1
00:00:03,320 --> 00:00:06,560
Speaker 1: Hello, Hello, Welcome to Smart Talks with IBM, a podcast

2
00:00:06,600 --> 00:00:12,520
Speaker 1: from Pushkin Industries, iHeartRadio and IBM. I'm Malcolm Glapwell. This season,

3
00:00:12,760 --> 00:00:15,920
Speaker 1: we're diving back into the world of artificial intelligence, but

4
00:00:16,000 --> 00:00:21,920
Speaker 1: with a focus on the powerful concept of open its possibilities, implications,

5
00:00:21,960 --> 00:00:25,439
Speaker 1: and misconceptions. We'll look at openness from a variety of

6
00:00:25,480 --> 00:00:29,240
Speaker 1: angles and explore how the concept is already reshaping industries,

7
00:00:29,760 --> 00:00:33,599
Speaker 1: ways of doing business and our very notion of what's possible.

8
00:00:34,400 --> 00:00:38,760
Speaker 1: In today's episode, Jacob Goldstein sat down with Mariam Ashuri,

9
00:00:39,120 --> 00:00:42,360
Speaker 1: the Director of Product Management and Head of Product for

10
00:00:42,440 --> 00:00:46,840
Speaker 1: IBM's Watson x dot AI, where she spearheads the product

11
00:00:46,920 --> 00:00:52,320
Speaker 1: strategy and delivery of IBM's Watson x foundation models. She

12
00:00:52,400 --> 00:00:55,840
Speaker 1: is a technologist with more than fifteen years of experience

13
00:00:56,200 --> 00:01:01,640
Speaker 1: developing data driven technologies. The conversation folks on how enterprises

14
00:01:01,880 --> 00:01:06,040
Speaker 1: can use technology to build and deliver greater transparency in

15
00:01:06,120 --> 00:01:11,240
Speaker 1: AI With Granite. Meriam explained how Grantite can be utilized

16
00:01:11,520 --> 00:01:16,600
Speaker 1: to improve efficiency across various domains. She discussed how these

17
00:01:16,640 --> 00:01:20,840
Speaker 1: models are being used in real world business applications, particularly

18
00:01:20,920 --> 00:01:25,440
Speaker 1: in areas like customer care, where AI can help enable quick,

19
00:01:25,920 --> 00:01:31,679
Speaker 1: accurate responses based on internal company data. Meriam provided a

20
00:01:31,720 --> 00:01:36,479
Speaker 1: fascinating look into how enterprises have moved from mere experimentation

21
00:01:36,720 --> 00:01:42,480
Speaker 1: with generative AI to actual production, navigating challenges such as

22
00:01:42,520 --> 00:01:47,880
Speaker 1: increased latency, cost, and energy consumption. She highlighted how the

23
00:01:47,920 --> 00:01:52,480
Speaker 1: emerging trend of smaller models customized with proprietary data can

24
00:01:52,520 --> 00:01:56,200
Speaker 1: potentially deliver high performance at a fraction of the cost,

25
00:01:56,880 --> 00:02:02,120
Speaker 1: marking a significant shift in how enterprises leverage AI. Whether

26
00:02:02,160 --> 00:02:04,960
Speaker 1: you're an AI enthusiast, we're a business leader looking to

27
00:02:05,000 --> 00:02:09,720
Speaker 1: harness the power of artificial intelligence, this episode is packed

28
00:02:10,240 --> 00:02:13,919
Speaker 1: with valuable insights and forward thinking strategies.

29
00:02:18,240 --> 00:02:20,200
Speaker 2: Let's just start with your background. How did you come

30
00:02:20,240 --> 00:02:21,400
Speaker 2: to work at IBM.

31
00:02:22,080 --> 00:02:25,560
Speaker 3: I joined IBM right after I graduated. I have an

32
00:02:25,560 --> 00:02:31,680
Speaker 3: AI background, and throughout the years, I've held many roles

33
00:02:31,680 --> 00:02:37,160
Speaker 3: in design, engineering, development, research, mostly focused on AI application

34
00:02:37,280 --> 00:02:41,400
Speaker 3: development and design. In my current job, I'm the product

35
00:02:41,440 --> 00:02:45,880
Speaker 3: owner for What's the Next DAYI, which is the IBM

36
00:02:45,919 --> 00:02:50,000
Speaker 3: platform for enterprise AI. What excites me about this job,

37
00:02:50,040 --> 00:02:53,840
Speaker 3: I would say, is the technology advancements over the last

38
00:02:53,840 --> 00:02:57,080
Speaker 3: eighteen months in the market. We've been witnessing how generative

39
00:02:57,120 --> 00:02:59,720
Speaker 3: II has been changing the market. The way that I

40
00:02:59,760 --> 00:03:03,160
Speaker 3: see that is JENAI has been perhaps one of the

41
00:03:03,280 --> 00:03:07,079
Speaker 3: largest paradigm shifts when we think about productivity. The same

42
00:03:07,120 --> 00:03:12,520
Speaker 3: way that Internet and personal computers impacted the productivity of workforce,

43
00:03:12,720 --> 00:03:17,560
Speaker 3: Now we are witnessing another wave of all those opportunities

44
00:03:17,600 --> 00:03:20,840
Speaker 3: that it can unlock for especially enterprise AI when it

45
00:03:20,880 --> 00:03:24,960
Speaker 3: comes to enhancing the productivity of the workforce and releasing

46
00:03:25,040 --> 00:03:29,680
Speaker 3: some time that can potentially be put into creating more

47
00:03:29,800 --> 00:03:34,239
Speaker 3: value work for enterprise. So that's the major part that

48
00:03:34,360 --> 00:03:38,200
Speaker 3: I picked this team to have an impact on the

49
00:03:38,240 --> 00:03:42,280
Speaker 3: market and the community, but also of course using the

50
00:03:42,880 --> 00:03:45,840
Speaker 3: skills that I gain through all these years through IBM

51
00:03:46,000 --> 00:03:49,960
Speaker 3: to help to establish IBM as the market leader for

52
00:03:50,120 --> 00:03:50,800
Speaker 3: enterprise AI.

53
00:03:51,440 --> 00:03:55,520
Speaker 2: So you talked about jenai as this sort of generational,

54
00:03:55,680 --> 00:04:00,760
Speaker 2: transformational technological force, and I'm curious just in terms of

55
00:04:01,200 --> 00:04:03,640
Speaker 2: how it's going to come into the world, Like, how

56
00:04:03,640 --> 00:04:07,720
Speaker 2: do you see market adoption of genai sort of evolving

57
00:04:07,760 --> 00:04:08,240
Speaker 2: from here?

58
00:04:09,040 --> 00:04:12,640
Speaker 3: Well, last year was the year of excitement about generative AI.

