1 00:00:04,440 --> 00:00:12,639 Speaker 1: Welcome to tech Stuff, a production from iHeartRadio. This season 2 00:00:12,720 --> 00:00:16,200 Speaker 1: on smart Talks with IBM, Malcolm Gladwell and team are 3 00:00:16,239 --> 00:00:19,479 Speaker 1: diving into the transformative world of artificial intelligence with a 4 00:00:19,520 --> 00:00:24,120 Speaker 1: fresh perspective on the concept of open What does open 5 00:00:24,560 --> 00:00:27,920 Speaker 1: really mean in the context of AI. It can mean 6 00:00:28,000 --> 00:00:31,600 Speaker 1: open source code or open data, but it also encompasses 7 00:00:31,720 --> 00:00:36,040 Speaker 1: fostering an ecosystem of ideas, ensuring diverse perspectives are heard, 8 00:00:36,159 --> 00:00:39,919 Speaker 1: and enabling new levels of transparency. Join hosts from your 9 00:00:39,960 --> 00:00:43,560 Speaker 1: favorite Pushkin podcasts as they explore how openness and AI 10 00:00:43,800 --> 00:00:49,040 Speaker 1: is reshaping industries, driving innovation, and redefining what's possible. You'll 11 00:00:49,080 --> 00:00:52,360 Speaker 1: hear from industry experts and leaders about the implications and 12 00:00:52,400 --> 00:00:56,160 Speaker 1: possibilities of open AI, and of course, Malcolm Gladwell will 13 00:00:56,200 --> 00:00:58,200 Speaker 1: be there to guide you through the season with his 14 00:00:58,360 --> 00:01:01,840 Speaker 1: unique insights. Look out for new episodes of Smart Talks 15 00:01:01,920 --> 00:01:05,360 Speaker 1: every other week on the iHeartRadio app, Apple Podcasts, or 16 00:01:05,360 --> 00:01:08,880 Speaker 1: wherever you get your podcasts, and learn more at IBM 17 00:01:09,080 --> 00:01:20,920 Speaker 1: dot com, slash smart Talks. 18 00:01:18,440 --> 00:01:26,280 Speaker 2: Pushkin Hello, Hello, Welcome to Smart Talks with IBM, a 19 00:01:26,319 --> 00:01:31,640 Speaker 2: podcast from Pushkin Industries, iHeartRadio and IBM. I'm Malcolm Glawell. 20 00:01:32,280 --> 00:01:35,800 Speaker 2: This season, we're diving back into the world of artificial intelligence, 21 00:01:36,120 --> 00:01:39,119 Speaker 2: but with a focus on the powerful concept of open 22 00:01:39,640 --> 00:01:45,119 Speaker 2: its possibilities, implications, and misconceptions. We'll look at openness from 23 00:01:45,200 --> 00:01:47,720 Speaker 2: a variety of angles and explore how the concept is 24 00:01:47,760 --> 00:01:52,080 Speaker 2: already reshaping industries, ways of doing business and our very 25 00:01:52,120 --> 00:01:57,280 Speaker 2: notion of what's possible. In today's episode, Jacob Goldstein sat 26 00:01:57,320 --> 00:02:01,440 Speaker 2: down with Mariam Ashuri, the Director of Product Management and 27 00:02:01,480 --> 00:02:05,240 Speaker 2: a Head of product for IBM's Watson x dot AI, 28 00:02:05,760 --> 00:02:09,680 Speaker 2: where she spearheads the product strategy and delivery of IBM's 29 00:02:09,720 --> 00:02:14,400 Speaker 2: watsonx foundation models. She is a technologist with more than 30 00:02:14,440 --> 00:02:19,960 Speaker 2: fifteen years of experience developing data driven technologies. The conversation 31 00:02:20,080 --> 00:02:24,200 Speaker 2: focused on how enterprises can use technology to build and 32 00:02:24,280 --> 00:02:30,240 Speaker 2: deliver greater transparency in AI. With Granite, Mariam explained how 33 00:02:30,240 --> 00:02:34,640 Speaker 2: Grantite can be utilized to improve efficiency across various domains. 34 00:02:35,280 --> 00:02:38,560 Speaker 2: She discussed how these models are being used in real 35 00:02:38,600 --> 00:02:43,800 Speaker 2: world business applications, particularly in areas like customer care, where 36 00:02:43,840 --> 00:02:48,920 Speaker 2: AI can help enable quick, accurate responses based on internal 37 00:02:49,160 --> 00:02:54,280 Speaker 2: company data. Mariam provided a fascinating look into how enterprises 38 00:02:54,320 --> 00:02:59,919 Speaker 2: have moved from mere experimentation with generative AI to actual production, 39 00:03:01,000 --> 00:03:05,880 Speaker 2: navigating challenges such as increased latency, cost and energy consumption. 40 00:03:06,880 --> 00:03:11,160 Speaker 2: She highlighted how the emerging trend of smaller models customized 41 00:03:11,200 --> 00:03:15,520 Speaker 2: with proprietary data can potentially deliver high performance at a 42 00:03:15,560 --> 00:03:19,440 Speaker 2: fraction of the cost, marking a significant shift in how 43 00:03:19,680 --> 00:03:24,160 Speaker 2: enterprises leverage AI. Whether you're an AI enthusiast or a 44 00:03:24,160 --> 00:03:27,640 Speaker 2: business leader looking to harness the power of artificial intelligence, 45 00:03:28,200 --> 00:03:34,200 Speaker 2: this episode is packed with valuable insights and forward thinking strategies. 46 00:03:38,520 --> 00:03:40,480 Speaker 3: Let's just start with your background. How did you come 47 00:03:40,520 --> 00:03:41,640 Speaker 3: to work at IBM. 48 00:03:42,360 --> 00:03:45,840 Speaker 4: I join IBM right after I graduated. I have an 49 00:03:45,840 --> 00:03:51,920 Speaker 4: AI background, and throughout the years, I've held many roles 50 00:03:51,960 --> 00:03:57,440 Speaker 4: in design, engineering, development research, mostly focused on AI application 51 00:03:57,560 --> 00:04:01,680 Speaker 4: development and design. In my current job, I'm the product 52 00:04:01,680 --> 00:04:04,840 Speaker 4: owner for What's the Next Day I, which is the 53 00:04:05,800 --> 00:04:10,280 Speaker 4: IBM platform for enterprise AI. What excites me about this job, 54 00:04:10,320 --> 00:04:14,120 Speaker 4: I would say, is the technology advancements over the last 55 00:04:14,120 --> 00:04:17,320 Speaker 4: eighteen months in the market. We've been witnessing how generative 56 00:04:17,360 --> 00:04:20,040 Speaker 4: ALI has been changing the market. The way that I 57 00:04:20,120 --> 00:04:23,440 Speaker 4: see that is JENNAI has been perhaps one of the 58 00:04:23,560 --> 00:04:27,360 Speaker 4: largest paradigm shifts. When we think about productivity the same 59 00:04:27,400 --> 00:04:32,800 Speaker 4: way that Internet and personal computers impacted the productivity of workforce. 