1 00:00:14,680 --> 00:00:18,560 Hello, this is Eva and you're listening to the In Between Tech 2 00:00:18,560 --> 00:00:21,200 and Trust podcast. In this week's episode, we dive 3 00:00:21,200 --> 00:00:25,320 into one of the most relevant topics that you can look into in 4 00:00:25,320 --> 00:00:29,880 the AI era, and that is the intersection of AI media and 5 00:00:29,880 --> 00:00:32,520 journalism. And for that topic, I have one 6 00:00:32,520 --> 00:00:36,320 of the leading voices with me in conversation, and that is Doctor 7 00:00:36,320 --> 00:00:39,880 Paul Elvas. And Paul is head of AI at funk 8 00:00:39,880 --> 00:00:42,920 comedian Cropper. He's also leading the AI lab 9 00:00:42,960 --> 00:00:47,120 within the company, so he brings his practical view into 10 00:00:47,120 --> 00:00:52,040 translating between stakeholders and implementing AI into the 11 00:00:52,040 --> 00:00:56,440 journalistic aspects of how to publish content and contain 12 00:00:56,440 --> 00:00:58,800 audience trust. So together we dive. 13 00:00:58,800 --> 00:01:03,520 Into how AI intersects with trust, transparency, and the 14 00:01:03,520 --> 00:01:06,200 journalistic. Practice, Paul also shares. 15 00:01:06,200 --> 00:01:10,160 His view on how an AI newsroom of the future should look like. 16 00:01:10,240 --> 00:01:16,120 We also look into the impacts of the so-called AI slop of what, 17 00:01:16,120 --> 00:01:20,360 for example, deepfakes, audience mistrust, and AI generated 18 00:01:20,360 --> 00:01:25,200 content does to the industry. But we also want to look into 19 00:01:25,200 --> 00:01:29,080 tangible ways of how to maintain credibility, democratic 20 00:01:29,080 --> 00:01:33,560 functions and navigating the rapidly changing technological 21 00:01:33,560 --> 00:01:37,320 conditions. Hi, Paul. 22 00:01:37,320 --> 00:01:39,240 Good morning. Good morning, Yifa, nice to be 23 00:01:39,240 --> 00:01:41,480 here. I'm particularly intrigued to 24 00:01:41,480 --> 00:01:45,080 dive into the intersection of tech, trust, and media with you 25 00:01:45,080 --> 00:01:47,200 today. But before we start into the 26 00:01:47,200 --> 00:01:51,560 topic particularly, I would love to hear your thought on the 27 00:01:51,560 --> 00:01:54,840 intersection of tech and trust. What is the first thing that 28 00:01:54,840 --> 00:01:56,640 comes to your mind when you hear that? 29 00:01:58,080 --> 00:02:02,160 So I think like trust and tech isn't really about blind faith. 30 00:02:02,200 --> 00:02:06,080 I think it's has a lot to do about understanding how the 31 00:02:06,080 --> 00:02:09,199 systems work and and who's accountable basically. 32 00:02:09,199 --> 00:02:12,560 So I think initially when especially tech is 33 00:02:12,560 --> 00:02:17,120 transformative, people tend to mistrust the technology and a 34 00:02:17,120 --> 00:02:21,960 lot of trust can be gained when you start working with it or 35 00:02:21,960 --> 00:02:25,760 just do like, yeah, do something with it and then you build up 36 00:02:25,760 --> 00:02:27,520 trust. You understand how the system 37 00:02:27,520 --> 00:02:32,160 works and then you also find out or figure out which components 38 00:02:32,160 --> 00:02:36,000 of the system you can trust and like which other parts of the 39 00:02:36,000 --> 00:02:40,480 system maybe are a little bit more ambiguous and you have to 40 00:02:40,480 --> 00:02:42,520 double check things in this domain. 41 00:02:43,120 --> 00:02:46,160 So I think it has a lot to do with understanding the 42 00:02:46,160 --> 00:02:49,720 technology to basically build up trust for things that are new. 43 00:02:50,480 --> 00:02:53,000 And the things that I knew, something that is like on your 44 00:02:53,000 --> 00:02:57,200 daily program, so to say, because as head of AI at one of 45 00:02:57,200 --> 00:03:01,720 the largest like media houses in Germany, you're basically at the 46 00:03:01,720 --> 00:03:05,880 linchpin of like newness and media and also the technology in 47 00:03:05,880 --> 00:03:07,720 itself of what shapes the industry. 48 00:03:07,840 --> 00:03:11,440 And you also sit at the core of digital transformation. 49 00:03:11,440 --> 00:03:15,440 So what is, according to your opinion and expertise of what 50 00:03:15,440 --> 00:03:18,400 you saw over the past couple of weeks and months, really the 51 00:03:18,400 --> 00:03:22,360 biggest misconception when we talk about AI and journalism 52 00:03:22,360 --> 00:03:25,800 today? I think we have to kind of 53 00:03:25,800 --> 00:03:30,520 differentiate between those within the field of media and 54 00:03:30,520 --> 00:03:34,720 and those who are basically more or less consumers of media. 55 00:03:35,320 --> 00:03:40,320 I would say that as people who are reading the news or 56 00:03:40,320 --> 00:03:44,800 consuming media primarily, there is this belief that that the 57 00:03:44,800 --> 00:03:49,680 media is like not putting a lot of thought into how they use AI. 58 00:03:49,680 --> 00:03:54,320 So I think some of them might have like the the conception at 59 00:03:54,320 --> 00:03:59,680 least that that media companies start using AI just to create 60 00:03:59,680 --> 00:04:03,480 content with it in a very naive manner. 61 00:04:03,560 --> 00:04:07,440 And I think part of this is also true for those who are in the 62 00:04:07,440 --> 00:04:11,840 field or working as, as journalists or in, in media 63 00:04:11,840 --> 00:04:16,959 companies that they initially think now the shift towards AI 64 00:04:17,120 --> 00:04:21,720 means that we have to naively adopt the technology and create 65 00:04:21,959 --> 00:04:26,520 just more content without supervision, without shaping 66 00:04:26,600 --> 00:04:30,360 this how the system work. And I would say the reality is 67 00:04:30,360 --> 00:04:36,440 more like that that now is a, is a time where you basically can 68 00:04:36,520 --> 00:04:41,240 reshape the whole process of creating media, creating 69 00:04:41,240 --> 00:04:44,720 content. And those who basically put a 70 00:04:44,720 --> 00:04:48,640 lot of effort into it and, and know how to integrate the 71 00:04:48,640 --> 00:04:52,760 technology well, they will succeed in a way that they will 72 00:04:52,760 --> 00:04:56,080 not just create a lot of AI slop, let's put it like that, 73 00:04:56,200 --> 00:05:00,440 but actually use AI to amplify also the quality of the media. 