59
00:04:12,760 --> 00:04:15,880
Speaker 3: Most of the companies were experimenting and exploring with GENI.

60
00:04:16,600 --> 00:04:20,200
Speaker 3: We see that energy shifted towards how to best monetize

61
00:04:20,200 --> 00:04:23,040
Speaker 3: that technology. Almost half of the market has moved from

62
00:04:23,360 --> 00:04:28,560
Speaker 3: investigation to pilots, ten percent has moved to production. When

63
00:04:28,600 --> 00:04:32,240
Speaker 3: you're exploring with this technology, you're looking for a valve factor,

64
00:04:32,520 --> 00:04:36,039
Speaker 3: You're looking for an AHA moment. That's why very large

65
00:04:36,080 --> 00:04:40,680
Speaker 3: general purpose models shine. But as companies move toward production

66
00:04:40,760 --> 00:04:43,680
Speaker 3: and scale, they soon realized the past success is not

67
00:04:43,720 --> 00:04:48,760
Speaker 3: that straightforward. For example, they're larger the model, the larger

68
00:04:48,839 --> 00:04:53,400
Speaker 3: computer resources it requires. That translates to increased latency that's

69
00:04:53,400 --> 00:04:57,800
Speaker 3: your response time. That translates to increased cost. That translates

70
00:04:57,839 --> 00:05:01,960
Speaker 3: to increase carbon, food print, energy consumption. So think about that.

71
00:05:02,120 --> 00:05:05,560
Speaker 3: At the scale of enterprise in production, some of them

72
00:05:05,920 --> 00:05:09,719
Speaker 3: can be a showstopper. Because of this reason, what actually

73
00:05:10,000 --> 00:05:15,200
Speaker 3: c is emerging in the market is instead of focusing

74
00:05:15,240 --> 00:05:20,520
Speaker 3: on very large general purpose models, coming back to very small,

75
00:05:21,520 --> 00:05:26,400
Speaker 3: trustworthy models that they can customize on their own proprietary

76
00:05:26,480 --> 00:05:29,279
Speaker 3: data that's the data about their customers, that the data

77
00:05:29,360 --> 00:05:33,960
Speaker 3: about their specific domains to create something differentiated that is

78
00:05:34,080 --> 00:05:38,440
Speaker 3: much smaller and delivers the performance that they want on

79
00:05:38,560 --> 00:05:41,159
Speaker 3: a target use case for a fraction of the costs.

80
00:05:41,480 --> 00:05:45,360
Speaker 2: Uh huh. So let's talk a little bit more specifically

81
00:05:45,400 --> 00:05:49,520
Speaker 2: about what you're working on. Let's talk about Granite. First

82
00:05:49,520 --> 00:05:51,200
Speaker 2: of all, tell me what is Granite.

83
00:05:51,800 --> 00:05:57,279
Speaker 3: Granite is our industrial leading family of models, flagship IBM models.

84
00:05:58,040 --> 00:06:02,239
Speaker 3: These are the models that we from scratch. When offered

85
00:06:02,240 --> 00:06:05,680
Speaker 3: to our platform, we offer indemnification and we stand behind

86
00:06:05,680 --> 00:06:12,680
Speaker 3: them today. It comes in four flavors, language, code, time series,

87
00:06:12,720 --> 00:06:19,240
Speaker 3: and geospecial models. Granite language series is covering English, Spanish, German,

88
00:06:19,680 --> 00:06:24,479
Speaker 3: Portuguese and Japanese. We have a combination of commercial and

89
00:06:24,720 --> 00:06:28,680
Speaker 3: open source language models on Granite. For example, we recently

90
00:06:28,880 --> 00:06:34,000
Speaker 3: released the Granite seven B language model, small powerful English model.

91
00:06:34,760 --> 00:06:38,080
Speaker 3: On the code front, our models are state of the

92
00:06:38,200 --> 00:06:42,480
Speaker 3: art models ranging from three billion to thirty four billion parameters.

93
00:06:43,120 --> 00:06:48,320
Speaker 3: These are very powerful models that performs or outperforms in

94
00:06:48,320 --> 00:06:52,159
Speaker 3: some cases the popular open source models in their right class.

95
00:06:52,200 --> 00:06:53,560
Speaker 3: So very powerful models.

96
00:06:53,760 --> 00:06:56,440
Speaker 2: So I get the idea a big picture about these models,

97
00:06:56,480 --> 00:06:58,200
Speaker 2: but it would be helpful to just get a sense

98
00:06:58,200 --> 00:07:00,599
Speaker 2: specifically of what they're doing. And you give me any

99
00:07:00,600 --> 00:07:04,920
Speaker 2: specific examples of how these models are being used in

100
00:07:05,000 --> 00:07:07,119
Speaker 2: businesses in the real world right now.

101
00:07:08,240 --> 00:07:11,520
Speaker 3: Well, the top use cases for generative AI are really

102
00:07:11,600 --> 00:07:18,480
Speaker 3: content generation, summarization, information extraction. Perhaps the most popular use

103
00:07:18,520 --> 00:07:22,200
Speaker 3: case that we are seeing in enterprise is content grounded

104
00:07:22,280 --> 00:07:26,520
Speaker 3: question and answering. So using these models as a base

105
00:07:26,800 --> 00:07:29,680
Speaker 3: to connect them to a body of information let's say,

106
00:07:29,720 --> 00:07:34,040
Speaker 3: their policies, their documents that is internal to the enterprise,

107
00:07:34,320 --> 00:07:38,520
Speaker 3: and get the model to provide answers based on that question.

108
00:07:38,920 --> 00:07:42,880
Speaker 3: One example of that is for customer agents customer care,

109
00:07:43,280 --> 00:07:47,840
Speaker 3: when a customer is asking a question. Previously, the agent

110
00:07:47,920 --> 00:07:51,440
Speaker 3: that responds to the customer had to answer the question

111
00:07:51,560 --> 00:07:54,360
Speaker 3: and if they don't know the answer escalated to the product.

112
00:07:54,480 --> 00:07:57,960
Speaker 3: Especially is keeping people on hold on the line to

113
00:07:58,120 --> 00:08:01,120
Speaker 3: go figure out the answer for them and then come back.

114
00:08:01,160 --> 00:08:03,520
Speaker 3: You can think of the time it takes to resolve

115
00:08:03,560 --> 00:08:07,240
Speaker 3: an issue. But now we llms, we have an opportunity

116
00:08:07,280 --> 00:08:11,320
Speaker 3: to automatically retrieve the information based on the internal documents

117
00:08:11,360 --> 00:08:14,360
Speaker 3: of the company, formulate an answer, show it to the

118
00:08:14,440 --> 00:08:17,800
Speaker 3: human agent, and then if they verify with the sources

119
00:08:17,840 --> 00:08:20,520
Speaker 3: of aries coming from, they can just translate it directly

120
00:08:20,560 --> 00:08:24,680
Speaker 3: to the customer. This is a very simple example of

121
00:08:24,720 --> 00:08:26,560
Speaker 3: how it's impacting the customer care.