60 00:04:32,960 --> 00:04:37,840 Speaker 4: Now we are witnessing another wave of all those opportunities 61 00:04:37,880 --> 00:04:41,120 Speaker 4: that it can unlock for especially enterprise AI when it 62 00:04:41,160 --> 00:04:45,240 Speaker 4: comes to enhancing the productivity of the workforce and releasing 63 00:04:45,320 --> 00:04:49,960 Speaker 4: some time that can potentially be put into creating more 64 00:04:50,080 --> 00:04:54,480 Speaker 4: value work for enterprise. So that's the major part that 65 00:04:54,640 --> 00:04:58,480 Speaker 4: I picked this team to have an impact on the 66 00:04:58,520 --> 00:05:02,560 Speaker 4: market and the community, but also of course using the 67 00:05:03,160 --> 00:05:06,080 Speaker 4: skills that I gain through all these years through IBM 68 00:05:06,279 --> 00:05:10,240 Speaker 4: to help to establish IBM as the market leader for 69 00:05:10,400 --> 00:05:11,159 Speaker 4: enterprise AI. 70 00:05:11,720 --> 00:05:15,800 Speaker 3: So you talked about jenai as this sort of generational, 71 00:05:15,960 --> 00:05:21,000 Speaker 3: transformational technological force, and I'm curious just in terms of 72 00:05:21,440 --> 00:05:23,920 Speaker 3: how it's going to come into the world, Like, how 73 00:05:23,920 --> 00:05:28,000 Speaker 3: do you see market adoption of genai sort of evolving 74 00:05:28,040 --> 00:05:28,520 Speaker 3: from here? 75 00:05:29,320 --> 00:05:32,880 Speaker 4: Well, last year was the year of excitement about generative AI. 76 00:05:33,040 --> 00:05:36,159 Speaker 4: Most of the companies were experimenting and exploring with GENI. 77 00:05:36,880 --> 00:05:40,440 Speaker 4: We see that energy shifted towards how to best monetize 78 00:05:40,480 --> 00:05:43,360 Speaker 4: that technology. Almost half of the market has moved from 79 00:05:43,640 --> 00:05:48,839 Speaker 4: investigation to pilots, ten percent has moved to production. When 80 00:05:48,880 --> 00:05:52,520 Speaker 4: you're exploring with this technology, you're looking for a valve factor, 81 00:05:52,800 --> 00:05:56,320 Speaker 4: You're looking for an AHA moment. That's why very large 82 00:05:56,320 --> 00:06:00,919 Speaker 4: general purpose models shine. But as companies move toward production 83 00:06:01,040 --> 00:06:03,960 Speaker 4: and scale, they soon realize the past success is not 84 00:06:04,000 --> 00:06:09,039 Speaker 4: that straightforward. For example, they're larger the model, the larger 85 00:06:09,120 --> 00:06:13,680 Speaker 4: computer resources it requires. That translates to increased latency that's 86 00:06:13,680 --> 00:06:18,080 Speaker 4: your response time. That translates to increased cost. That translates 87 00:06:18,120 --> 00:06:21,680 Speaker 4: to increase carbon food print, and energy consumption. So think 88 00:06:21,720 --> 00:06:25,360 Speaker 4: about that. At the scale of enterprise in production, some 89 00:06:25,440 --> 00:06:28,880 Speaker 4: of them can be a showstopper. Because of this reason, 90 00:06:29,160 --> 00:06:33,640 Speaker 4: what actually c is emerging in the market is instead 91 00:06:33,680 --> 00:06:39,520 Speaker 4: of focusing on very large general purpose models, coming back 92 00:06:39,600 --> 00:06:45,280 Speaker 4: to very small, trustworthy models that they can customize on 93 00:06:45,360 --> 00:06:48,920 Speaker 4: their own proprietary data. That's the data about their customers, 94 00:06:48,960 --> 00:06:52,719 Speaker 4: that the data about their specific domains to create something 95 00:06:52,800 --> 00:06:57,400 Speaker 4: differentiated that is much smaller and delivers the performance that 96 00:06:57,440 --> 00:07:00,760 Speaker 4: they want on a target use is for a fraction 97 00:07:00,839 --> 00:07:01,360 Speaker 4: of the cost. 98 00:07:01,760 --> 00:07:05,640 Speaker 3: Uh huh. So let's talk a little bit more specifically 99 00:07:05,680 --> 00:07:09,800 Speaker 3: about what you're working on. Let's talk about granite. First 100 00:07:09,800 --> 00:07:11,480 Speaker 3: of all, tell me what is granite. 101 00:07:12,040 --> 00:07:17,560 Speaker 4: Granite is our industrial leading family of models, flagship IBM models. 102 00:07:18,320 --> 00:07:22,160 Speaker 4: These are the models that we train from scratch. When 103 00:07:22,240 --> 00:07:25,640 Speaker 4: offered to our platform, we offer indemnification and we stand 104 00:07:25,640 --> 00:07:31,360 Speaker 4: behind them today. It comes in four flavors, language, code, 105 00:07:32,040 --> 00:07:39,520 Speaker 4: time series and geospecial models. Granite language series is covering English, Spanish, German, 106 00:07:39,960 --> 00:07:44,800 Speaker 4: Portuguese and Japanese. We have a combination of commercial and 107 00:07:45,000 --> 00:07:48,960 Speaker 4: open source language models on Granite. For example, we recently 108 00:07:49,120 --> 00:07:54,280 Speaker 4: released the Granite seven B language model small powerful English model. 109 00:07:55,040 --> 00:07:58,360 Speaker 4: On the code front, our models are state of the 110 00:07:58,480 --> 00:08:01,640 Speaker 4: art models ranging from from three billion to thirty four 111 00:08:01,680 --> 00:08:07,560 Speaker 4: billion parameters. These are very powerful models that performs or 112 00:08:07,680 --> 00:08:11,560 Speaker 4: outperforms in some cases the popular open source models in 113 00:08:11,600 --> 00:08:13,840 Speaker 4: their weight class. So very powerful models. 114 00:08:14,040 --> 00:08:16,720 Speaker 3: So I get the idea a big picture about these models, 115 00:08:16,760 --> 00:08:18,440 Speaker 3: but it would be helpful to just get a sense 116 00:08:18,480 --> 00:08:20,600 Speaker 3: specifically of what they're doing, Like, can you give me 117 00:08:20,640 --> 00:08:24,280 Speaker 3: any specific examples of how these models are being used 118 00:08:25,080 --> 00:08:27,360 Speaker 3: in businesses in the real world right now? 119 00:08:28,520 --> 00:08:31,800 Speaker 4: Well, the top use cases for generative AI are really 120 00:08:31,880 --> 00:08:38,760 Speaker 4: content generation, summarization, information extraction. Perhaps the most popular use 121 00:08:38,800 --> 00:08:42,480 Speaker 4: case that we are seeing in enterprise is content grounded 122 00:08:42,559 --> 00:08:46,800 Speaker 4: question and answering, So using these models as a base 123 00:08:47,080 --> 00:08:49,960 Speaker 4: to connect them to a body of information let's say 124 00:08:50,000 --> 00:08:54,320 Speaker 4: their policies, their documents that is internal to the enterprise, 125 00:08:54,600 --> 00:08:58,880 Speaker 4: and get the model to provide answers based on that questions. 