –> 00:20:25.640

sarah: And I think one of the challenges with the output of large language model systems

154

00:20:26.220 –> 00:20:28.580

is that we’re calling them.

155

00:20:29.170 –> 00:20:31.290

sarah: We’re calling that output information.

156

00:20:32.440 –> 00:20:37.259

sarah: And I’m not sure if that’s a a great word.

157

00:20:38.270 –> 00:20:42.870

There’s a researcher I spoke with last week

158

00:20:43.480 –> 00:20:50.090

sarah: who would like us to remember how large language models work on the inside

159

00:20:50.200 –> 00:20:59.430

sarah: and view the output as synthetic media. because information is constructed by a human.

160

00:20:59.570 –> 00:21:00.540

sarah: And if

161

00:21:01.090 –> 00:21:05.790

sarah: if words are not constructed by a human.

162

00:21:06.490 –> 00:21:08.999

is it information?

163

00:21:09.110 –> 00:21:26.929

sarah: And and this is a this is a very interesting discussion, and I think that this is nothing that we’re going to solve today. But it goes. It touches on another topic in this space, which is anthropomorphizing these these systems, as you as you noted, you know, they’re

164

00:21:27.480 –> 00:21:33.260

they’re impressive. And if you don’t know the inner workings. Or if you’re new to this space.

165

00:21:33.310 –> 00:21:38.250

sarah: or actually, that’s not fair. If you’re a human. you you

166

00:21:38.490 –> 00:21:40.560

automatically

167

00:21:40.710 –> 00:21:44.489

sarah: give intention to text

168

00:21:44.650 –> 00:21:56.639

sarah: you, you give you make meaning out of the kind of bag of words that you get back. And this bag of words is really good. And so it’s our natural tendency

169

00:21:56.830 –> 00:22:08.139

sarah: to to find meaning. I mean, think of conspiracy theories like we try to find meaning in in desperate facts. And this is this goes back to

170

00:22:08.390 –> 00:22:09.510

sarah: our

171

00:22:10.830 –> 00:22:15.670

sarah: our core. Human animal.

172

00:22:15.690 –> 00:22:26.339

sarah: backgrounds. You know, this is this is how we probably survived. on the African belt. Right? We needed to understand signal. So

173

00:22:26.670 –> 00:22:31.510

sarah: when I I hear the word hallucination.

174

00:22:31.640 –> 00:22:37.190

sarah: I also get a little bit concerned that we’re anthropomorphizing these systems.

175

00:22:39.220 –> 00:22:50.759

sarah: and I think maybe a better word would be error and answer. And by answer for more. 5 z these systems. And in using the word hallucination we’re about, we’re evoking

176

00:22:50.940 –> 00:22:55.560

sarah: spirituality, creativity. very

177

00:22:56.750 –> 00:23:05.190

sarah: ideas and words that are very close to human souls. And again, I want to take a step back and say, this is a machine people.

178

00:23:05.260 –> 00:23:07.920

it’s it’s an error.

179

00:23:07.990 –> 00:23:13.610

sarah: And so I just I just kind of went on that tirade because of you mentioning intention. But

180

00:23:13.790 –> 00:23:17.690

sarah: I think. I think, harnessing

181

00:23:18.300 –> 00:23:26.820

sarah: our human intention to complete tasks, to work together with each other. you know, to build companies and

182

00:23:26.930 –> 00:23:29.760

and support customers.

183

00:23:30.000 –> 00:23:42.770

sarah: We can use these tools to do that. But I I want to push back on hallucinations and perhaps information as as output.

184

00:23:42.990 –> 00:23:45.060

sarah: And then

185

00:23:46.140 –> 00:24:06.549

sarah: yeah, sorry, Alex, please. No, no, no need to apologize. Quite quite the contrary, Sarah. I am always looking for a good reason to weave the word anthropomorphize into a conversation, and I’m glad that you took it.