122
00:08:27,040 --> 00:08:31,520
Speaker 2: So one big theme of this season is this idea

123
00:08:31,520 --> 00:08:34,920
Speaker 2: of open and one of the things that's interesting to

124
00:08:35,040 --> 00:08:38,839
Speaker 2: me about the work you're doing is you are using

125
00:08:38,920 --> 00:08:42,560
Speaker 2: not only granted this model IBM developed, but you're also

126
00:08:42,720 --> 00:08:46,320
Speaker 2: using third party models right from other places. So tell

127
00:08:46,360 --> 00:08:48,360
Speaker 2: me about that work and how that is sort of

128
00:08:48,400 --> 00:08:52,480
Speaker 2: fitting into your kind of real world typically enterprise GENAI work.

129
00:08:53,160 --> 00:08:56,559
Speaker 3: When it comes to model strategy, our strategy is really

130
00:08:56,600 --> 00:09:01,120
Speaker 3: focused on two pillars, multimodel and multidiplom It means that

131
00:09:01,200 --> 00:09:04,840
Speaker 3: we don't believe one single model rules all the use cases.

132
00:09:05,080 --> 00:09:07,240
Speaker 3: And I think at this point the market has also

133
00:09:07,320 --> 00:09:10,640
Speaker 3: realized the enterprise markets in average today are using five

134
00:09:10,679 --> 00:09:15,560
Speaker 3: to ten different models for different use cases. Oh interesting,

135
00:09:15,880 --> 00:09:18,160
Speaker 3: So in our portfolio, if you look into what's on

136
00:09:18,240 --> 00:09:21,040
Speaker 3: Extra DAYI today, we are offering a large sets of

137
00:09:21,280 --> 00:09:24,120
Speaker 3: high performing, state of the art models coming from open

138
00:09:24,160 --> 00:09:28,359
Speaker 3: source commercial models that we are bringing through our partners

139
00:09:28,679 --> 00:09:32,559
Speaker 3: and also IBM developed models. In addition to all of these,

140
00:09:32,760 --> 00:09:35,760
Speaker 3: we also have an option for bring your own model

141
00:09:36,080 --> 00:09:39,040
Speaker 3: from outside the platform. Let's say you have a custom

142
00:09:39,120 --> 00:09:41,800
Speaker 3: model that you made it yourself, you can bring it

143
00:09:41,840 --> 00:09:46,720
Speaker 3: to the platform and really helping the customers to navigate

144
00:09:46,760 --> 00:09:50,440
Speaker 3: through avoid range of models and pick the right model

145
00:09:50,679 --> 00:09:54,320
Speaker 3: for their target use case. Throughout that we've been heavily

146
00:09:54,360 --> 00:09:57,560
Speaker 3: working with our partners, and you know, this is the

147
00:09:57,600 --> 00:10:01,080
Speaker 3: market that is evolving rapidly. We've been at the forefront

148
00:10:01,080 --> 00:10:03,240
Speaker 3: of a spit to delivery. One example that I like

149
00:10:03,320 --> 00:10:08,760
Speaker 3: to highlight is recently Metal released Lama four or five billion,

150
00:10:09,080 --> 00:10:11,640
Speaker 3: such a powerful model. On the same day that it

151
00:10:11,800 --> 00:10:14,760
Speaker 3: was released to the market, we made it available in

152
00:10:14,840 --> 00:10:17,880
Speaker 3: our platform to our customers the same day. And not

153
00:10:17,960 --> 00:10:20,400
Speaker 3: only we delivered it on the same day, we are

154
00:10:20,400 --> 00:10:24,880
Speaker 3: offering competitive pricing but also for flexibility in where to deploy.

155
00:10:25,000 --> 00:10:27,920
Speaker 3: So we are giving an option to enterprise to deploy

156
00:10:28,000 --> 00:10:32,240
Speaker 3: these models on the platform of dat choice, either multi

157
00:10:32,240 --> 00:10:36,120
Speaker 3: cloud it can be gcpaws as youre IBM cloud, or

158
00:10:36,160 --> 00:10:41,520
Speaker 3: on premises. The same for mistrall Ai. Mistrall Ai recently

159
00:10:41,720 --> 00:10:44,680
Speaker 3: released the model misroll large two on the same day

160
00:10:44,960 --> 00:10:47,960
Speaker 3: we delivered that through the platform. That's an example of

161
00:10:48,000 --> 00:10:52,560
Speaker 3: a commercial model Lama but open source, but large two

162
00:10:52,720 --> 00:10:56,120
Speaker 3: is a commercial model that we made available through the platform.

163
00:10:56,720 --> 00:11:02,320
Speaker 2: Great. So I want to talk about enterprise grade foundation models,

164
00:11:03,000 --> 00:11:05,880
Speaker 2: just to get into it briefly, what's a foundation model.

165
00:11:06,360 --> 00:11:10,080
Speaker 3: People associate foundation models with a large language model, but

166
00:11:10,200 --> 00:11:13,520
Speaker 3: large language models are really a subset of foundation models.

167
00:11:13,679 --> 00:11:17,600
Speaker 3: Large language models are focused on language, but foundation models

168
00:11:17,640 --> 00:11:21,480
Speaker 3: can be code generators, can be focused on time series

169
00:11:21,520 --> 00:11:24,080
Speaker 3: model we talked about, they can be images, it can

170
00:11:24,120 --> 00:11:29,239
Speaker 3: be jewispecial models. So foundation model, as the term suggests,

171
00:11:29,400 --> 00:11:34,640
Speaker 3: your foundations to create a series of subsequent models that

172
00:11:34,800 --> 00:11:38,719
Speaker 3: can be customized for a downstream use case. And that's

173
00:11:38,760 --> 00:11:42,079
Speaker 3: why they are calling them foundation models. LM is a

174
00:11:42,120 --> 00:11:44,720
Speaker 3: good example of that as a subset for language that

175
00:11:44,800 --> 00:11:49,000
Speaker 3: you can further customize on your specific data to get

176
00:11:49,040 --> 00:11:52,600
Speaker 3: the model to do other works. So the core of

177
00:11:52,600 --> 00:11:56,120
Speaker 3: these foundation models, they are basically trained on an ab

178
00:11:56,200 --> 00:12:00,280
Speaker 3: third amount of data r data sets that of the

179
00:12:00,320 --> 00:12:03,560
Speaker 3: institutions today are sourcing them from the Internet, So you

180
00:12:03,600 --> 00:12:06,440
Speaker 3: can imagine what can potentially go to those models, and

181
00:12:06,480 --> 00:12:10,080
Speaker 3: then it comes to the enterprise and they start using it.