126 00:08:59,200 --> 00:09:03,160 Speaker 4: One example of that is for customer agents customer care, 127 00:09:03,559 --> 00:09:08,120 Speaker 4: when a customer is asking a question. Previously, the agent 128 00:09:08,200 --> 00:09:11,720 Speaker 4: that responds to the customer had to answer the question 129 00:09:11,840 --> 00:09:14,600 Speaker 4: and if they don't know the answer escalated to the product, 130 00:09:14,760 --> 00:09:18,559 Speaker 4: especially keeping people on hold on the line to go 131 00:09:18,640 --> 00:09:21,360 Speaker 4: figure out the answer for that and then come back. 132 00:09:21,440 --> 00:09:23,800 Speaker 4: You can think of the time it takes to resolve 133 00:09:23,840 --> 00:09:27,520 Speaker 4: an issue. But now with LLMS, we have an opportunity 134 00:09:27,559 --> 00:09:31,600 Speaker 4: to automatically retrieve the information based on the internal documents 135 00:09:31,640 --> 00:09:34,640 Speaker 4: of the company, formulate an answer, show it to the 136 00:09:34,720 --> 00:09:38,080 Speaker 4: human agent, and then if they verify with the sources 137 00:09:38,080 --> 00:09:40,800 Speaker 4: of varies coming from, they can just translate it directly 138 00:09:40,840 --> 00:09:44,959 Speaker 4: to the customer. This is a very simple example of 139 00:09:45,000 --> 00:09:46,840 Speaker 4: how it's impacting the customer care. 140 00:09:47,320 --> 00:09:51,800 Speaker 3: So one big theme of this season is this idea 141 00:09:51,800 --> 00:09:55,200 Speaker 3: of open And one of the things that's interesting to 142 00:09:55,320 --> 00:09:59,120 Speaker 3: me about the work you're doing is you are using 143 00:09:59,200 --> 00:10:02,440 Speaker 3: not only ran at this model IBM developed, but you're 144 00:10:02,520 --> 00:10:06,360 Speaker 3: also using third party models right from other places. So 145 00:10:06,400 --> 00:10:08,520 Speaker 3: tell me about that work and how that is sort 146 00:10:08,559 --> 00:10:11,840 Speaker 3: of fitting into your kind of real world typically enterprise 147 00:10:11,960 --> 00:10:12,760 Speaker 3: GENAI work. 148 00:10:13,400 --> 00:10:16,840 Speaker 4: When it comes to model strategy, our strategy is really 149 00:10:16,880 --> 00:10:21,199 Speaker 4: focused on two pillars, multimodel and multi deployment. It means 150 00:10:21,240 --> 00:10:24,280 Speaker 4: that we don't believe one single model rules all the 151 00:10:24,400 --> 00:10:27,000 Speaker 4: use cases. And I think at this point the market 152 00:10:27,000 --> 00:10:30,320 Speaker 4: has also realized the enterprise markets in average today are 153 00:10:30,400 --> 00:10:34,680 Speaker 4: using five to ten different models for different use cases. 154 00:10:34,880 --> 00:10:35,840 Speaker 3: Oh interesting. 155 00:10:36,160 --> 00:10:38,439 Speaker 4: So in our portfolio, if you look into what's on 156 00:10:38,480 --> 00:10:42,280 Speaker 4: extradai today, we are offering a large sets of high performing, 157 00:10:42,400 --> 00:10:45,880 Speaker 4: state of the art models coming from open source commercial 158 00:10:45,960 --> 00:10:49,480 Speaker 4: models that we are bringing through our partners and also 159 00:10:49,559 --> 00:10:53,160 Speaker 4: IBM developed models. In addition to all of these, we 160 00:10:53,280 --> 00:10:56,600 Speaker 4: also have an option for bring your own model from 161 00:10:56,679 --> 00:10:59,760 Speaker 4: outside the platform. Let's say you have a custom model 162 00:11:00,160 --> 00:11:02,200 Speaker 4: that you made it yourself, you can bring it to 163 00:11:02,240 --> 00:11:07,320 Speaker 4: the platform and really helping the customers to navigate through 164 00:11:07,440 --> 00:11:11,040 Speaker 4: avoid range of models and pick the right model for 165 00:11:11,120 --> 00:11:15,040 Speaker 4: their target use case. Throughout that, we've been heavily working 166 00:11:15,160 --> 00:11:18,200 Speaker 4: with our partners and you know, this is the market 167 00:11:18,200 --> 00:11:21,440 Speaker 4: that is evolving rapidly. We've been at the forefront of 168 00:11:21,440 --> 00:11:23,680 Speaker 4: a spit to delivery. One example that I like to 169 00:11:23,960 --> 00:11:29,040 Speaker 4: highlight is recently Metal released Lama four or five billion, 170 00:11:29,360 --> 00:11:31,920 Speaker 4: such a powerful model. On the same day that it 171 00:11:32,080 --> 00:11:35,040 Speaker 4: was released to the market, we made it available in 172 00:11:35,120 --> 00:11:38,160 Speaker 4: our platform to our customers the same day, and not 173 00:11:38,240 --> 00:11:40,679 Speaker 4: only we delivered it on the same day. We are 174 00:11:40,679 --> 00:11:45,160 Speaker 4: offering competitive pricing but also flexibility in where to deploy. 175 00:11:45,280 --> 00:11:48,199 Speaker 4: So we are giving an option to enterprise to deploy 176 00:11:48,280 --> 00:11:52,520 Speaker 4: these models on the platform of dage choice, either multi 177 00:11:52,520 --> 00:11:56,400 Speaker 4: cloud it can be gcpaws as youre IBM cloud, or 178 00:11:56,440 --> 00:12:02,320 Speaker 4: on premises the same for mistrall Ai MISTROLEI recently released 179 00:12:02,320 --> 00:12:05,320 Speaker 4: the model Misrol large two on the same day we 180 00:12:05,440 --> 00:12:08,360 Speaker 4: delivered that through the platform. That's an example of a 181 00:12:08,400 --> 00:12:13,080 Speaker 4: commercial model. Lama as open source, but large two is 182 00:12:13,200 --> 00:12:17,280 Speaker 4: a commercial model that we made available through the platform Great. 183 00:12:18,080 --> 00:12:22,559 Speaker 3: So I want to talk about enterprise grade foundation models, 184 00:12:23,280 --> 00:12:26,200 Speaker 3: just to get into it briefly. What's a foundation model. 