74 00:05:00,520 --> 00:05:03,360 I think you can like if you do it right, you can, you can 75 00:05:03,360 --> 00:05:07,080 integrate AI at the right steps in the process of creating 76 00:05:07,080 --> 00:05:11,600 various kinds of content that actually do amplify by the final 77 00:05:11,600 --> 00:05:15,600 results instead of like lowering or damping what comes out. 78 00:05:15,680 --> 00:05:18,680 And media companies shouldn't like just naively adopt the 79 00:05:18,680 --> 00:05:20,720 technology. They should focus on on this 80 00:05:20,720 --> 00:05:25,240 part of figuring out how you can create good content that people 81 00:05:25,240 --> 00:05:27,600 are interested in with the technology. 82 00:05:28,600 --> 00:05:33,080 And you said like a word that does actually all over the news. 83 00:05:33,080 --> 00:05:37,280 It's like the AI slop. Well, does that mean to you 84 00:05:37,280 --> 00:05:39,840 particularly? And also, could you explain a 85 00:05:39,840 --> 00:05:42,960 little bit more of how it impacts your work or also 86 00:05:43,520 --> 00:05:46,600 another work of like the media house that you're part of? 87 00:05:47,040 --> 00:05:50,960 Yeah, I mean, AI slob, I kind of dislike the word a little bit, 88 00:05:50,960 --> 00:05:54,720 but it it's just around there and people use it to describe, I 89 00:05:54,720 --> 00:05:59,520 would say mediocre or bad content that is just created 90 00:05:59,560 --> 00:06:04,600 with AI tools. Could be a short video for 91 00:06:04,600 --> 00:06:09,640 social platforms could be news articles that are published on 92 00:06:09,800 --> 00:06:12,280 pages that are not really trustworthy. 93 00:06:12,760 --> 00:06:16,040 So this is typically the kind of content that we refer to with AI 94 00:06:16,040 --> 00:06:20,240 slop, but also sometimes you can kind of like put this as a label 95 00:06:20,240 --> 00:06:22,720 on a lot of different kinds of content, and sometimes you're 96 00:06:22,720 --> 00:06:27,200 also not really appreciating certain kinds of AI content that 97 00:06:27,200 --> 00:06:30,520 are actually good. I mean, I just recently talked 98 00:06:30,520 --> 00:06:34,800 about the the cat that is playing various instruments on 99 00:06:34,800 --> 00:06:37,640 on the front porch. And I actually believe this is 100 00:06:37,880 --> 00:06:40,120 this is a unique kind of quality content. 101 00:06:40,120 --> 00:06:43,760 Actually, while it is completely created by by AII think it's 102 00:06:43,760 --> 00:06:48,240 VO3A video model that created these scenes where where a cat 103 00:06:48,240 --> 00:06:50,160 is playing. I don't know, a lot of different 104 00:06:50,160 --> 00:06:53,040 kinds of instruments, which obviously you could never film 105 00:06:53,040 --> 00:06:54,440 because a cat would never do this. 106 00:06:54,960 --> 00:06:58,760 But it's really entertaining to to see this kind of idea and 107 00:06:58,760 --> 00:07:01,960 it's funny and, and it has a lot of things that that people would 108 00:07:02,080 --> 00:07:04,920 usually put into content when they created. 109 00:07:05,640 --> 00:07:08,000 So I think we have to be careful with the AI slot label. 110 00:07:08,000 --> 00:07:12,160 But on the other hand, I think it really good, nicely describes 111 00:07:12,160 --> 00:07:16,840 how consumers of media are now have a hard time navigating the 112 00:07:16,840 --> 00:07:21,120 landscape of media and basically figuring out what is trustworthy 113 00:07:21,120 --> 00:07:24,680 and what is not trustworthy. And it surfaces this challenge 114 00:07:24,680 --> 00:07:29,160 that AI content gives us to basically also judge what is 115 00:07:29,160 --> 00:07:32,440 true, what is not true, what can I believe and what is just like 116 00:07:32,440 --> 00:07:35,760 fake news. And I think there are two sides 117 00:07:35,760 --> 00:07:37,480 of like the fear that people have. 118 00:07:37,480 --> 00:07:41,480 So on the one hand side, it's the fear of trusting the media 119 00:07:41,480 --> 00:07:44,800 output and contents in terms of information that you consume. 120 00:07:45,280 --> 00:07:48,800 And on the other side, it's also like losing the trust in 121 00:07:48,800 --> 00:07:54,560 journalism overall and the craft of journalism and in itself, 122 00:07:54,560 --> 00:07:57,720 because also there is a lot of talks that journalists won't 123 00:07:57,720 --> 00:08:00,040 have a job. If you just would like put it 124 00:08:00,040 --> 00:08:02,960 really like frankly, and that the quality is erode. 125 00:08:03,040 --> 00:08:06,920 What do you see AI actually enabling, particularly when you 126 00:08:06,920 --> 00:08:12,480 talk about quality or quantity? And where do you see legitimate 127 00:08:12,480 --> 00:08:16,520 risks that could impact the way of how we the news that we 128 00:08:16,520 --> 00:08:19,360 consume? I think AI, like in The 129 00:08:19,360 --> 00:08:24,360 Newsroom, is really good at automating tedious tasks, tasks 130 00:08:24,360 --> 00:08:28,520 that nobody really likes but you have to do as a journalist. 