182
00:12:10,480 --> 00:12:15,400
Speaker 3: So for us also, when we started looking into in particular,

183
00:12:15,520 --> 00:12:19,600
Speaker 3: it was triggered by customers asking us to provide client

184
00:12:19,720 --> 00:12:23,000
Speaker 3: protections on these models, and we started thinking about, let's

185
00:12:23,000 --> 00:12:25,800
Speaker 3: look into how the models are trained and if we

186
00:12:25,880 --> 00:12:29,760
Speaker 3: are comfortable of fering client protections on the models that

187
00:12:29,800 --> 00:12:32,480
Speaker 3: are available in the market. And guess what, for a

188
00:12:32,559 --> 00:12:36,640
Speaker 3: majority of these models. There is absolutely no visibility into

189
00:12:36,720 --> 00:12:40,160
Speaker 3: what data vent into those models, not much transparency into

190
00:12:40,200 --> 00:12:44,240
Speaker 3: how the model trains, and the responsibility lies on you

191
00:12:44,320 --> 00:12:46,600
Speaker 3: as the customers we start using those models.

192
00:12:46,640 --> 00:12:50,440
Speaker 2: So just to be clear, that is presenting like potential risk,

193
00:12:50,600 --> 00:12:53,840
Speaker 2: real potential risk to a company that is using these models.

194
00:12:54,120 --> 00:12:56,960
Speaker 3: It is. It is a potential risk in particular for

195
00:12:57,200 --> 00:13:01,840
Speaker 3: the customers in highly regulated industries. So what we did

196
00:13:01,880 --> 00:13:05,960
Speaker 3: for Granite was when we started training these models from scratch,

197
00:13:06,040 --> 00:13:09,080
Speaker 3: Basically we went to the corpus of data that was

198
00:13:09,120 --> 00:13:12,640
Speaker 3: available to us. So, for example, the very first version

199
00:13:12,720 --> 00:13:16,920
Speaker 3: of Granite was exposed to twenty persons of its data

200
00:13:16,960 --> 00:13:19,720
Speaker 3: from finance and legal because we have a lot of

201
00:13:20,480 --> 00:13:25,000
Speaker 3: financial institutions as our clients. We worked directly with our

202
00:13:25,080 --> 00:13:30,240
Speaker 3: IBM research to identify detectors for harmful information like haytype

203
00:13:30,320 --> 00:13:31,959
Speaker 3: use and profanity detectors.

204
00:13:32,520 --> 00:13:34,839
Speaker 2: Okay, so we're talking about Granted, we're talking about this

205
00:13:35,080 --> 00:13:38,360
Speaker 2: set of models IBM has developed. Let's talk about using

206
00:13:38,400 --> 00:13:43,200
Speaker 2: Granite on Watson X compared to downloading open source models,

207
00:13:43,320 --> 00:13:44,240
Speaker 2: Like how do those differ?

208
00:13:44,880 --> 00:13:48,520
Speaker 3: By using Granite and Whatson X, you get two things.

209
00:13:48,920 --> 00:13:52,400
Speaker 3: The first one is the client protection and in themification

210
00:13:52,559 --> 00:13:54,760
Speaker 3: that we talked about. You get that if the model

211
00:13:54,800 --> 00:13:58,520
Speaker 3: is consumed through our platform. And the second one is

212
00:13:58,800 --> 00:14:02,600
Speaker 3: really the equos of platform capabilities that we are offering

213
00:14:02,720 --> 00:14:05,920
Speaker 3: to help you create value on top of those data,

214
00:14:06,000 --> 00:14:09,960
Speaker 3: so for example, bringing your data to customize granted for

215
00:14:10,040 --> 00:14:13,360
Speaker 3: your own specific use case. But also one thing that

216
00:14:13,400 --> 00:14:16,880
Speaker 3: I like to highlight in particular is the AI governance.

217
00:14:17,440 --> 00:14:20,120
Speaker 3: So when you get one of these pre train models,

218
00:14:20,360 --> 00:14:23,520
Speaker 3: you put it in front of your own users. Through

219
00:14:23,560 --> 00:14:28,200
Speaker 3: the input and instructions that the user provides for the model,

220
00:14:28,680 --> 00:14:32,920
Speaker 3: they can nodge the model to potentially create undesired behavior

221
00:14:33,240 --> 00:14:35,920
Speaker 3: and change the behavior of the model. And because of

222
00:14:35,960 --> 00:14:40,200
Speaker 3: this is extremely important to automatically document the lineage of

223
00:14:40,280 --> 00:14:44,760
Speaker 3: who touched the model at one point, so if something happens,

224
00:14:44,800 --> 00:14:47,400
Speaker 3: you can trace it back and see where it's coming from.

225
00:14:47,800 --> 00:14:52,600
Speaker 3: And that's what's an extra governance is offering automatically documenting

226
00:14:52,720 --> 00:14:56,280
Speaker 3: the lineage. When you use the grantite within the platform,

227
00:14:56,320 --> 00:14:58,640
Speaker 3: you get all of those you can have the end

228
00:14:58,680 --> 00:15:01,960
Speaker 3: to end governance, you can and have access to all

229
00:15:01,960 --> 00:15:05,720
Speaker 3: these scalable deployment opportunities that is available for you, like

230
00:15:05,840 --> 00:15:08,720
Speaker 3: to allow you deploy them on the platform of your

231
00:15:08,800 --> 00:15:12,440
Speaker 3: choice that we talked about, either multiple cloud or on

232
00:15:12,520 --> 00:15:15,600
Speaker 3: prem and it also helps you to have access to

233
00:15:15,600 --> 00:15:20,880
Speaker 3: avoid range of model customizations approaches, promptuning, fine tuning, retrieval,

234
00:15:20,920 --> 00:15:24,440
Speaker 3: augmented generations agents. There is a series of them available

235
00:15:24,760 --> 00:15:26,360
Speaker 3: to use an apply to your model.

236
00:15:27,120 --> 00:15:31,600
Speaker 1: This distinction between large language models and foundation models is

237
00:15:31,640 --> 00:15:36,120
Speaker 1: eye opening. Miriam emphasized that foundation models can be tailored

238
00:15:36,120 --> 00:15:41,120
Speaker 1: to specific tasks, but with that versatility comes a significant

239
00:15:41,200 --> 00:15:45,600
Speaker 1: challenge the lack of transparency and how these models are trained.