185 00:12:26,640 --> 00:12:30,319 Speaker 4: People associate foundation models with a large language model, but 186 00:12:30,480 --> 00:12:33,800 Speaker 4: large language models are really a subset of foundation models. 187 00:12:33,960 --> 00:12:37,880 Speaker 4: Large language models are focused on language, but foundation models 188 00:12:37,920 --> 00:12:41,760 Speaker 4: can be code generators, can be focused on time series 189 00:12:41,760 --> 00:12:44,360 Speaker 4: model we talked about, they can be images, it can 190 00:12:44,400 --> 00:12:48,840 Speaker 4: be jewy special models. So foundation model, as the term 191 00:12:48,960 --> 00:12:54,079 Speaker 4: suggests that your foundations to create a series of subsequent 192 00:12:54,280 --> 00:12:58,640 Speaker 4: models that can be customized for a downstream use case. 193 00:12:58,679 --> 00:13:02,079 Speaker 4: And that's why they are calling that foundation models. LM 194 00:13:02,240 --> 00:13:04,240 Speaker 4: is a good example of that as a subset for 195 00:13:04,400 --> 00:13:08,480 Speaker 4: language that you can further customize on your specific data 196 00:13:08,880 --> 00:13:12,439 Speaker 4: to get the model to do other works. So the 197 00:13:12,480 --> 00:13:16,000 Speaker 4: core of these foundation models, they are basically trained on 198 00:13:16,040 --> 00:13:20,320 Speaker 4: an ab third amount of data data sets that most 199 00:13:20,320 --> 00:13:23,400 Speaker 4: of the institutions today are sourcing them from the internet, 200 00:13:23,559 --> 00:13:26,560 Speaker 4: So you can imagine what can potentially go to those models, 201 00:13:26,600 --> 00:13:29,839 Speaker 4: and then it comes to the enterprise and they start 202 00:13:29,920 --> 00:13:33,840 Speaker 4: using it. So for us also, when we started looking 203 00:13:33,920 --> 00:13:38,800 Speaker 4: into in particular, it was triggered by customers asking us 204 00:13:38,800 --> 00:13:42,280 Speaker 4: to provide client protections on these models, and we started 205 00:13:42,280 --> 00:13:45,679 Speaker 4: thinking about, let's look into how the models are trained 206 00:13:45,679 --> 00:13:49,200 Speaker 4: and if you are comfortable of fering client protections on 207 00:13:49,440 --> 00:13:51,920 Speaker 4: the models that are available in the market. And guess what, 208 00:13:52,400 --> 00:13:55,480 Speaker 4: For a majority of these models, there is absolutely no 209 00:13:55,640 --> 00:13:59,520 Speaker 4: visibility into what data vent into those models, not much 210 00:13:59,559 --> 00:14:03,840 Speaker 4: transparent into how the model trains and the responsibility lies 211 00:14:03,880 --> 00:14:06,880 Speaker 4: on you as the customers to start using those models. 212 00:14:06,920 --> 00:14:10,720 Speaker 3: So just to be clear, that is presenting like potential risk, 213 00:14:10,880 --> 00:14:14,239 Speaker 3: real potential risk to a company that is using these models. 214 00:14:14,360 --> 00:14:17,360 Speaker 4: It is. It is a potential risk in particular for 215 00:14:17,480 --> 00:14:22,080 Speaker 4: the customers in highly regulated industries. So what we did 216 00:14:22,160 --> 00:14:26,240 Speaker 4: for Granite was when we started training these models from scratch, 217 00:14:26,320 --> 00:14:29,360 Speaker 4: basically we went to the corpus of data that was 218 00:14:29,400 --> 00:14:32,920 Speaker 4: available to us. So, for example, the very first version 219 00:14:33,000 --> 00:14:37,160 Speaker 4: of Granite was exposed to twenty persons of its data 220 00:14:37,240 --> 00:14:40,000 Speaker 4: from finance and legal because we have a lot of 221 00:14:40,760 --> 00:14:45,280 Speaker 4: financial institutions as our clients. We worked directly with our 222 00:14:45,360 --> 00:14:50,480 Speaker 4: IBM research to identify detectors for harmful information like haytyp 223 00:14:50,600 --> 00:14:52,239 Speaker 4: use and profanity detectors. 224 00:14:52,800 --> 00:14:55,120 Speaker 3: Okay, so we're talking about Granted, we're talking about this 225 00:14:55,320 --> 00:14:58,640 Speaker 3: set of models IBM has developed. Let's talk about using 226 00:14:58,680 --> 00:15:03,680 Speaker 3: Granite on WATSONEX compared to downloading open source models, Like, 227 00:15:03,720 --> 00:15:04,520 Speaker 3: how do those differ? 228 00:15:05,160 --> 00:15:08,800 Speaker 4: By using Granite and whats on X, you get two things. 229 00:15:09,160 --> 00:15:12,720 Speaker 4: The first one is the client protection and in thementification 230 00:15:12,840 --> 00:15:15,040 Speaker 4: that we talked about, you get that if the model 231 00:15:15,080 --> 00:15:18,840 Speaker 4: is consumed through our platform and the second one is 232 00:15:19,080 --> 00:15:22,880 Speaker 4: really the ecosystem of platform capabilities that we are offering 233 00:15:23,000 --> 00:15:26,200 Speaker 4: to help you create value on top of those data. 234 00:15:26,280 --> 00:15:30,240 Speaker 4: So for example, bringing your data to customize granted for 235 00:15:30,320 --> 00:15:33,640 Speaker 4: your own specific use case. But also one thing that 236 00:15:33,680 --> 00:15:37,160 Speaker 4: I like to highlight in particular is the AI governance. 237 00:15:37,720 --> 00:15:40,400 Speaker 4: So when you get one of these pre train models, 238 00:15:40,640 --> 00:15:43,800 Speaker 4: you put it in front of your own users. Through 239 00:15:43,840 --> 00:15:48,480 Speaker 4: the input and instructions that the user provides for the model, 240 00:15:48,960 --> 00:15:53,200 Speaker 4: they can notdge the model to potentially create undesired behavior 241 00:15:53,520 --> 00:15:56,200 Speaker 4: and change the behavior of the model. And because of 242 00:15:56,240 --> 00:16:00,400 Speaker 4: this is extremely important to automatically document the lead age 243 00:16:00,440 --> 00:16:04,320 Speaker 4: of who touched the model at what point, so if 244 00:16:04,320 --> 00:16:06,760 Speaker 4: something happens, you can trace it back and see where 245 00:16:06,800 --> 00:16:10,760 Speaker 4: it's coming from. And that's what's an extra governance is 246 00:16:10,800 --> 00:16:15,360 Speaker 4: offering