131 00:08:28,520 --> 00:08:32,919 So transcribing things, summarizing things, doing the 132 00:08:32,919 --> 00:08:35,760 research for your papers. AI can assist with a lot of 133 00:08:35,760 --> 00:08:40,320 these processes and basically can give you more time to focus 134 00:08:40,320 --> 00:08:44,000 on those parts of the process that should always be done by a, 135 00:08:44,000 --> 00:08:48,480 by a human, by by somebody who is crafting the content and who 136 00:08:48,520 --> 00:08:51,760 is in in the driver's seat for the final result. 137 00:08:51,760 --> 00:08:55,680 So this is also something that we at for Comedian Corpora 138 00:08:55,680 --> 00:08:59,480 inherently believe in that it is really important to keep the 139 00:08:59,480 --> 00:09:03,360 human the center of the process at the beginning and the end. 140 00:09:03,360 --> 00:09:07,120 So whenever content is created, a human should be the one who's 141 00:09:07,120 --> 00:09:10,040 deciding how to create it. What, what message do you want 142 00:09:10,040 --> 00:09:11,960 to convey? And also how should the final 143 00:09:11,960 --> 00:09:14,280 piece look? And, but obviously AI can help 144 00:09:14,280 --> 00:09:17,680 us to be a lot more efficient. And that's something that we 145 00:09:17,680 --> 00:09:20,000 should appreciate and, and integrate. 146 00:09:20,000 --> 00:09:23,520 But then I mean, on the other hand, we should always focus 147 00:09:23,520 --> 00:09:26,800 that we will keep some things with the human in the loop. 148 00:09:26,800 --> 00:09:31,080 And then you also asked about like trust, How can we keep the 149 00:09:31,080 --> 00:09:34,000 trust basically with us? And I would say in the situation 150 00:09:34,000 --> 00:09:37,240 right now where you have a lot of AI slot and users have 151 00:09:37,240 --> 00:09:40,680 trouble like navigating to landscape brands and especially 152 00:09:40,680 --> 00:09:44,680 also people become like very important to basically build up 153 00:09:44,680 --> 00:09:47,200 trust. Because trust in media also for 154 00:09:47,200 --> 00:09:51,880 me means to build up a relationship with either people 155 00:09:51,880 --> 00:09:54,600 or it could also be a relationship to to a news brand 156 00:09:54,600 --> 00:09:57,640 where you basically know that they work thoroughly. 157 00:09:57,640 --> 00:10:00,800 And I think for media companies, this is kind of like something 158 00:10:00,800 --> 00:10:02,920 that they should put on their top priority. 159 00:10:03,000 --> 00:10:07,080 We have to like make sure that we always keep the trust with 160 00:10:07,080 --> 00:10:10,280 our audience, that we keep this relationship to our audience, 161 00:10:10,280 --> 00:10:14,400 and be really transparent about how we use AI to basically make 162 00:10:14,400 --> 00:10:19,600 sure that we are the ones that people go to if they want to 163 00:10:19,760 --> 00:10:24,960 know that they can trust us with how we talk about what's 164 00:10:24,960 --> 00:10:31,160 happening in the world. If we talk transparency, it's 165 00:10:31,160 --> 00:10:35,200 one of like the most crucial parts of of the whole discussion 166 00:10:35,200 --> 00:10:39,200 of how we set up and interact with AI, particularly in the way 167 00:10:39,200 --> 00:10:42,480 that content is shaped. Who do you think is responsible, 168 00:10:42,480 --> 00:10:47,760 the consumer or the provider or like a third party organization 169 00:10:48,080 --> 00:10:52,040 that provides, you know, frameworks for it? 170 00:10:53,640 --> 00:10:58,560 I think it should be a shared responsibility that is with 171 00:10:58,720 --> 00:11:03,640 everybody who's every party that is part of the process of 172 00:11:04,720 --> 00:11:12,400 shaping or creating content. I think the those who trained 173 00:11:12,440 --> 00:11:16,840 the big AI models, they, they could obviously do a lot more in 174 00:11:16,840 --> 00:11:20,280 terms of transparency. We still don't really know which 175 00:11:20,280 --> 00:11:24,840 data went to went into the models, into the big LLMS that 176 00:11:24,840 --> 00:11:27,520 were trained. So we need a lot more 177 00:11:27,520 --> 00:11:29,680 transparency about the training data. 178 00:11:31,440 --> 00:11:37,080 But then also us as intermediaries who are using 179 00:11:37,080 --> 00:11:41,000 this technology and either incorporate this into workflows, 180 00:11:41,000 --> 00:11:44,080 we need to make sure that we be that we are transparent about 181 00:11:44,800 --> 00:11:48,200 how we use it. And for instance, if we like, we 182 00:11:48,200 --> 00:11:52,120 have a certain set of rules, for instance, that if something is, 183 00:11:53,640 --> 00:11:57,520 is, it's created by AI. And then we, we label this piece 184 00:11:57,520 --> 00:12:00,360 of content that is an AI created piece. 185 00:12:00,440 --> 00:12:03,600 And we also make transparent about like what, what, what part 186 00:12:03,600 --> 00:12:08,320 of this is created by an AI. So to give you an example, we 187 00:12:08,320 --> 00:12:13,560 have a daily newsletter where people can now also listen to 188 00:12:13,560 --> 00:12:16,800 the newsletter. And the newsletter is written by 189 00:12:16,960 --> 00:12:18,800 one of the journalists in our newsroom. 190 00:12:19,160 --> 00:12:25,960 And now they can hear the newsletter that is is an audio 191 00:12:26,560 --> 00:12:29,200 spoken by the person who wrote the newsletter. 192 00:12:29,920 --> 00:12:35,200 But we use AI to basically turn the text into audio. 193 00:12:35,760 --> 00:12:39,840 So we use 11 apps here to basically create an audio 194 00:12:39,840 --> 00:12:44,320 podcast for the newsletter. And this is really good because 195 00:12:44,320 --> 00:12:49,520 we basically reach our readers or listeners in this case in 196 00:12:49,520 --> 00:12:54,000 another type of media. And we can like transform the 197 00:12:54,000 --> 00:12:57,720 original content piece into something else with very little 198 00:12:57,720 --> 00:13:01,000 effort. We can use AI for this, but we 199 00:13:01,000 --> 00:13:05,640 make sure that this is transparent and that that the 200 00:13:05,720 --> 00:13:08,280 the the user knows like what is happening. 