186

00:24:06.730 –> 00:24:12.050

[email protected]: Well, it’s interesting, right? Like it’s really interesting how we do this. And

187

00:24:12.880 –> 00:24:19.570

sarah: it’s not like it’s not a bad thing. It’s a natural thing. And this is why we

188

00:24:19.790 –> 00:24:26.220

sarah: you know, this is how we relate to our pets. I mean, you know, this is like

189

00:24:26.690 –> 00:24:36.510

sarah: you, you know. I’ll fight you if you don’t think my cat totally understands what I you know this is a totally human behavior, but

190

00:24:37.190 –> 00:24:39.880

sarah: I it is, and it’s something that that

191

00:24:39.980 –> 00:24:43.049

sarah: that has a

192

00:24:43.360 –> 00:24:52.900

sarah: It has it. It’s a core part of who we are, and there’s a reason that it’s in our the fabric of our humanity. but

193

00:24:53.190 –> 00:24:54.660

sarah: it also

194

00:24:54.820 –> 00:24:58.880

sarah: in this case confers trust

195

00:24:59.260 –> 00:25:03.180

and trust, is something that should be earned.

196

00:25:03.580 –> 00:25:06.500

sarah: and I like the fact

197

00:25:06.790 –> 00:25:08.930

sarah: that you know.

198

00:25:08.940 –> 00:25:23.180

as you said, a lot of people are saying, Oh, chat you! It is bad for education. hey? Chat Gpt has gotten a lot of folks who had not previously played with AI playing with AI. It’s lowered the barrier to entry

199

00:25:23.200 –> 00:25:24.740

sarah: in this space.

200

00:25:24.900 –> 00:25:30.390

There’s a ton of things that I had to learn in school that are now, you know, solved problems.

201

00:25:30.830 –> 00:25:31.970

sarah: you.

202

00:25:31.980 –> 00:25:50.090

sarah: you know the ensemble learning of the past is completely out the window. It’s great that I understand what’s going on. But if the tools are out there, there’s great documentation. There’s a great open source community. Then we have a lot of people entering this workspace

203

00:25:50.110 –> 00:26:04.230

sarah: who are going to be because because some of these current iterations are so fun they’re sticky. They’re good products, as as you and I in in the Bay Area would say.

204

00:26:04.420 –> 00:26:09.280

sarah: But as you also mentioned, that means that the

205

00:26:09.420 –> 00:26:13.330

sarah: that’s that’s in conflict with the macroeconomic trend

206

00:26:13.360 –> 00:26:20.859

sarah: of letting people go in a lot of mid-size companies, even the larger companies. And I think that there’s

207

00:26:21.120 –> 00:26:24.400

sarah: a an essence of

208

00:26:24.570 –> 00:26:39.430

sarah: a lot of mid-size companies that have AI teams are taking a step back and saying we were going to build this internally. And now we can buy this. And so we don’t need some of these internal people.

209

00:26:39.720 –> 00:26:40.560

and

210

00:26:41.180 –> 00:26:42.550

sarah: I think

211

00:26:42.850 –> 00:26:44.910

that if you look at

212

00:26:45.520 –> 00:26:50.770

sarah: the current strategic ecosystem.

213

00:26:51.100 –> 00:26:59.410

sarah: the strategically, this is exactly what open AI and some of the these companies want to do.

214

00:26:59.760 –> 00:27:03.920

sarah: they’re looking for customers. and

215

00:27:04.140 –> 00:27:26.679

sarah: to some degree the the price point has been so low that people that it’s really lowered the barrier in that dimension as well. It’s an accessible price point to learn these technologies, and then, when you’re playing with them at home. You bring them into the workspace and say, Hey, I’ve got some great ways that we can automate.

216

00:27:26.900 –> 00:27:36.320

sarah: Some workflows. but more broadly if you look at You know these, these macroeconomic trends.

217

00:27:37.710 –> 00:27:44.260

sarah: the open AI lost 54 million dollars and 22, and it has a hundred employees.

218

00:27:44.840 –> 00:27:45.640

sarah: The

219

00:27:45.940 –> 00:27:53.719

sarah: the technologies that we’ve been talking about are extremely extremely compute, intensive, which means

220

00:27:53.760 –> 00:28:05.160

sarah: they’re probably 2 companies who are making money out of Lms right now. One is Nvidia. and the other is whatever power company they were.