automatically documenting the lineage. When you use the granite 247 00:16:15,520 --> 00:16:18,120 Speaker 4: within the platform you get all of those, you can 248 00:16:18,360 --> 00:16:21,760 Speaker 4: have the end to end governance, You can have access 249 00:16:21,800 --> 00:16:25,840 Speaker 4: to all these scalable deployment opportunities that is available for you, 250 00:16:25,920 --> 00:16:28,800 Speaker 4: like to allow you deploy them on the platform of 251 00:16:28,840 --> 00:16:32,320 Speaker 4: your choice that we talked about, either multiple cloud or 252 00:16:32,600 --> 00:16:35,760 Speaker 4: on prem and it also helps you to have access 253 00:16:35,800 --> 00:16:40,440 Speaker 4: to avoid range of model customizations, approaches, prompt tuning, fine tuning, 254 00:16:40,680 --> 00:16:44,000 Speaker 4: retrival augmented generations agents. There is a series of them 255 00:16:44,120 --> 00:16:46,600 Speaker 4: available to use an apply to your model. 256 00:16:47,400 --> 00:16:51,880 Speaker 2: This distinction between large language models and foundation models is 257 00:16:51,920 --> 00:16:56,400 Speaker 2: eye opening. Mariam emphasized that foundation models can be tailored 258 00:16:56,400 --> 00:17:01,400 Speaker 2: to specific tasks, but with that versatility comes a significant 259 00:17:01,480 --> 00:17:05,880 Speaker 2: challenge the lack of transparency and how these models are trained. 260 00:17:06,680 --> 00:17:10,679 Speaker 2: This composed a real risk, especially in highly regulated industries 261 00:17:11,000 --> 00:17:16,440 Speaker 2: like finance. Essentially, by using Granite and watsonex together, enterprises 262 00:17:16,480 --> 00:17:19,200 Speaker 2: get powerful and customizable tools. 263 00:17:20,400 --> 00:17:22,600 Speaker 3: So let's talk about the future a little bit. What 264 00:17:22,680 --> 00:17:24,720 Speaker 3: do you think are some of the big developments we're 265 00:17:24,840 --> 00:17:27,520 Speaker 3: likely to see in the realm of AI models? 266 00:17:28,040 --> 00:17:32,359 Speaker 4: Very good question. I feel like the generative AI of 267 00:17:32,440 --> 00:17:37,560 Speaker 4: the past was powered by large language models. The generative 268 00:17:37,640 --> 00:17:41,919 Speaker 4: AI of the future is going to reason, plan, act 269 00:17:42,280 --> 00:17:43,080 Speaker 4: and reflect. 270 00:17:43,600 --> 00:17:47,000 Speaker 3: Huh, and so I mean in the context of Granite 271 00:17:47,200 --> 00:17:50,639 Speaker 3: in particular, like, what are we likely to see both 272 00:17:50,800 --> 00:17:52,680 Speaker 3: you know, in the near term and in the sort 273 00:17:52,720 --> 00:17:54,000 Speaker 3: of medium to long term. 274 00:17:54,560 --> 00:17:58,879 Speaker 4: There are multiple elements to implement an agentic workflow that 275 00:17:58,920 --> 00:18:02,440 Speaker 4: I just mentioned. One element of that is the LLM 276 00:18:02,520 --> 00:18:06,639 Speaker 4: itself to be able to do the planning and reasoning 277 00:18:06,720 --> 00:18:11,080 Speaker 4: and acting and doing something that we call tool calling. 278 00:18:11,480 --> 00:18:15,080 Speaker 4: So basically, a series of tools are available to the model. 279 00:18:15,640 --> 00:18:18,520 Speaker 4: You ask the model to call those and make a call. 280 00:18:18,680 --> 00:18:21,840 Speaker 4: For example, we can say, hey, Granted, what is the 281 00:18:21,840 --> 00:18:27,600 Speaker 4: weather like where Jacob lives. It's connect to web search API, 282 00:18:28,160 --> 00:18:30,919 Speaker 4: look up your location. Then it's going to connect to 283 00:18:31,359 --> 00:18:35,720 Speaker 4: weather API, calculate the weather and come back and formulate 284 00:18:35,760 --> 00:18:39,320 Speaker 4: an answer and respond to that. So during this process, 285 00:18:39,880 --> 00:18:42,359 Speaker 4: it first has to plan the task of how to 286 00:18:42,400 --> 00:18:45,280 Speaker 4: answer that question, look into what are the tools that 287 00:18:45,320 --> 00:18:48,000 Speaker 4: are available to it and call them, and that's an 288 00:18:48,000 --> 00:18:50,680 Speaker 4: ability of the model to do that. What we did 289 00:18:50,720 --> 00:18:54,800 Speaker 4: with Granted was we expanded the granite capabilities to be 290 00:18:54,880 --> 00:18:58,520 Speaker 4: able to do function calling. So for example, today we 291 00:18:58,880 --> 00:19:01,960 Speaker 4: have an open source granted to an eb function calling 292 00:19:02,040 --> 00:19:05,000 Speaker 4: that is available on hugging face to try on and 293 00:19:05,040 --> 00:19:07,640 Speaker 4: you can grab the model and the model has capability 294 00:19:07,680 --> 00:19:11,000 Speaker 4: to do the tool callings. I'm anticipating that in the 295 00:19:11,040 --> 00:19:15,280 Speaker 4: near future the planning and reasoning and acting and reflecting 296 00:19:15,320 --> 00:19:18,400 Speaker 4: capabilities of the large language models are going to continue 297 00:19:18,440 --> 00:19:18,919 Speaker 4: to evolve. 298 00:19:20,320 --> 00:19:24,359 Speaker 3: So thinking now from the point of view of buyers 299 00:19:24,400 --> 00:19:28,040 Speaker 3: and users of AIS really people who are listening from 300 00:19:28,040 --> 00:19:34,480 Speaker 3: that perspective. As people are evaluating AI tools and solutions, 301 00:19:35,119 --> 00:19:38,000 Speaker 3: what is the most important thing they should be thinking about? 302 00:19:38,080 --> 00:19:40,520 Speaker 3: How do you think about kind of that process? 