240
00:15:46,400 --> 00:15:50,400
Speaker 1: This composed a real risk, especially in highly regulated industries

241
00:15:50,720 --> 00:15:56,160
Speaker 1: like finance. Essentially, by using Granite and watsonex together, enterprises

242
00:15:56,200 --> 00:15:58,960
Speaker 1: get powerful and customizable tools.

243
00:16:00,120 --> 00:16:02,320
Speaker 2: So let's talk about the future a little bit. What

244
00:16:02,400 --> 00:16:04,440
Speaker 2: do you think are some of the big developments we're

245
00:16:04,560 --> 00:16:07,240
Speaker 2: likely to see in the realm of AI models?

246
00:16:07,760 --> 00:16:12,080
Speaker 3: Very good question. I feel like the generative AI of

247
00:16:12,160 --> 00:16:17,280
Speaker 3: the past was powered by large language models. The generative

248
00:16:17,360 --> 00:16:21,640
Speaker 3: AI of the future is going to reason, plan, act

249
00:16:22,040 --> 00:16:22,800
Speaker 3: and reflect.

250
00:16:23,320 --> 00:16:26,720
Speaker 2: Huh, and so I mean in the context of Granite

251
00:16:26,920 --> 00:16:30,360
Speaker 2: in particular, like, what are we likely to see both

252
00:16:30,520 --> 00:16:32,400
Speaker 2: you know, in the near term and in the sort

253
00:16:32,440 --> 00:16:33,680
Speaker 2: of medium to long term.

254
00:16:34,320 --> 00:16:38,600
Speaker 3: There are multiple elements to implement an agentic workflow that

255
00:16:38,640 --> 00:16:42,160
Speaker 3: I just mentioned. One element of that is the LLM

256
00:16:42,240 --> 00:16:46,360
Speaker 3: itself to be able to do the planning and reasoning

257
00:16:46,440 --> 00:16:50,800
Speaker 3: and acting and doing something that we call tool calling.

258
00:16:51,200 --> 00:16:54,800
Speaker 3: So basically, a series of tools are available to the model.

259
00:16:55,360 --> 00:16:58,240
Speaker 3: You ask the model to call those and make a call.

260
00:16:58,400 --> 00:17:01,560
Speaker 3: For example, we can say, hey, Granted, what is the

261
00:17:01,600 --> 00:17:07,320
Speaker 3: weather like where Jacob lives. It's connect to web search API,

262
00:17:07,880 --> 00:17:10,639
Speaker 3: look up your location. Then it's going to connect to

263
00:17:11,119 --> 00:17:15,440
Speaker 3: weather API, calculate the weather and come back and formulate

264
00:17:15,480 --> 00:17:19,040
Speaker 3: an answer and respond to that. So during this process,

265
00:17:19,600 --> 00:17:22,119
Speaker 3: it first has to plan the task of how to

266
00:17:22,119 --> 00:17:25,000
Speaker 3: answer that question, look into what are the tools that

267
00:17:25,040 --> 00:17:27,720
Speaker 3: are available to it, and call them, and that's an

268
00:17:27,720 --> 00:17:30,400
Speaker 3: ability of the model to do that. What we did

269
00:17:30,440 --> 00:17:34,560
Speaker 3: with Granted was we expanded the granite capabilities to be

270
00:17:34,600 --> 00:17:38,240
Speaker 3: able to do function callings. So for example, today we

271
00:17:38,600 --> 00:17:41,679
Speaker 3: have an open source granted to an eb function calling

272
00:17:41,760 --> 00:17:44,719
Speaker 3: that is available on hugging face to try on and

273
00:17:44,760 --> 00:17:47,360
Speaker 3: you can grab the model and the model has capability

274
00:17:47,400 --> 00:17:50,679
Speaker 3: to do the tool calling. I'm anticipating that in the

275
00:17:50,760 --> 00:17:55,000
Speaker 3: near future the planning and reasoning and acting and reflecting

276
00:17:55,040 --> 00:17:58,120
Speaker 3: capabilities of the large language models are going to continue

277
00:17:58,160 --> 00:17:58,639
Speaker 3: to evolve.

278
00:18:00,119 --> 00:18:04,080
Speaker 2: So thinking now from the point of view of buyers

279
00:18:04,119 --> 00:18:07,280
Speaker 2: and users of AIS, really people who are listening kind

280
00:18:07,520 --> 00:18:14,200
Speaker 2: from that perspective, as people are evaluating AI tools and solutions,

281
00:18:14,840 --> 00:18:17,720
Speaker 2: what is the most important thing they should be thinking about?

282
00:18:17,800 --> 00:18:20,240
Speaker 2: How do you think about kind of that process?

283
00:18:21,320 --> 00:18:24,560
Speaker 3: I think they should always start with the area at

284
00:18:24,680 --> 00:18:28,760
Speaker 3: which they think it would benefit from AI, and then

285
00:18:29,080 --> 00:18:33,080
Speaker 3: within that area, look into what data they have available

286
00:18:33,240 --> 00:18:37,440
Speaker 3: to potentially fit into those AI service architects do they

287
00:18:37,440 --> 00:18:41,000
Speaker 3: have access to quality data? And the second question that

288
00:18:41,040 --> 00:18:42,919
Speaker 3: they have to ask themselves is do I have a

289
00:18:42,960 --> 00:18:46,919
Speaker 3: trusted partner that can supply what I need to be

290
00:18:46,960 --> 00:18:50,679
Speaker 3: able to implement AI. That can be a collection of

291
00:18:50,720 --> 00:18:53,280
Speaker 3: the foundation models that you're going to need, that can

292
00:18:53,320 --> 00:18:57,360
Speaker 3: be a collection of the platform capabilities that the trusted

293
00:18:57,400 --> 00:19:01,240
Speaker 3: partner can offer you to implement such a The thirting

294
00:19:01,560 --> 00:19:06,919
Speaker 3: is go and evaluate the regulations. Does regulation allow you

295
00:19:07,000 --> 00:19:11,520
Speaker 3: to applyoy AI to the specific area that you are

296
00:19:11,680 --> 00:19:15,320
Speaker 3: investigating and you're targeting for AI. And the last part,

297
00:19:15,680 --> 00:19:18,919
Speaker 3: but not least, is back to the principles of design thinking,

298
00:19:19,080 --> 00:19:23,840
Speaker 3: what is the problem in that area? I'm solving with AI,

299
00:19:24,359 --> 00:19:27,680
Speaker 3: and if AI is even appropriate, because we want to

300
00:19:27,720 --> 00:19:30,000
Speaker 3: make sure that you use AI not just because it's

301
00:19:30,040 --> 00:19:32,840
Speaker 3: a cool, hot toy in the market, but you are

302
00:19:33,040 --> 00:19:37,679
Speaker 3: convinced that it can significantly enhance the user experience of

303
00:19:37,720 --> 00:19:40,760
Speaker 3: your customers in that area. And once you have an

304
00:19:40,760 --> 00:19:43,639
Speaker 3: answer to those all these four questions, then maybe you

305
00:19:43,680 --> 00:19:46,360
Speaker 3: have a good candidates to start applying aiit.