221

00:28:05.270 –> 00:28:10.600

sarah: They were accessing to run these these scrapes of the web.

222

00:28:13.060 –> 00:28:27.159

sarah: I doubt it’s Pg. And the as a Pg. And a customer So, Sarah, I want to touch on you. You can cut that. I don’t want them coming after me. That sounds good.

223

00:28:27.260 –> 00:28:35.190

[email protected]: Sarah, I want to touch on a themed. We have alluded to throughout the episode.

224

00:28:35.800 –> 00:28:38.150

[email protected]: and that is the team trust.

225

00:28:39.090 –> 00:28:42.250

sarah: as I think. So, yeah.

226

00:28:42.290 –> 00:28:45.080

[email protected]: as these companies

227

00:28:45.650 –> 00:28:56.389

[email protected]: be them. Large enterprises start ups even providers in this space. as these companies embark on their journey

228

00:28:56.400 –> 00:29:01.410

[email protected]: to either build Lms. Incorporate generative AI

229

00:29:01.540 –> 00:29:12.080

[email protected]: into their go to market. How can companies and the trust of their consumer base

230

00:29:12.140 –> 00:29:27.820

[email protected]: and use this technology in a way that will not feel invasive or not feel like. it is infringing on the privacy or the the rights of the consumer.

231

00:29:30.360 –> 00:29:34.259

Great question. I I think that this is going to be.

232

00:29:35.930 –> 00:29:42.250

sarah: You know, I do want to give respect to open AI for creating a great product because

233

00:29:43.190 –> 00:29:46.960

sarah: Lms have have been around vectorized.

234

00:29:46.990 –> 00:29:49.609

sarah: data

235

00:29:49.640 –> 00:29:54.439

word embeddings have been around, but they they created a front end. That that was.

236

00:29:54.580 –> 00:29:57.870

sarah: It’s a great experience. and

237

00:29:58.210 –> 00:30:09.370

sarah: I think that that in gender trust, in a way that they can parlay into building a

238

00:30:10.670 –> 00:30:25.570

sarah: partner and a customer business that will have some really big, impressive names. What does that mean? You know, when the rubber hits the road for everyone else. No.

239

00:30:25.740 –> 00:30:27.780

sarah: when you come into

240

00:30:28.060 –> 00:30:33.770

sarah: a situation where you’re trying to win a sale from

241

00:30:33.890 –> 00:30:38.129

sarah: a customer, I think it’s really important to say, what?

242

00:30:38.930 –> 00:30:57.289

sarah: How can I define my company? And what do I have that makes my company unique? And in most cases this question, before November thirtieth, 2,022 would be, I have a a relationship.

243

00:30:57.400 –> 00:31:07.420

sarah: and I with this customer. I have relationships with customers like this customer, and I have a long standing

244

00:31:07.740 –> 00:31:09.470

data set

245

00:31:09.580 –> 00:31:11.430

sarah: that includes

246

00:31:11.490 –> 00:31:29.060

sarah: well documented metadata of these relationships. Okay? So maybe that was a computer scientist describing an enterprise customer engagement. But I think that that fundamentally has not changed. If you came into a

247

00:31:29.100 –> 00:31:34.260

sarah: space where you are providing a service, and you have

248

00:31:34.370 –> 00:31:45.980

sarah: a great customer field. And because of that ongoing engagement over years, you have a ton of data that reflects your customers.

249

00:31:46.090 –> 00:31:49.849

sarah: Then you are going to be in a great situation

250

00:31:50.160 –> 00:31:52.319

sarah: in the large language model realm.