201 00:13:08,280 --> 00:13:14,840 What's going on here? I think that is so like tangible 202 00:13:14,840 --> 00:13:20,120 of how you incorporated it into your processes or the way that 203 00:13:20,120 --> 00:13:24,640 you approach AI. From my experience, it's hard to 204 00:13:24,640 --> 00:13:28,720 be that clear as the way that you also just reflected upon of 205 00:13:28,720 --> 00:13:32,000 how you as an organization are. How did you get there? 206 00:13:32,000 --> 00:13:37,560 Like how did you come up with this clarity of how you use AI 207 00:13:37,720 --> 00:13:42,400 and who did you also loop into the preparation to get there? 208 00:13:44,560 --> 00:13:47,600 Yeah, I think that's a, that's a, that's a big challenge 209 00:13:47,600 --> 00:13:51,040 actually for for media companies, probably also for 210 00:13:51,240 --> 00:13:57,720 companies in other domains because you AI is basically 211 00:13:58,400 --> 00:14:02,520 everywhere and it it has they're like every part in your company 212 00:14:02,520 --> 00:14:06,720 basically is more or less involved into this process of 213 00:14:07,720 --> 00:14:15,000 agreeing on let's say a certain consensus how to use AI in the 214 00:14:15,000 --> 00:14:18,480 company. And so, so in our case, we 215 00:14:18,480 --> 00:14:23,000 basically had to make sure that we getting agreement between the 216 00:14:23,000 --> 00:14:26,200 Workers Union, the, the, the high level executive board, 217 00:14:26,480 --> 00:14:30,560 those who work in the unused room and, and, and basically 218 00:14:30,560 --> 00:14:34,680 make sure that every party is involved in this process and can 219 00:14:34,680 --> 00:14:36,840 like incorporate their perspective on this. 220 00:14:37,760 --> 00:14:43,360 And then basically, you know, try to, to, to agree on 221 00:14:43,360 --> 00:14:46,560 something certain set of rules that we all think are valid. 222 00:14:46,920 --> 00:14:50,640 But also to be honest, I think because the technology is just 223 00:14:50,640 --> 00:14:54,880 evolving so rapidly, this is not really for a long duration and 224 00:14:54,880 --> 00:14:57,680 you basically always have to refine these things. 225 00:14:57,680 --> 00:15:02,520 If you just think like how much has happened in this year alone 226 00:15:03,200 --> 00:15:05,880 and also how much has the industry moved? 227 00:15:05,880 --> 00:15:09,320 How much more are we like moving towards AI content? 228 00:15:09,320 --> 00:15:14,720 If you look at video, for instance, like video AI at the 229 00:15:14,720 --> 00:15:17,560 beginning of the year wasn't really a thing that you like, 230 00:15:17,840 --> 00:15:21,040 think is is around there. But, but now it's like 231 00:15:21,160 --> 00:15:23,200 everywhere on social media, you can find video AI. 232 00:15:24,000 --> 00:15:29,200 So that just shows us like how fast things are moving and that 233 00:15:29,200 --> 00:15:33,720 you constantly need to like re evaluate also the, the, the set 234 00:15:33,720 --> 00:15:40,880 of consensus rules or points that you agree on and and make 235 00:15:40,880 --> 00:15:46,600 sure that you're always in line with how the technology evolves. 236 00:15:47,920 --> 00:15:51,480 But doesn't that make you like not only the head of AI, but 237 00:15:52,240 --> 00:15:55,520 more of an in like integrator, orchestrator between the 238 00:15:55,520 --> 00:15:59,480 different kinds of, you know, skill sets, stakeholders? 239 00:15:59,520 --> 00:16:04,440 I mean, how you just walked us through of who's like needed to 240 00:16:04,440 --> 00:16:08,520 be looped in. That's like quite the pool of 241 00:16:08,800 --> 00:16:11,560 people that could form an opinion or so how do you 242 00:16:11,560 --> 00:16:15,720 navigate that? Yeah, it's, it's really, it's 243 00:16:15,720 --> 00:16:19,920 really crazy to basically have all these hats on and, and, and 244 00:16:19,920 --> 00:16:23,800 shift between them. So basically in between a tech 245 00:16:23,800 --> 00:16:30,920 lab that I'm also managing. And then basically this, this 246 00:16:30,920 --> 00:16:34,200 role, which is more like an internal, I don't know, policy 247 00:16:34,200 --> 00:16:38,840 team or whatever, who is trying to put every or get everybody on 248 00:16:38,840 --> 00:16:45,680 the table. I think, I think this is just 249 00:16:45,680 --> 00:16:48,120 part of the role. And also what I, what I see in 250 00:16:48,120 --> 00:16:52,760 the industry is that this is an emerging role that many 251 00:16:52,760 --> 00:16:57,760 companies now have. And there are people in other 252 00:16:57,760 --> 00:17:00,640 companies who have similar challenges. 253 00:17:00,640 --> 00:17:03,200 And, and I think it's, it's, it's just necessary. 254 00:17:03,200 --> 00:17:09,400 Somebody needs to like, like connect the dots and make sure 255 00:17:09,400 --> 00:17:14,720 that the Dai topic is just steered in the right way. 256 00:17:14,720 --> 00:17:19,000 And, and this is you, you cannot like the, the, the topic is just 257 00:17:19,000 --> 00:17:20,760 so big. You cannot, like, you cannot do 258 00:17:20,760 --> 00:17:23,160 it this on your own. You cannot just like make 259 00:17:23,160 --> 00:17:24,960 decisions on your own. You have to get everybody 260 00:17:24,960 --> 00:17:28,560 involved. So it's it's just you have to do 261 00:17:28,560 --> 00:17:31,400 it this way or I think it will fail, you know? 262 00:17:32,960 --> 00:17:37,160 And when you now build or deploy the AI systems with the lab that 263 00:17:37,160 --> 00:17:41,320 you have also the, you know, partnerships that you form, what 264 00:17:41,320 --> 00:17:47,720 are key decisions of where like transparency, ethics and like 265 00:17:47,720 --> 00:17:51,080 the technological infrastructure that's needed collide? 