303 00:19:41,560 --> 00:19:44,880 Speaker 4: I think they should always start with the area at 304 00:19:44,960 --> 00:19:49,040 Speaker 4: which they think it would benefit from AI, and then 305 00:19:49,359 --> 00:19:53,360 Speaker 4: within that area, look into what data they have available 306 00:19:53,520 --> 00:19:57,720 Speaker 4: to potentially fit into those AI service architects do they 307 00:19:57,720 --> 00:20:01,280 Speaker 4: have access to quality data? And the second question that 308 00:20:01,320 --> 00:20:03,199 Speaker 4: they have to ask themselves is do I have a 309 00:20:03,240 --> 00:20:07,199 Speaker 4: trusted partner that can supply what I need to be 310 00:20:07,240 --> 00:20:10,959 Speaker 4: able to implement AI. That can be a collection of 311 00:20:11,000 --> 00:20:13,560 Speaker 4: the foundation models that you're going to need, That can 312 00:20:13,600 --> 00:20:17,640 Speaker 4: be a collection of the platform capabilities that the trusted 313 00:20:17,680 --> 00:20:21,040 Speaker 4: partner can offer you to implement such a thing. The 314 00:20:21,119 --> 00:20:26,240 Speaker 4: third thing is go and evaluate the regulations. Does regulation 315 00:20:26,600 --> 00:20:30,880 Speaker 4: allow you to applyoy AI to the specific area that 316 00:20:31,400 --> 00:20:34,800 Speaker 4: you are investigating and you're targeting for AI. And the 317 00:20:34,920 --> 00:20:38,160 Speaker 4: last part, but not least, is back to the principles 318 00:20:38,200 --> 00:20:41,840 Speaker 4: of design thinking what is the problem in that area 319 00:20:42,320 --> 00:20:46,720 Speaker 4: I'm solving with AI? And if AI is even appropriate? 320 00:20:47,280 --> 00:20:49,280 Speaker 4: Because we want to make sure that you use AI 321 00:20:49,440 --> 00:20:52,320 Speaker 4: not just because it's a cool, hot toy in the market, 322 00:20:52,359 --> 00:20:56,640 Speaker 4: but you are convinced that it can significantly enhance the 323 00:20:56,760 --> 00:21:00,600 Speaker 4: user experience of your customers in that area. And once 324 00:21:00,640 --> 00:21:03,160 Speaker 4: you have an answer to those all these four questions, 325 00:21:03,240 --> 00:21:06,640 Speaker 4: then maybe you have a good candidates to start applying AIT. 326 00:21:07,560 --> 00:21:11,320 Speaker 3: And what about from the side of project managers who 327 00:21:11,320 --> 00:21:14,359 Speaker 3: are trying to just keep up with how fast things 328 00:21:14,400 --> 00:21:18,720 Speaker 3: are changing, how fast innovation is happening, Like, what advice 329 00:21:18,720 --> 00:21:19,919 Speaker 3: would you give those people? 330 00:21:20,520 --> 00:21:24,800 Speaker 4: My advice would be focus on agility. This is a 331 00:21:24,840 --> 00:21:28,520 Speaker 4: market that is evolving rapidly and the winners of the 332 00:21:28,600 --> 00:21:32,080 Speaker 4: market would be those that are able to take advantage 333 00:21:32,080 --> 00:21:35,320 Speaker 4: of the best the market can offer at any point 334 00:21:35,320 --> 00:21:38,320 Speaker 4: of time. So in order to do that, they need 335 00:21:38,359 --> 00:21:46,639 Speaker 4: to be open to experimentation, continuous learning, and to rapidly 336 00:21:46,960 --> 00:21:48,520 Speaker 4: adopting the new ideas. 337 00:21:49,720 --> 00:21:53,159 Speaker 3: And when you think about the future and GENAI, is 338 00:21:53,240 --> 00:21:57,120 Speaker 3: there a particular, say problem that you are most excited 339 00:21:57,160 --> 00:21:57,679 Speaker 3: to solve. 340 00:21:58,359 --> 00:22:01,160 Speaker 4: I think that would be productive. If you look into 341 00:22:01,280 --> 00:22:04,680 Speaker 4: the stats that are out there, there are surveys that 342 00:22:04,960 --> 00:22:08,639 Speaker 4: confirm that sixty to seventy percents of the time of 343 00:22:08,640 --> 00:22:14,639 Speaker 4: our employees can be potentially enhanced to the productivity gains 344 00:22:14,640 --> 00:22:18,600 Speaker 4: of generative For example, I personally myself use my product 345 00:22:18,640 --> 00:22:21,800 Speaker 4: for content generation a lot, so the time that it 346 00:22:21,920 --> 00:22:27,080 Speaker 4: frees up can be potentially put into generating a higher 347 00:22:27,200 --> 00:22:31,199 Speaker 4: value work. And because of that, I'm super excited with 348 00:22:31,680 --> 00:22:36,240 Speaker 4: all the opportunities that it represents for enterprises to go 349 00:22:36,320 --> 00:22:40,280 Speaker 4: and dedicate the time of the employees to higher value items. 350 00:22:40,520 --> 00:22:45,280 Speaker 3: Great. Okay, a couple Granite specific questions. So what are 351 00:22:45,359 --> 00:22:49,280 Speaker 3: like the key things you want the world to know about. 352 00:22:48,960 --> 00:22:55,720 Speaker 4: Granted Granite is open, trusted, and targeted. Two ways to 353 00:22:55,800 --> 00:22:59,760 Speaker 4: think about openness. One open as open weights it's a 354 00:22:59,840 --> 00:23:02,960 Speaker 4: very lebe for public to download. And the second one 355 00:23:03,320 --> 00:23:08,720 Speaker 4: is open as in there is less restrictions on how 356 00:23:08,800 --> 00:23:12,120 Speaker 4: the customers can legally use these models for a range 357 00:23:12,160 --> 00:23:15,440 Speaker 4: of use cases. We have released Grantite open source models 358 00:23:15,440 --> 00:23:19,280 Speaker 4: on their Apache license that is enabling a large range 359 00:23:19,320 --> 00:23:23,240 Speaker 4: of use cases. The second one was trusted. We talked 360 00:23:23,240 --> 00:23:27,480 Speaker 4: about that like it's rooted in the trustworthy governance process 361 00:23:27,520 --> 00:23:30,880 Speaker 4: that we established thereund how we are training these models 362 00:23:31,280 --> 00:23:34,199 Speaker 4: and the responsibility that we take for these models. And 363 00:23:34,280 --> 00:23:38,080 Speaker 4: the third one is targeted, targeted for enterprise. We talked 364 00:23:38,119 --> 00:23:42,320 Speaker 4: about like exposing Granted to enterprise data or the domain 365 00:23:42,440 --> 00:23:46,159 Speaker 4: specific Granted some of them like Cobalt Java Translation that 366 00:23:46,280 --> 00:23:51,320 Speaker 4: is targeting to solve specific enterprise needs and that's granite, 367 00:23:51,359 --> 00:23:53,200 Speaker 4: so open, trusted and targeted. 368 00:23:53,920 --> 00:23:55,720 Speaker 3: So there are a lot of models out in the 369 00:23:55,760 --> 00:23:58,880 Speaker 3: world all of a sudden, right, it's a crowded market. 370 00:23:59,440 --> 00:24:02,320 Speaker 3: Where does it fit in that universe? What is the 371 00:24:02,359 --> 00:24:03,240 Speaker 3: market for granted? 