251

00:31:52.450 –> 00:32:05.930

sarah: Why? Because you’ve already gained the trust, and you’ve already created great results for your customers. And now you’re just laying a a technical

252

00:32:06.110 –> 00:32:13.129

sarah: technology on top of that, a technical layer that says we can

253

00:32:13.280 –> 00:32:29.280

sarah: optimize this past data even more, we can surface new insights. We can respond to your questions in a more naturalistic way. And we can have deeper

254

00:32:29.450 –> 00:32:53.260

sarah: information discovery from our pre existing information. Using some of these, vector, based approaches. Right? So when you think about trust, I don’t think about someone who’s coming to the market tomorrow with, I can solve X, y, and Z with large language models or chat Gpt, for

255

00:32:53.680 –> 00:33:07.300

sarah: you know, road cleaning, whatever it may be. But I do think that trust is about relationships, and from a technical perspective, it’s about data. And if you can leverage that data

256

00:33:07.550 –> 00:33:20.730

sarah: to create naturalistic experiences for your existing customers, you’re going to make them happy. And you’re going to gain more data is key and data is what

257

00:33:20.990 –> 00:33:31.749

sarah: is really amplified. The large language model scene. So I think a lot of companies need to understand that their data is gold.

258

00:33:31.790 –> 00:33:50.039

sarah: At Orange we, we take our customer engagement extremely seriously. We are customer-centric. Our our concern is great customer experience. And we know that we’re sitting on a gold mine of data. We know past preference.

259

00:33:50.280 –> 00:34:04.840

sarah: And now we have a tool that allows us to personalize and and create even better experiences, you know. And that’s we’re amplifying that trust

260

00:34:06.480 –> 00:34:20.589

[email protected]: fantastic. Well, Sarah, I know we’ve we’ve covered a lot of ground today. and thank you for all of the insight. I have one final question now that we’re on the topic of Orange and the projects that you’re working on.

261

00:34:21.150 –> 00:34:28.960

[email protected]: What What? That’s on your plate right now or in your pipeline excites you the most that you’re working on.

262

00:34:30.170 –> 00:34:33.969

sarah: thank you. Because

263

00:34:34.580 –> 00:34:39.539

sarah: I’ve been talking a lot about generative, and

264

00:34:39.770 –> 00:34:50.789

sarah: I I I hope this is as articulate as as as needed. But the thing that excites me the most is working on

265

00:34:50.969 –> 00:34:56.769

low resource language projects. We have a lot of customers

266

00:34:56.810 –> 00:35:00.150

sarah: in our West African. Well.

267

00:35:00.290 –> 00:35:11.150

sarah: okay, let me start at the beginning, Hi, my name is Sarah, and I’m really interested in low resource language projects. This is why I joined Orange 5 years ago.

268

00:35:11.780 –> 00:35:18.680

Oranges in a particularly well suited position to support

269

00:35:19.480 –> 00:35:34.140

sarah: our customers in numerous areas. We have a bank, we have video game company. We’ve entertainment streaming services. We lay Internet cable. We have satellites, we you know. What don’t we do? Well.

270

00:35:34.210 –> 00:35:35.330

sarah: we

271

00:35:35.420 –> 00:35:41.349

sarah: are focused on providing great consumer experiences to everyone.

272

00:35:41.450 –> 00:35:49.150

sarah: and many of our customers are in places that have underserved languages.

273

00:35:49.450 –> 00:36:01.050

sarah: So orange is in Francophone, Africa. But a lot of people in these regions no longer speak French is not is not considered

274

00:36:01.460 –> 00:36:06.989

a necessary language as as much as new trends in

275

00:36:07.000 –> 00:36:11.779

sarah: in political and community

276

00:36:12.170 –> 00:36:19.770

sarah: economics spring up, you know, there’s ability to to buy and sell products locally.

277

00:36:19.890 –> 00:36:31.249

sarah: And with that local focus a lot of people have been returning and to local languages that are spoken

278

00:36:31.640 –> 00:36:48.650

sarah: or comparatively underserved. And so what does that mean? It means that with the shifting dynamics of these areas as well as other, you know, macroeconomic forces, we have a lot of customers who do not speak French.

279

00:36:48.690 –> 00:36:52.640

sarah: or would prefer not to speak French.

280

00:36:53.030 –> 00:36:53.910

sarah: Now.