266 00:17:54,080 --> 00:17:57,560 So, yeah, so I think there are like obviously like several 267 00:17:57,880 --> 00:18:04,080 layers of or at several layers, you basically have to integrate 268 00:18:04,320 --> 00:18:09,160 safety and also make sure that the system basically responds in 269 00:18:09,160 --> 00:18:16,200 the way that is aligned with how you want it to respond and also 270 00:18:16,200 --> 00:18:18,920 make sure that your ethical standards are met. 271 00:18:20,200 --> 00:18:26,120 I think on a very low level, you basically have like the 272 00:18:26,120 --> 00:18:29,640 decisions for the the models that you use in your system. 273 00:18:29,640 --> 00:18:33,800 So basically we we make sure that we use models that we think 274 00:18:34,400 --> 00:18:40,680 put a lot of effort into meeting ethics standards and were 275 00:18:40,680 --> 00:18:45,520 transparent about, you know how they are at least a little bit 276 00:18:45,520 --> 00:18:48,000 more transparent about how they train their models. 277 00:18:48,720 --> 00:18:52,880 And also that they put efforts into making sure that these 278 00:18:52,880 --> 00:18:58,880 models to at least to certain extent comply with some ethical 279 00:18:58,880 --> 00:19:01,240 standards. So like we're not like 280 00:19:01,480 --> 00:19:04,240 triggering hate speech or harmful content. 281 00:19:07,680 --> 00:19:10,920 So that's on the on the model base where you basically select 282 00:19:10,960 --> 00:19:13,040 certain types of model that you want to integrate into your 283 00:19:13,040 --> 00:19:15,160 systems. And then basically on top of 284 00:19:15,160 --> 00:19:16,640 that, you just build up the system. 285 00:19:17,080 --> 00:19:21,280 So, and then it really depends on like what kind of system you 286 00:19:21,280 --> 00:19:22,640 built. Do you build something for 287 00:19:22,640 --> 00:19:27,440 internal use that is basically, I don't know, drafting something 288 00:19:27,440 --> 00:19:34,920 or used in The Newsroom for like research purpose or for giving 289 00:19:34,920 --> 00:19:38,040 you feedback on, on your article or anything like that. 290 00:19:39,160 --> 00:19:41,280 Then we have different constraints and then we 291 00:19:41,360 --> 00:19:43,280 basically can do like some internal testing. 292 00:19:43,280 --> 00:19:45,240 We have feedback loops, these kind of things. 293 00:19:46,120 --> 00:19:48,040 We also have the human in the loop principles. 294 00:19:48,040 --> 00:19:52,160 So everything that where AI is used will always be reviewed by 295 00:19:52,160 --> 00:19:55,320 human. And then we have other systems 296 00:19:55,320 --> 00:19:59,920 where we basically built AI systems that are user facing. 297 00:20:00,480 --> 00:20:04,080 So when users interact with AI systems, you cannot really make 298 00:20:04,080 --> 00:20:07,720 sure that every answer is proof read by humans. 299 00:20:07,720 --> 00:20:11,080 So basically then you have to double check that the system is 300 00:20:11,080 --> 00:20:14,680 working. Accordingly, so we do intensive 301 00:20:14,680 --> 00:20:20,160 red teaming. So basically we try to to get 302 00:20:20,520 --> 00:20:23,000 like the worst out of the bot, so to speak. 303 00:20:23,400 --> 00:20:27,560 So we're trying to trick the bot into giving us answers that the 304 00:20:27,560 --> 00:20:30,560 bot shouldn't give us. We're trying to jailbreak the 305 00:20:30,560 --> 00:20:36,200 bot in a way and only if like this red teaming and and several 306 00:20:36,200 --> 00:20:40,800 other tests are positive, then we are rolling the system out to 307 00:20:40,800 --> 00:20:43,240 the users. And then even if we rolled the 308 00:20:43,240 --> 00:20:47,960 system out, we always make sure that we have feedback loop and 309 00:20:47,960 --> 00:20:51,920 can check, check that the quality is good. 310 00:20:54,040 --> 00:20:56,280 So yeah, several levels. Yeah. 311 00:20:56,760 --> 00:20:59,240 I mean, that's amazing. And I think particularly when 312 00:20:59,240 --> 00:21:03,400 you talk about the quality and coming back also to the craft 313 00:21:03,760 --> 00:21:10,040 part of like output in itself or in the way that like news are 314 00:21:10,040 --> 00:21:13,480 being brought to the market. Basically, we live in an era 315 00:21:13,480 --> 00:21:16,560 where deep fake has never been that easy to produce. 316 00:21:16,560 --> 00:21:21,760 And also like synthetic content that is actually not real but 317 00:21:21,760 --> 00:21:27,160 artificial like logically, but is supposed to be hyper 318 00:21:27,160 --> 00:21:29,320 personalized. It's also supposed to be in real 319 00:21:29,320 --> 00:21:31,600 time. But still, there's this like 320 00:21:31,920 --> 00:21:37,920 underlying feeling that I cannot trust of what I see in the news 321 00:21:37,920 --> 00:21:41,480 that I consume anymore. And where is the tipping point 322 00:21:41,480 --> 00:21:46,160 at which, like AI begins to distort reality more than it 323 00:21:46,160 --> 00:21:50,200 supports it for you and also from your experience in the 324 00:21:50,200 --> 00:21:55,480 field? I think we, we already reached a 325 00:21:55,480 --> 00:21:59,440 point where this is like really a topic that we should take 326 00:21:59,440 --> 00:22:04,760 seriously on several levels, also on our societal level. 327 00:22:04,760 --> 00:22:08,280 And we need to like bring these discussions to the table. 328 00:22:08,920 --> 00:22:12,080 There are studies that are showing that the majority of 329 00:22:12,080 --> 00:22:16,720 content that is now being published online is AI created. 