306
00:19:47,320 --> 00:19:51,040
Speaker 2: And what about from the side of project managers who

307
00:19:51,040 --> 00:19:54,080
Speaker 2: are trying to just keep up with how fast things

308
00:19:54,119 --> 00:19:58,439
Speaker 2: are changing, how fast innovation is happening, Like, what advice

309
00:19:58,440 --> 00:19:59,639
Speaker 2: would you give those people?

310
00:20:00,280 --> 00:20:04,520
Speaker 3: My advice would be focused on agility. This is a

311
00:20:04,560 --> 00:20:08,240
Speaker 3: market that is evolving rapidly and the winners of the

312
00:20:08,320 --> 00:20:11,800
Speaker 3: market would be those that are able to take advantage

313
00:20:11,800 --> 00:20:15,040
Speaker 3: of the best the market can offer at any point

314
00:20:15,080 --> 00:20:18,040
Speaker 3: of time. So in order to do that, they need

315
00:20:18,119 --> 00:20:26,359
Speaker 3: to be open to experimentation, continuous learning, and to rapidly

316
00:20:26,680 --> 00:20:28,240
Speaker 3: adopting the new ideas.

317
00:20:29,440 --> 00:20:32,919
Speaker 2: And when you think about the future and GENAI, is

318
00:20:32,960 --> 00:20:36,840
Speaker 2: there a particular, say problem that you are most excited

319
00:20:36,880 --> 00:20:37,399
Speaker 2: to solve.

320
00:20:38,080 --> 00:20:40,879
Speaker 3: I think that would be productivity. If you look into

321
00:20:41,000 --> 00:20:44,400
Speaker 3: the stats that are out there, there are surveys that

322
00:20:44,680 --> 00:20:48,359
Speaker 3: confirm that sixty to seventy percents of the time of

323
00:20:48,400 --> 00:20:54,359
Speaker 3: our employees can be potentially enhanced to the productivity gains

324
00:20:54,359 --> 00:20:57,800
Speaker 3: of generative I for example, I personally myself use my

325
00:20:57,880 --> 00:21:01,399
Speaker 3: product for content generation a lot, so the time that

326
00:21:01,440 --> 00:21:06,440
Speaker 3: it frees up can be potentially put into generating a

327
00:21:06,520 --> 00:21:10,720
Speaker 3: higher value work. And because of that, I'm super excited

328
00:21:10,840 --> 00:21:15,280
Speaker 3: with all the opportunities that it represents for enterprises to

329
00:21:15,720 --> 00:21:18,840
Speaker 3: go and dedicate the time of their employees to higher

330
00:21:18,920 --> 00:21:20,000
Speaker 3: value items.

331
00:21:20,240 --> 00:21:24,800
Speaker 2: Great. Okay, a couple of Granite specific questions. So what

332
00:21:24,880 --> 00:21:27,639
Speaker 2: are like the key things you want the world to

333
00:21:27,760 --> 00:21:29,640
Speaker 2: know about Granite.

334
00:21:29,680 --> 00:21:35,680
Speaker 3: Granite is open, trusted, and targeted. Two ways to think

335
00:21:35,720 --> 00:21:40,199
Speaker 3: about openness. One open as open weights it's available for

336
00:21:40,280 --> 00:21:44,479
Speaker 3: public to download, and the second one is open as

337
00:21:44,560 --> 00:21:49,439
Speaker 3: in there is less restrictions on how the customers can

338
00:21:49,560 --> 00:21:52,640
Speaker 3: legally use these models for a range of use cases.

339
00:21:52,760 --> 00:21:56,119
Speaker 3: We have released Granite open source models on their Apache

340
00:21:56,320 --> 00:22:00,200
Speaker 3: license that is enabling a large range of use cases.

341
00:22:00,600 --> 00:22:03,800
Speaker 3: The second one was trusted. We talked about that like

342
00:22:03,880 --> 00:22:08,119
Speaker 3: it's rooted in the trustworthy governance process that we established

343
00:22:08,119 --> 00:22:12,119
Speaker 3: thereund how we are training these models and the responsibility

344
00:22:12,160 --> 00:22:14,640
Speaker 3: that we take for these models. And the third one

345
00:22:14,680 --> 00:22:19,160
Speaker 3: is targeted, targeted for enterprise. We talked about like exposing

346
00:22:19,160 --> 00:22:23,520
Speaker 3: Granted to enterprise data or the domain specific Granted some

347
00:22:23,600 --> 00:22:26,960
Speaker 3: of them like Cobalt Java Translation that is targeting to

348
00:22:27,160 --> 00:22:32,200
Speaker 3: solve specific enterprise needs. And that's Granite, so open, trusted

349
00:22:32,280 --> 00:22:32,920
Speaker 3: and targeted.

350
00:22:33,640 --> 00:22:35,439
Speaker 2: So there are a lot of models out in the

351
00:22:35,480 --> 00:22:38,600
Speaker 2: world all of a sudden, right, it's a crowded market.

352
00:22:39,200 --> 00:22:42,040
Speaker 2: Where does Granite fit in that universe? What is the

353
00:22:42,080 --> 00:22:42,960
Speaker 2: market for granted?

354
00:22:43,960 --> 00:22:47,840
Speaker 3: We talked about the enterprise market shifting away from very

355
00:22:47,960 --> 00:22:53,480
Speaker 3: large general purpose models to targeted, smaller models, and Granted

356
00:22:53,960 --> 00:22:58,159
Speaker 3: is a small model that enterprise can pick up and

357
00:22:58,320 --> 00:23:03,159
Speaker 3: customize on their proprietary data to create something that is

358
00:23:03,200 --> 00:23:07,679
Speaker 3: differentiated for a target use case. So Granted is alsuited

359
00:23:07,760 --> 00:23:12,520
Speaker 3: as a small domain specific business ready tailored for business

360
00:23:12,760 --> 00:23:17,159
Speaker 3: and trained on enterprise data to solve enterprise questions.

361
00:23:17,560 --> 00:23:20,440
Speaker 2: You mentioned small as one of the things that granted

362
00:23:20,560 --> 00:23:25,600
Speaker 2: is why is that useful in some contexts for enterprise

363
00:23:25,680 --> 00:23:26,720
Speaker 2: for businesses.