372 00:24:04,240 --> 00:24:08,120 Speaker 4: We talked about the enterprise market shifting away from very 373 00:24:08,240 --> 00:24:13,199 Speaker 4: large general purpose models to target a smaller models, and 374 00:24:13,320 --> 00:24:18,040 Speaker 4: Granted is a small model that enterprise can pick up 375 00:24:18,320 --> 00:24:23,240 Speaker 4: and customize on their proprietary data to create something that 376 00:24:23,359 --> 00:24:27,359 Speaker 4: is differentiated for a target use case. So Granted is 377 00:24:27,400 --> 00:24:32,120 Speaker 4: well suited as a small, domain specific business, ready tailored 378 00:24:32,160 --> 00:24:37,360 Speaker 4: for business and trained on enterprise data to solve enterprise questions. 379 00:24:37,840 --> 00:24:40,720 Speaker 3: You mentioned small as one of the things that granted 380 00:24:40,840 --> 00:24:45,880 Speaker 3: is why is that useful in some contexts for enterprise 381 00:24:45,960 --> 00:24:46,959 Speaker 3: for businesses. 382 00:24:47,800 --> 00:24:52,280 Speaker 4: The larger the model, the larger computer resources it requires. 383 00:24:52,960 --> 00:24:58,199 Speaker 4: It translates to increased latency that's your response time. It 384 00:24:58,240 --> 00:25:04,840 Speaker 4: translates to increase costs, and in translates to increase carbon 385 00:25:04,880 --> 00:25:09,399 Speaker 4: footprint and energy consumption. So at the scale of enterprise transactions. 386 00:25:09,440 --> 00:25:11,959 Speaker 4: When you move to production and you want to scale, 387 00:25:12,640 --> 00:25:17,800 Speaker 4: some of these challenges can be multiple times stronger, like 388 00:25:17,920 --> 00:25:21,200 Speaker 4: costs can add up, the energy consumption can be a 389 00:25:21,280 --> 00:25:24,879 Speaker 4: serious thing, and the latency is depending on the application, 390 00:25:25,560 --> 00:25:31,840 Speaker 4: can be a showstopper and blockier because for longer, larger models, 391 00:25:31,840 --> 00:25:35,359 Speaker 4: more powerful models, it just takes the way longer time 392 00:25:35,560 --> 00:25:37,440 Speaker 4: to process and calculate the output. 393 00:25:37,520 --> 00:25:41,359 Speaker 3: For you, we are going to finish up with a 394 00:25:41,400 --> 00:25:45,879 Speaker 3: speed round and I want you to just answer with 395 00:25:45,960 --> 00:25:48,359 Speaker 3: the first thing that comes to mind. Don't overthink this, Okay, 396 00:25:48,640 --> 00:25:54,440 Speaker 3: complete this sentence. In five years, AI will be invisible. Ah, 397 00:25:54,480 --> 00:25:56,480 Speaker 3: I like that. What do you mean by that? 398 00:25:56,960 --> 00:26:01,520 Speaker 4: Today? AI is everywhere? But if you ask my kids 399 00:26:01,560 --> 00:26:05,240 Speaker 4: at home, they know AI. But if you say very like, 400 00:26:05,280 --> 00:26:07,680 Speaker 4: how do you use AI, they don't know the answer 401 00:26:07,800 --> 00:26:12,040 Speaker 4: because it's so blended in their life that they don't 402 00:26:12,080 --> 00:26:15,800 Speaker 4: feel like it's something that they are using. They are 403 00:26:15,840 --> 00:26:18,240 Speaker 4: getting used to that. So when I think of next 404 00:26:18,280 --> 00:26:22,240 Speaker 4: generation and the years to come, that generation is so 405 00:26:23,160 --> 00:26:26,880 Speaker 4: used to AI being part of their life that they 406 00:26:26,880 --> 00:26:29,920 Speaker 4: feel like it's just there. That's one and the second 407 00:26:29,960 --> 00:26:33,119 Speaker 4: one is the simplicity of interaction. With AI that you 408 00:26:33,200 --> 00:26:36,560 Speaker 4: don't feel like you're interacting with the system. It's just there, 409 00:26:36,640 --> 00:26:39,800 Speaker 4: like you talk to AI. Everything is automated. So I 410 00:26:39,840 --> 00:26:44,639 Speaker 4: would say the simplicity and being blended to solve the 411 00:26:44,800 --> 00:26:49,159 Speaker 4: right problems is the part that I'm referring to as invisible. 412 00:26:49,280 --> 00:26:52,600 Speaker 4: Like Internet is everywhere and it's invisible. But we used 413 00:26:52,600 --> 00:26:55,760 Speaker 4: to dial in, like you remember the dialing zone to 414 00:26:55,960 --> 00:27:00,760 Speaker 4: connect the Internet. It's gone. The Internet is completely invisible today. 415 00:27:00,640 --> 00:27:03,360 Speaker 3: Right, Like we used to talk about logging on, right, 416 00:27:03,400 --> 00:27:06,359 Speaker 3: and you don't log on anymore because you're always logged on. 417 00:27:07,040 --> 00:27:08,360 Speaker 4: Yep, you're always connected. 418 00:27:08,480 --> 00:27:13,040 Speaker 3: Yeah, what's the number one thing that people misunderstand about AI? 419 00:27:13,640 --> 00:27:18,439 Speaker 4: AI is an ivitable but should not be feared. 420 00:27:19,440 --> 00:27:22,199 Speaker 3: What advice would you give yourself ten years ago to 421 00:27:22,440 --> 00:27:24,320 Speaker 3: better prepare you for today? 422 00:27:25,280 --> 00:27:28,800 Speaker 4: I would say, develop a broad range of skills. Even 423 00:27:29,040 --> 00:27:32,719 Speaker 4: if you think they will not help you today, they 424 00:27:32,800 --> 00:27:34,320 Speaker 4: may be valuable in the future. 425 00:27:34,920 --> 00:27:38,119 Speaker 3: So on the consumer side, right now, we hear a 426 00:27:38,119 --> 00:27:43,439 Speaker 3: lot about chatbots and image generators. But on the business side, 427 00:27:43,480 --> 00:27:46,000 Speaker 3: what do you think is the next big business application? 428 00:27:46,560 --> 00:27:49,120 Speaker 4: AI? Influencers generating content? 429 00:27:49,560 --> 00:27:52,080 Speaker 3: Huh? How do you use AI in your day to 430 00:27:52,160 --> 00:27:52,879 Speaker 3: day life today. 431 00:27:53,760 --> 00:27:57,679 Speaker 4: One simple example is linked in posts. I love it 432 00:27:57,720 --> 00:27:59,800 Speaker 4: to just go to my product. I'll give you an 433 00:27:59,800 --> 00:28:03,040 Speaker 4: example which is my favorite one. Lama three point one 434 00:28:03,119 --> 00:28:06,320 Speaker 4: four A five b the post that I announced on 435 00:28:06,400 --> 00:28:09,760 Speaker 4: LinkedIn on hey, IBM is releasing the model on the 436 00:28:09,800 --> 00:28:12,800 Speaker 4: same day it was generated by Lama three point one 437 00:28:12,840 --> 00:28:16,200 Speaker 4: four five billion. So using the same model to post 438 00:28:16,240 --> 00:28:20,040 Speaker 4: the generate the announcement note very elegant. 