281

00:36:54.180 –> 00:37:07.009

sarah: I describe to the uninitiated low resource languages as languages where there’s no reddit, although in the past week that illusion is less powerful. But

282

00:37:07.040 –> 00:37:26.570

sarah: it means that there’s not a lot of data online. So I am very passionate about selling people products in whatever language they would like, and providing goods and services and whatever language they would like. So recently, I ran a machine translation competition between French and Bombbara. That was text based. We had

283

00:37:26.930 –> 00:37:34.860

sarah: 20 participants cash prizes. The winters are from all over the globe, and well.

284

00:37:34.960 –> 00:37:48.069

sarah: only a couple of the participants actually spoke Bambara. Almost all of the participants in the competition spoke, or had family members who spoke a different low resource language. And

285

00:37:48.140 –> 00:37:58.590

sarah: this was a really impressive experience, because it reminded us that no matter how much the large language model hype cycle

286

00:37:58.830 –> 00:38:02.260

sarah: you know, brings us cool, sticky

287

00:38:02.280 –> 00:38:04.259

sarah: products and

288

00:38:04.280 –> 00:38:16.720

sarah: and crazy new applications, and really And you know the Pope and a puffer coat whatever it may be. I think it’s really important to remember that

289

00:38:16.840 –> 00:38:31.330

sarah: natural language processing is not a solved domain. And there are a lot of challenges out there for the global the global population that are not English centric or not French centric.

290

00:38:31.360 –> 00:38:38.210

sarah: And I’m hoping that you know, when we were working on this last iteration of the translation project.

291

00:38:38.240 –> 00:38:51.019

sarah: Everyone was saying, Okay, large language models have dropped. How relevant are they for for this domain. And then so our next step is to experiment with that and figure out, how can you

292

00:38:51.030 –> 00:39:02.640

sarah: augment large language models for languages that are basically small language models and and you know, have their own orthographic challenges.

293

00:39:02.710 –> 00:39:12.449

sarah: And at the end of the day it’s really about customer service and people. So I think it’s it’s been a really great experience. And I’ve met a lot of really interesting people.

294

00:39:12.570 –> 00:39:20.619

sarah: and I trust that that will continue to be a great way to leverage new technologies

295

00:39:20.640 –> 00:39:23.330

sarah: with a global

296

00:39:23.470 –> 00:39:29.129

sarah: economy and global engagement. But I do want to reiterate that

297

00:39:29.580 –> 00:39:39.040

sarah: Llms have not solved AI. We there’s probably going to be a plateau right now, where

298

00:39:39.240 –> 00:39:46.870

sarah: open AI has not has mentioned. They don’t have plans to release a Gpt. 5 barred

299

00:39:47.010 –> 00:39:51.170

sarah: different, you know the llamas

300

00:39:51.190 –> 00:39:52.590

different

301

00:39:52.660 –> 00:39:58.190

sarah: models are out there, and there’s been a lot of optimization around

302

00:39:58.210 –> 00:40:15.360

sarah: compute and running these models on on smaller and smaller memory devices. And so I would watch for to see what Apple is up to, because I think they’re going to. They’re going to jump on the scene with a pretty interesting privacy.

303

00:40:15.440 –> 00:40:18.919

sarah: oriented solution as is their brand.

304

00:40:19.160 –> 00:40:30.080

sarah: But I’m I’m really excited about the power of communication and the power of speaking to people in their own language. So watch the space for more machine translation

305

00:40:30.240 –> 00:40:34.400

sarah: and low resource language innovations.

306

00:40:35.490 –> 00:40:44.690

Well, well, thank you, Sarah, and you know this has been a a fascinating discussion, one which I think the audience is really going to appreciate, not only for the

307

00:40:44.800 –> 00:40:49.650

[email protected]: information and insight that you’ve provided, but I think also a very

308

00:40:49.830 –> 00:40:59.740

[email protected]: actionable optimistic and pragmatic point of view. again for the listeners. This was Sarah Luger.

309

00:40:59.880 –> 00:41:12.280

[email protected]: Sarah, thank you very much for joining. I really enjoyed our conversation. and I’m looking forward to the next one.

310

00:41:14.060 –> 00:41:16.500

sarah: Thank you, Alex. It was a pleasure.