364
00:23:27,520 --> 00:23:32,000
Speaker 3: The larger the model, the larger computer resources it requires,

365
00:23:32,680 --> 00:23:37,919
Speaker 3: it translates to increased latency. That's your response time. It

366
00:23:37,960 --> 00:23:44,600
Speaker 3: translates to increased cost and in translates to increased carbon

367
00:23:44,600 --> 00:23:49,119
Speaker 3: footprint and energy consumption. So at the scale of enterprise transactions,

368
00:23:49,160 --> 00:23:51,520
Speaker 3: when you move to production and you want to scale,

369
00:23:52,359 --> 00:23:57,560
Speaker 3: some of these challenges can be multiple times stronger. Like

370
00:23:57,640 --> 00:24:00,920
Speaker 3: costs can add up, the energy assumption can be a

371
00:24:01,000 --> 00:24:04,600
Speaker 3: serious thing, and the latency is depending on the application,

372
00:24:05,320 --> 00:24:11,560
Speaker 3: can be a showstopper and blocker because for longer, larger models,

373
00:24:11,560 --> 00:24:15,119
Speaker 3: more powerful models, it just takes a way longer time

374
00:24:15,280 --> 00:24:17,159
Speaker 3: to process and calculate the output.

375
00:24:17,240 --> 00:24:21,080
Speaker 2: For you, we are going to finish up with a

376
00:24:21,119 --> 00:24:25,600
Speaker 2: speed round and I want you to just answer with

377
00:24:25,680 --> 00:24:28,080
Speaker 2: the first thing that comes to mind. Don't overthink this, Okay,

378
00:24:28,359 --> 00:24:31,280
Speaker 2: complete this sentence. In five years, AI.

379
00:24:31,160 --> 00:24:33,600
Speaker 3: Will be invisible.

380
00:24:33,920 --> 00:24:36,200
Speaker 2: Ah, I like that. What do you mean by that?

381
00:24:36,720 --> 00:24:41,240
Speaker 3: Today? AI is everywhere, But if you ask my kids

382
00:24:41,280 --> 00:24:44,600
Speaker 3: at home, they know AI. But if you say very a,

383
00:24:44,680 --> 00:24:46,840
Speaker 3: I like, how do you use AI? They don't know

384
00:24:46,880 --> 00:24:51,320
Speaker 3: the answer because it's so blended in their life that

385
00:24:51,359 --> 00:24:55,159
Speaker 3: they don't feel like it's something that they are using.

386
00:24:55,240 --> 00:24:57,520
Speaker 3: They are getting used to that. So when I think

387
00:24:57,560 --> 00:25:01,800
Speaker 3: of next generation and the years to that generation is

388
00:25:01,880 --> 00:25:06,400
Speaker 3: so used to AI being part of their life that

389
00:25:06,440 --> 00:25:09,199
Speaker 3: they feel like it's just there. That's one, and the

390
00:25:09,240 --> 00:25:12,679
Speaker 3: second one is the simplicity of interaction with AI, that

391
00:25:12,760 --> 00:25:15,680
Speaker 3: you don't feel like you're interacting with the system. It's

392
00:25:15,880 --> 00:25:19,159
Speaker 3: just there. Like you talk to AI. Everything is automated.

393
00:25:19,280 --> 00:25:23,119
Speaker 3: So I would say the simplicity and being blended to

394
00:25:23,880 --> 00:25:27,240
Speaker 3: solve the right problems is the part that I'm referring

395
00:25:27,280 --> 00:25:31,399
Speaker 3: to as invisible. Like Internet is everywhere and it's invisible.

396
00:25:31,680 --> 00:25:34,240
Speaker 3: But we used to dial in, like you remember the

397
00:25:34,320 --> 00:25:38,720
Speaker 3: dialing zone to connect the Internet. It's gone. Internet is

398
00:25:38,760 --> 00:25:40,480
Speaker 3: completely invisible today.

399
00:25:40,359 --> 00:25:43,080
Speaker 2: Right, Like we used to talk about logging on, right,

400
00:25:43,119 --> 00:25:46,080
Speaker 2: and you don't log on anymore because you're always logged on.

401
00:25:46,760 --> 00:25:48,080
Speaker 3: Yep, you're always connected.

402
00:25:48,200 --> 00:25:52,760
Speaker 2: Yeah. What's the number one thing that people misunderstand about AI?

403
00:25:53,359 --> 00:25:58,159
Speaker 3: AI is an irritable but should not be feared.

404
00:25:59,160 --> 00:26:01,919
Speaker 2: What advice would you give yourself ten years ago to

405
00:26:02,160 --> 00:26:04,040
Speaker 2: better prepare you for today?

406
00:26:05,000 --> 00:26:08,520
Speaker 3: I would say, develop a broad range of skills. Even

407
00:26:08,760 --> 00:26:12,480
Speaker 3: if you think they will not help you today, they

408
00:26:12,520 --> 00:26:14,040
Speaker 3: may be valuable in the future.

409
00:26:14,640 --> 00:26:17,840
Speaker 2: So on the consumer side, right now we hear a

410
00:26:17,840 --> 00:26:23,160
Speaker 2: lot about chatbots and image generators. But on the business side,

411
00:26:23,200 --> 00:26:26,280
Speaker 2: what do you think is the next big business application.

412
00:26:26,280 --> 00:26:28,840
Speaker 3: AI influencers generating content.

413
00:26:29,280 --> 00:26:31,800
Speaker 2: Huh. How do you use AI in your day to

414
00:26:31,880 --> 00:26:32,600
Speaker 2: day life today?

415
00:26:33,480 --> 00:26:37,560
Speaker 3: One simple example is LinkedIn posts. I love it to

416
00:26:37,680 --> 00:26:40,000
Speaker 3: just go to my product. I'll give you an example

417
00:26:40,040 --> 00:26:43,000
Speaker 3: which is my favorite one. Lama three point one four

418
00:26:43,040 --> 00:26:46,520
Speaker 3: A five b the post that I announced on LinkedIn

419
00:26:46,720 --> 00:26:49,720
Speaker 3: on hey, IBM is releasing the model on the same

420
00:26:49,800 --> 00:26:52,760
Speaker 3: day it was generated by Lama three point one four

421
00:26:52,760 --> 00:26:55,879
Speaker 3: A five billion. So using the same model to post

422
00:26:56,000 --> 00:26:59,760
Speaker 3: the generate the announcement note very elegant.

423
00:27:00,359 --> 00:27:01,920
Speaker 2: Is there anything else I should ask you?