330 00:22:17,480 --> 00:22:22,720 And from like anecdotal consumption of social media, you 331 00:22:22,720 --> 00:22:26,560 can really see that there are a lot of instances where I'm not 332 00:22:26,560 --> 00:22:30,320 able or where at at least I takes me several minutes to 333 00:22:30,320 --> 00:22:33,760 basically think about if videos are real or not. 334 00:22:33,760 --> 00:22:37,200 And, and I can't really tell. And these are obviously only the 335 00:22:37,200 --> 00:22:41,040 examples that trigger this. There might be a lot of videos 336 00:22:41,040 --> 00:22:46,400 that I see that are AI created and I'm I'm, I just don't 337 00:22:46,400 --> 00:22:50,000 realize it anymore. So so I think this is really 338 00:22:50,000 --> 00:22:54,080 really an important topic that we need to bring to the table 339 00:22:54,400 --> 00:22:59,280 also to be discussed by our politicians, by those who shape 340 00:22:59,280 --> 00:23:03,320 basically the guardrail. It's the, the, the, the laws 341 00:23:03,320 --> 00:23:07,200 also for, for how basically the media landscape is, is created 342 00:23:07,200 --> 00:23:12,120 because I think ultimately also like our democracy needs that we 343 00:23:12,120 --> 00:23:20,760 have trustworthy media that basically is, is helping the 344 00:23:20,760 --> 00:23:25,240 people in the country to shape their opinions in a way that 345 00:23:25,240 --> 00:23:28,280 they're not just influenced by AI content. 346 00:23:28,280 --> 00:23:34,520 And if we look at a very at the political, like the global 347 00:23:34,520 --> 00:23:38,520 political landscape right now, there are many actors that can 348 00:23:38,560 --> 00:23:44,800 definitely use AI for purposes that are very bad and that can 349 00:23:44,960 --> 00:23:49,560 really harm our society. So, so this is not, not 350 00:23:49,560 --> 00:23:54,440 something that maybe we need to talk about. 351 00:23:54,440 --> 00:23:57,040 It's this is a serious topic that we need to talk about. 352 00:23:57,800 --> 00:24:01,600 And I think like for us as a media company, we also have a 353 00:24:01,600 --> 00:24:06,800 responsibility to, to, to participate into mitigating the 354 00:24:06,800 --> 00:24:10,400 situation and basically helping people to build trust. 355 00:24:10,400 --> 00:24:14,080 And I think for us this, this like, I mean, we have to 356 00:24:14,080 --> 00:24:22,040 basically fight against this huge amount of synthetic 357 00:24:22,040 --> 00:24:25,560 content. We have to be visible. 358 00:24:25,560 --> 00:24:31,680 We have to convince our readers that we are trustworthy every 359 00:24:31,680 --> 00:24:35,440 day again and again and build up our communities and our 360 00:24:35,440 --> 00:24:37,560 audiences. I think we, we put a lot of 361 00:24:37,560 --> 00:24:44,640 effort into, in into doing this and, and this like to, to be, to 362 00:24:44,640 --> 00:24:48,960 be relevant in the future means to basically use AI but use it 363 00:24:48,960 --> 00:24:56,160 in a good way and also show others that AI is it's, it's 364 00:24:56,160 --> 00:24:59,640 not, it's not ultimately bad. It's, it's also, it's, it's in a 365 00:24:59,640 --> 00:25:02,680 way also like tool. So if you're a bad actor and use 366 00:25:02,680 --> 00:25:05,200 a tool for bad purposes, it can be very harmful. 367 00:25:05,200 --> 00:25:09,360 But if you are somebody who tries to build up trust and be 368 00:25:09,360 --> 00:25:11,560 trustworthy, you can also use AI. 369 00:25:11,680 --> 00:25:14,760 But you have to make sure to use it in a transparent way. 370 00:25:14,760 --> 00:25:19,000 Because then we can basically compete with this AI slot that 371 00:25:19,000 --> 00:25:25,200 is out there and be visible and show users that we have 372 00:25:25,200 --> 00:25:28,360 trustworthy information that they can consume. 373 00:25:29,360 --> 00:25:33,240 And if you now think about like the way that you work, it's 374 00:25:33,240 --> 00:25:40,240 mostly like summarized under the like newsroom that you like are 375 00:25:40,280 --> 00:25:43,160 sometimes even literally in the room with. 376 00:25:43,240 --> 00:25:47,640 And the way that you just reflected upon the impact of 377 00:25:47,640 --> 00:25:52,880 trustworthy news versus like the fake news and the way that also 378 00:25:52,880 --> 00:25:58,160 media like companies need to bring this to the consumer. 379 00:25:58,760 --> 00:26:04,160 Does it have an impact on The Newsroom in an AI world, so to 380 00:26:04,160 --> 00:26:08,760 say? And if so, if you yourself could 381 00:26:08,840 --> 00:26:13,520 redesign a newsroom today for like the future, how would you 382 00:26:13,840 --> 00:26:15,280 like? How would you set it up? 383 00:26:15,320 --> 00:26:20,280 Why and how to get there? That's a really good question. 384 00:26:22,840 --> 00:26:25,880 Yeah, I think that there could be a whole podcast episode in in 385 00:26:25,880 --> 00:26:28,360 itself. So maybe I'll focus on on two 386 00:26:28,360 --> 00:26:29,240 points. I think. 387 00:26:29,600 --> 00:26:35,560 I think one truth that a lot of media companies still struggle 388 00:26:35,560 --> 00:26:41,560 with is that news is inherently shifting from text to other 389 00:26:41,560 --> 00:26:44,000 forms. It's more today, it's more about 390 00:26:44,120 --> 00:26:51,320 audio, it's podcasts, it's video on social media platforms that 391 00:26:51,320 --> 00:26:54,120 are consumed as a news source, especially by younger 392 00:26:54,120 --> 00:26:58,360 generations. And we have to acknowledge this. 393 00:26:58,760 --> 00:27:05,480 And I think in newsroom of the future would be a lot more video 394 00:27:05,480 --> 00:27:11,240 first, podcast 1st and article second, I would say. 395 00:27:13,160 --> 00:27:16,000 So that's one thing. And then also maybe the other 396 00:27:16,000 --> 00:27:24,640 thing would be to also, yeah, design the The Newsroom AI 397 00:27:24,640 --> 00:27:27,680 first. Because as you can imagine, 398 00:27:27,680 --> 00:27:34,080 there are people who have been in journalism for 20 or 30 years 399 00:27:34,080 --> 00:27:38,720 and they have a typical way how they, how they work, how they, 400 00:27:39,720 --> 00:27:44,640 how they structure the work. And I think if you look at young 401 00:27:44,640 --> 00:27:49,080 startups, for instance, at Morning Crunch, which is a 402 00:27:49,160 --> 00:27:58,600 newsletter first news company, they build up a newsroom that is 403 00:27:59,240 --> 00:28:03,000 basically AI first. So, and what that means 404 00:28:03,000 --> 00:28:08,360 basically is to rethink every component of the process in 405 00:28:08,360 --> 00:28:13,440 terms of where can we maximize the efficiency of AI for this 406 00:28:13,440 --> 00:28:17,080 process without compromising any of the quality. 