439 00:28:20,640 --> 00:28:22,200 Speaker 3: Is there anything else I should ask you? 440 00:28:22,640 --> 00:28:25,600 Speaker 4: Oh, we didn't talk about instruct lab. So when you 441 00:28:25,680 --> 00:28:28,560 Speaker 4: grab a model, you start from the model, but you 442 00:28:28,680 --> 00:28:33,679 Speaker 4: need to then customize it on your proprietary data to 443 00:28:33,720 --> 00:28:37,359 Speaker 4: create value on top of that. So instruct lab is 444 00:28:37,400 --> 00:28:43,880 Speaker 4: giving you a method based on open source contributions to 445 00:28:43,920 --> 00:28:50,160 Speaker 4: collectively contribute to improve the base model. So if you're 446 00:28:50,200 --> 00:28:56,360 Speaker 4: an enterprise, you can leverage your internal employees to collectively 447 00:28:56,480 --> 00:29:00,800 Speaker 4: all contribute to improve the model. Give you an example 448 00:29:00,840 --> 00:29:03,000 Speaker 4: of why it matters, Like if you go to hiking 449 00:29:03,040 --> 00:29:06,400 Speaker 4: pace today and look for Lama, there are about fifty 450 00:29:06,480 --> 00:29:10,520 Speaker 4: thousand different lamas coming up. And the reason is because 451 00:29:10,560 --> 00:29:13,600 Speaker 4: there is no way to contribute to the base model. 452 00:29:14,040 --> 00:29:16,080 Speaker 4: If you're a developer, you have to make a colon 453 00:29:16,160 --> 00:29:18,600 Speaker 4: of the copy of the model and finding need for 454 00:29:18,640 --> 00:29:21,760 Speaker 4: your own purpose. We figure the method that we call 455 00:29:21,960 --> 00:29:25,920 Speaker 4: instruct lab to be able to collectively collect all that 456 00:29:26,000 --> 00:29:29,560 Speaker 4: information and contribute to the base model and enhance. So 457 00:29:29,720 --> 00:29:33,360 Speaker 4: that's instruct lab. I just wanted to highlight the value 458 00:29:33,400 --> 00:29:37,200 Speaker 4: of being open because that's another topic that has been 459 00:29:37,200 --> 00:29:40,480 Speaker 4: emerging in the market over the past eighteen months. In particular, 460 00:29:41,160 --> 00:29:44,080 Speaker 4: I believe the future of AI is open and we've 461 00:29:44,120 --> 00:29:49,280 Speaker 4: been seeing how the open source markets has been changing, 462 00:29:49,600 --> 00:29:53,640 Speaker 4: how the models are accessible to a wider audience, and 463 00:29:54,080 --> 00:29:58,560 Speaker 4: good things typically happen when you make technology pieces accessible 464 00:29:58,600 --> 00:30:01,920 Speaker 4: to a broader range of community to stress test that, 465 00:30:02,640 --> 00:30:05,440 Speaker 4: and that's the direction that we've been adopting with granted, 466 00:30:05,560 --> 00:30:07,480 Speaker 4: and I feel like that's really the adoption that the 467 00:30:07,520 --> 00:30:10,240 Speaker 4: market is going to emerge to moving forward. 468 00:30:10,440 --> 00:30:14,800 Speaker 3: Yeah, there's this interesting I think maybe naively unintuitive, but 469 00:30:14,880 --> 00:30:17,280 Speaker 3: it makes sense once you think about it, thing that 470 00:30:18,040 --> 00:30:21,080 Speaker 3: open source things are safer. You might naively think, oh no, 471 00:30:21,240 --> 00:30:23,000 Speaker 3: put it in a box so nobody can see it, 472 00:30:23,000 --> 00:30:25,160 Speaker 3: and that'll be safer. But like it turns out, of 473 00:30:25,160 --> 00:30:27,560 Speaker 3: the world, if you let everybody poke at it, the 474 00:30:27,600 --> 00:30:30,320 Speaker 3: world will find the vulnerabilities for you and you can. 475 00:30:30,240 --> 00:30:33,880 Speaker 4: Fix them right, That's exactly what's going to happen. Yeah. 476 00:30:34,120 --> 00:30:36,600 Speaker 3: Great, it was lovely to talk with you. Thank you 477 00:30:36,640 --> 00:30:37,520 Speaker 3: so much for your time. 478 00:30:38,000 --> 00:30:41,840 Speaker 4: The same here, thanks Jacob, and. 479 00:30:41,760 --> 00:30:44,480 Speaker 2: That wraps up this episode. A huge thanks to Marian 480 00:30:44,560 --> 00:30:48,280 Speaker 2: and Jacob. Today's conversation open my eyes as to how 481 00:30:48,360 --> 00:30:53,000 Speaker 2: open technology and AI are intersecting to create more transparent 482 00:30:53,320 --> 00:30:57,640 Speaker 2: and efficient systems for enterprises. From the power of smaller, 483 00:30:57,720 --> 00:31:00,800 Speaker 2: more targeted models like granted to the importance of trust 484 00:31:00,920 --> 00:31:05,760 Speaker 2: and governance in AI, these developments are reshaping how businesses 485 00:31:05,840 --> 00:31:09,840 Speaker 2: operate at their core. As we continue to unpack the 486 00:31:09,920 --> 00:31:15,120 Speaker 2: complexities of artificial intelligence, it's clear that openness, whether in data, 487 00:31:15,560 --> 00:31:19,800 Speaker 2: technology or collaboration, is not just a concept, but a 488 00:31:19,920 --> 00:31:26,280 Speaker 2: driving force that can unlock new possibilities. Smart Talks with 489 00:31:26,320 --> 00:31:29,600 Speaker 2: IBM is produced by Matt Romano, Joey fish Ground, Amy 490 00:31:29,640 --> 00:31:33,720 Speaker 2: Gains McQuaid and Jacob Goldstein, who are edited by Lydia 491 00:31:33,800 --> 00:31:37,520 Speaker 2: Jane kott Or. Engineers are Sarah Brugerer and Ben Tolliday. 492 00:31:38,040 --> 00:31:40,880 Speaker 2: Theme song by Gramoscope. Special thanks to the eight Bar 493 00:31:41,000 --> 00:31:44,880 Speaker 2: and IBM teams, as well as the Pushkin marketing team. 494 00:31:45,320 --> 00:31:47,960 Speaker 2: Smart Talks with IBM is a production of Pushkin Industries 495 00:31:48,200 --> 00:31:52,520 Speaker 2: and Ruby Studio at iHeartMedia. To find more Pushkin podcasts, 496 00:31:52,800 --> 00:31:56,320 Speaker 2: listen on the iHeartRadio app, Apple Podcasts, or wherever you 497 00:31:56,440 --> 00:32:02,880 Speaker 2: listen to podcasts. I'm Malcolm Glauwell. This is a paid 498 00:32:02,920 --> 00:32:07,080 Speaker 2: advertisement from ib M. The conversations on this podcast don't 499 00:32:07,120 --> 00:32:22,520 Speaker 2: necessarily represent IBM's positions, strategies, or opinions.