424
00:27:02,359 --> 00:27:05,320
Speaker 3: Oh, we didn't talk about instruct lab. So when you

425
00:27:05,400 --> 00:27:08,280
Speaker 3: grab a model, you start from the model, but you

426
00:27:08,400 --> 00:27:13,399
Speaker 3: need to then customize it on your proprietary data to

427
00:27:13,440 --> 00:27:17,080
Speaker 3: create value on top of that. So instruct lab is

428
00:27:17,119 --> 00:27:23,600
Speaker 3: giving you a method based on open source contributions to

429
00:27:23,640 --> 00:27:29,880
Speaker 3: collectively contribute to improve the base model. So if you're

430
00:27:29,920 --> 00:27:36,080
Speaker 3: an enterprise, you can leverage your internal employees to collectively

431
00:27:36,200 --> 00:27:39,960
Speaker 3: all contribute to improve the model. And I'll give you

432
00:27:39,960 --> 00:27:42,199
Speaker 3: an example of why it matters. Like if you go

433
00:27:42,240 --> 00:27:45,160
Speaker 3: to higging face today and look for Lama, there are

434
00:27:45,359 --> 00:27:49,479
Speaker 3: about fifty thousand different lamas coming up and the reason

435
00:27:49,680 --> 00:27:52,240
Speaker 3: is because there is no way to contribute to the

436
00:27:52,280 --> 00:27:55,360
Speaker 3: base model. If you're a developer, you have to make

437
00:27:55,400 --> 00:27:57,720
Speaker 3: a colon of the copy of the model and find

438
00:27:57,800 --> 00:28:00,720
Speaker 3: you need for your own purpose. We are the method

439
00:28:00,760 --> 00:28:04,679
Speaker 3: that we call instruct lab to be able to collectively

440
00:28:04,920 --> 00:28:08,320
Speaker 3: collect all that information and contribute to the base model

441
00:28:08,359 --> 00:28:12,159
Speaker 3: and enhance. So that's instruct lab. I just wanted to

442
00:28:12,240 --> 00:28:16,359
Speaker 3: highlight the value of being open because that's another topic

443
00:28:16,400 --> 00:28:18,800
Speaker 3: that has been emerging in the market over the past

444
00:28:18,840 --> 00:28:22,240
Speaker 3: eighteen months. In particular, I believe the future of AI

445
00:28:22,400 --> 00:28:26,119
Speaker 3: is open, and we've been seeing how the open source

446
00:28:26,240 --> 00:28:31,359
Speaker 3: markets has been changing, how the models are accessible to

447
00:28:31,400 --> 00:28:35,960
Speaker 3: a wider audience, and good things typically happen when you

448
00:28:36,400 --> 00:28:40,240
Speaker 3: make technology pieces accessible to a broader range of community

449
00:28:40,400 --> 00:28:43,800
Speaker 3: to stress test that and that's the direction that we've

450
00:28:43,840 --> 00:28:46,360
Speaker 3: been adopting with granted, and I feel like that's really

451
00:28:46,400 --> 00:28:48,800
Speaker 3: the adoption that the market is going to emerge to

452
00:28:49,240 --> 00:28:49,920
Speaker 3: moving forward.

453
00:28:50,160 --> 00:28:54,719
Speaker 2: Yeah, this interesting, I think, maybe naively unintuitive, but it

454
00:28:54,720 --> 00:28:58,040
Speaker 2: makes sense once you think about it. Thing that open

455
00:28:58,080 --> 00:29:01,280
Speaker 2: source things are safer. Evily think, oh no, put it

456
00:29:01,320 --> 00:29:03,080
Speaker 2: in a box so nobody can see it, and that'll

457
00:29:03,120 --> 00:29:05,160
Speaker 2: be safer. But like it turns out of the world,

458
00:29:05,200 --> 00:29:07,840
Speaker 2: if you let everybody poke at it, the world will

459
00:29:07,840 --> 00:29:10,480
Speaker 2: find the vulnerabilities for you and you can fix them.

460
00:29:10,560 --> 00:29:13,040
Speaker 3: Right, That's exactly what's going to happen. Yeah.

461
00:29:13,800 --> 00:29:16,320
Speaker 2: Great, it was lovely to talk with you. Thank you

462
00:29:16,360 --> 00:29:17,320
Speaker 2: so much for your time.

463
00:29:17,720 --> 00:29:19,840
Speaker 3: The same here, thanks Jacob.

464
00:29:21,320 --> 00:29:23,880
Speaker 1: And that wraps up this episode. A huge thanks to

465
00:29:23,920 --> 00:29:27,760
Speaker 1: Mariam and Jacob. Today's conversation open my eyes as to

466
00:29:27,800 --> 00:29:32,000
Speaker 1: how open technology and AI are intersecting to create more

467
00:29:32,040 --> 00:29:37,360
Speaker 1: transparent and efficient systems for enterprises. From the power of smaller,

468
00:29:37,440 --> 00:29:40,520
Speaker 1: more targeted models like granted to the importance of trust

469
00:29:40,640 --> 00:29:45,480
Speaker 1: and governance in AI, these developments are reshaping how businesses

470
00:29:45,560 --> 00:29:49,560
Speaker 1: operate at their core. As we continue to unpack the

471
00:29:49,640 --> 00:29:54,840
Speaker 1: complexities of artificial intelligence, it's clear that openness, whether in data,

472
00:29:55,280 --> 00:29:59,520
Speaker 1: technology or collaboration, is not just a concept, but a

473
00:29:59,600 --> 00:30:06,000
Speaker 1: driving force that can unlock new possibilities. Smart Talks with

474
00:30:06,040 --> 00:30:09,320
Speaker 1: IBM is produced by Matt Romano, Joey fish Ground, Amy

475
00:30:09,360 --> 00:30:13,440
Speaker 1: Gains McQuaid, and Jacob Goldstein, who are edited by Lydia

476
00:30:13,520 --> 00:30:17,280
Speaker 1: Jean kott Or. Engineers are Sarah Brugerer and Ben Tolliday.

477
00:30:17,760 --> 00:30:20,640
Speaker 1: Theme song by Gramoscope special thanks to the eight Bar

478
00:30:20,760 --> 00:30:24,720
Speaker 1: and IBM teams, as well as the Pushkin marketing team.

479
00:30:25,040 --> 00:30:27,680
Speaker 1: Smart Talks with IBM is a production of Pushkin Industries

480
00:30:27,920 --> 00:30:32,240
Speaker 1: and Ruby Studio at iHeartMedia. To find more Pushkin podcasts,

481
00:30:32,560 --> 00:30:36,040
Speaker 1: listen on the iHeartRadio app, Apple Podcasts, or wherever you

482
00:30:36,160 --> 00:30:42,600
Speaker 1: listen to podcasts. I'm Malcolm Glauwell. This is a paid

483
00:30:42,640 --> 00:30:47,400
Speaker 1: advertisement from IBM. The conversations on this podcast don't necessarily

484
00:30:47,440 --> 00:31:02,960
Speaker 1: represent IBM's positions, strategies, or opinions.