407 00:28:17,680 --> 00:28:21,960 And I mean, obviously it's always easy to design this on 408 00:28:21,960 --> 00:28:28,040 Greenfield and it's much more difficult to basically transform 409 00:28:31,320 --> 00:28:34,120 legacy media into a new newsroom. 410 00:28:34,120 --> 00:28:38,200 But but yeah, maybe maybe two things that I would that I would 411 00:28:38,360 --> 00:28:41,680 emphasize is to, to go on on something like video 1st or the 412 00:28:41,680 --> 00:28:47,240 1st and then basically create a create a more streamlined AI 413 00:28:47,240 --> 00:28:51,800 first process of of how to work in The Newsroom. 414 00:28:52,840 --> 00:28:58,040 And I like that you are so like specific about the forms of 415 00:28:58,040 --> 00:29:02,280 content and like how it depends on the like, generations, 416 00:29:02,280 --> 00:29:04,400 consumption, also the habits that we've formed. 417 00:29:05,160 --> 00:29:08,720 But doesn't that also inherently mean that we need to think about 418 00:29:08,720 --> 00:29:14,960 distribution of content? So currently, I would almost say 419 00:29:14,960 --> 00:29:20,720 everyone's consuming media through social media and like on 420 00:29:20,720 --> 00:29:25,360 a, let's say classical way that you go to a website or app of a 421 00:29:25,360 --> 00:29:30,920 specific like news house. However, if we I now understand 422 00:29:30,920 --> 00:29:34,720 you correctly, you also propose that the way that we like 423 00:29:34,800 --> 00:29:39,640 consume it via the distribution channels will massively change 424 00:29:39,640 --> 00:29:42,760 because the forms of contents change. 425 00:29:42,760 --> 00:29:51,520 Or how would you also view that of like a prerequisite for media 426 00:29:51,520 --> 00:29:55,040 houses to link it to the production? 427 00:29:55,400 --> 00:30:00,480 And how would you then also regard it to be enabled through 428 00:30:00,520 --> 00:30:06,360 AI or maybe also like challenged by AI because it's not possible 429 00:30:06,360 --> 00:30:13,400 to for whatever reason? Yeah, I think, I think the 430 00:30:13,400 --> 00:30:17,320 distribution also matters. In an ideal world, and I would 431 00:30:17,320 --> 00:30:21,640 say like in a especially in a world before the introduction of 432 00:30:21,640 --> 00:30:25,440 Cha Chi PT and Google AI overviews, Google AI mode, these 433 00:30:25,440 --> 00:30:31,040 kind of things, we always as publishing houses after the 434 00:30:31,040 --> 00:30:34,560 introduction of Internet, we were aiming for having direct 435 00:30:34,560 --> 00:30:37,400 relationships to our consumers and basically making sure that 436 00:30:38,200 --> 00:30:42,080 people will find us with relevant keywords that they 437 00:30:42,080 --> 00:30:45,840 search for and then basically click on a link and then they 438 00:30:45,840 --> 00:30:50,840 are on our website. What we see now today is that 439 00:30:50,840 --> 00:30:54,800 this is becoming increasingly difficult and more and more 440 00:30:54,800 --> 00:31:00,280 difficult basically because this this old contract of the 441 00:31:00,400 --> 00:31:06,560 Internet, the search contract that you basically get referral 442 00:31:06,560 --> 00:31:09,320 clicks for your content. This is not not really working 443 00:31:09,320 --> 00:31:12,840 anymore. We see that ChatGPT and Google 444 00:31:13,280 --> 00:31:17,600 AI overviews are having significantly lower click 445 00:31:17,600 --> 00:31:19,640 through rates and referral traffic. 446 00:31:19,640 --> 00:31:23,680 So, so it's becoming harder and harder to basically get people 447 00:31:23,680 --> 00:31:31,200 on your, on your websites. And, and this is also true for, 448 00:31:31,200 --> 00:31:35,720 for, for social media because it's, it's really difficult to, 449 00:31:35,720 --> 00:31:38,960 to get out of there and, and, and the platforms make sure to 450 00:31:38,960 --> 00:31:41,640 build up this water garden basically that's keeping people 451 00:31:41,640 --> 00:31:47,360 inside the, the platform. However, I think it's, it's 452 00:31:47,360 --> 00:31:52,160 still relevant to basically try getting people onto your 453 00:31:52,160 --> 00:31:55,440 websites and to your products and to build up direct 454 00:31:55,440 --> 00:31:58,520 relationships with them. I think there are a lot more 455 00:32:00,240 --> 00:32:02,760 opportunities that media companies now explore. 456 00:32:02,760 --> 00:32:07,000 This could be live events, this could be, I don't know, 457 00:32:07,080 --> 00:32:09,880 mailings. This could be WhatsApp channels 458 00:32:09,880 --> 00:32:13,400 that are like directly connected to your journalists and editors. 459 00:32:13,880 --> 00:32:18,160 So basically like I think in the core is to build up this 460 00:32:18,160 --> 00:32:22,920 relationship between you and and your readers and your audience 461 00:32:23,400 --> 00:32:27,480 and make the best of how this how the the media landscape is 462 00:32:27,480 --> 00:32:30,080 right now. And then I mean, on another 463 00:32:30,080 --> 00:32:35,080 level, I think, and this is more on, on the political level, I 464 00:32:35,080 --> 00:32:39,640 think we have to make sure to basically have a media landscape 465 00:32:39,640 --> 00:32:46,000 that is, that is taking into account to basically have a, a 466 00:32:46,600 --> 00:32:54,440 level playing field and have a, have a fair, fair market where 467 00:32:54,440 --> 00:33:00,560 basically users are still able to, to get to, to the sources 468 00:33:00,560 --> 00:33:02,840 and, and to get to the publishing companies. 469 00:33:02,840 --> 00:33:07,920 Because as as a citizen of the country, I I strongly believe we 470 00:33:07,920 --> 00:33:15,840 need independent media and we have to make sure that this that 471 00:33:15,840 --> 00:33:18,560 this market still exists and will exist in the future. 472 00:33:19,280 --> 00:33:23,920 And sorry, I I need to go. That was like a jump off point 473 00:33:23,920 --> 00:33:27,840 for me into one of my passion topics, independent media. 474 00:33:28,360 --> 00:33:32,640 And you linked it to because I'm a citizen of this country. 475 00:33:32,880 --> 00:33:36,480 So you inherently linked like independent media to the 476 00:33:36,480 --> 00:33:40,720 political landscape. And one key question that I 477 00:33:40,720 --> 00:33:43,440 always have myself is how do we get there? 478 00:33:44,040 --> 00:33:48,040 What would you propose? That's a. 479 00:33:48,360 --> 00:33:52,120 That's a good question. If I would have the solution, 480 00:33:52,120 --> 00:33:56,760 I'd probably have a. Different head on your head. 481 00:33:57,160 --> 00:34:01,640 Yes, that would be really great if if somebody had the solution. 482 00:34:03,880 --> 00:34:08,840 No, I think, I think, I think we have to at one point we have to 483 00:34:08,840 --> 00:34:12,840 fight for like for, for a European standpoint and the 484 00:34:12,840 --> 00:34:17,199 whole ecosystem because a lot of the media landscape today is 485 00:34:17,199 --> 00:34:21,679 also shaped by big tech companies from the US and also 486 00:34:22,199 --> 00:34:24,960 China. So basically we have to 487 00:34:24,960 --> 00:34:27,040 basically build up our own standpoint. 488 00:34:27,120 --> 00:34:33,960 This doesn't really mean to like cut off every connection to 489 00:34:33,960 --> 00:34:37,280 them, but to basically find an agreement. 490 00:34:37,280 --> 00:34:43,960 And I think, I think as I mentioned, the technology's 491 00:34:44,239 --> 00:34:47,520 evolving so rapidly and I, I think like within this journey 492 00:34:47,520 --> 00:34:53,880 also, I myself have like change my perspective on certain 493 00:34:53,880 --> 00:34:56,159 things. And, and also on this question, 494 00:34:56,159 --> 00:35:01,240 I would, I'm, I'm still a little bit optimistic that that I think 495 00:35:01,240 --> 00:35:05,360 there is the potential that after a very rapid and 496 00:35:05,360 --> 00:35:11,920 disruptive introduction of AI platforms like Chechi PT or also 497 00:35:11,920 --> 00:35:17,480 the new AI summaries in Google, that we now can basically get 498 00:35:17,480 --> 00:35:22,600 back to the table and negotiate with those companies also on 499 00:35:22,600 --> 00:35:26,480 different levels, politics, the media companies and everybody 500 00:35:26,480 --> 00:35:30,520 who was involved to basically find solutions that work for 501 00:35:30,520 --> 00:35:36,520 everybody. Because I think in the end, this 502 00:35:36,520 --> 00:35:40,280 experience shift that we observed with AI that was 503 00:35:40,280 --> 00:35:44,520 introduced by chair GPT mainly and now is you can see in many 504 00:35:44,520 --> 00:35:46,880 platforms, this is not going away. 505 00:35:46,880 --> 00:35:49,720 I think because as a consumer, I love, I love this. 506 00:35:49,720 --> 00:35:53,640 It's really, really nice way of getting information in a very 507 00:35:53,640 --> 00:35:56,160 condensed way. And now we just have to 508 00:35:56,160 --> 00:36:01,520 basically figure out how we can work with this new technology in 509 00:36:01,520 --> 00:36:05,000 a way that independent media can still survive. 510 00:36:07,120 --> 00:36:11,920 So right now I already have at least two another episodes that 511 00:36:11,920 --> 00:36:14,480 I would want to do to the like first, The Newsroom of the 512 00:36:14,480 --> 00:36:16,880 Future and 2nd Independent Media. 513 00:36:16,880 --> 00:36:21,040 How do we get there? But for now, we're almost at the 514 00:36:21,040 --> 00:36:24,320 end of this one. So I do want to cherish the 515 00:36:24,320 --> 00:36:28,320 moment and have some so-called in between moments with you. 516 00:36:28,680 --> 00:36:32,240 They are intuitive ones. And these are the ones where you 517 00:36:32,240 --> 00:36:36,200 literally just share what comes to your mind 1st. 518 00:36:36,240 --> 00:36:39,880 And I do want to start with like a key question for me to you, 519 00:36:40,640 --> 00:36:42,840 what does trust and media feel like? 520 00:36:44,480 --> 00:36:48,040 I would say it feels like a handshake that must be renewed 521 00:36:48,040 --> 00:36:54,080 every single day. That is such a great visual. 522 00:36:55,040 --> 00:36:56,400 Let's see how we can include that. 523 00:36:56,400 --> 00:37:02,960 Anyways, what should I never do? I think it should never make 524 00:37:03,400 --> 00:37:07,400 decisions that are for good reason, still in the hands of 525 00:37:07,400 --> 00:37:09,760 people, like hiring people for instance. 526 00:37:10,800 --> 00:37:13,080 And what will journalists always do? 527 00:37:14,520 --> 00:37:17,920 I think journalists will always provide the context, they will 528 00:37:18,600 --> 00:37:23,840 bring in the empathy and also their very own and unique 529 00:37:23,840 --> 00:37:28,360 standpoint on how things are in the world. 530 00:37:29,920 --> 00:37:38,040 What is the future of news? To make producing news very 531 00:37:38,040 --> 00:37:43,800 efficient and high quality thanks to AI with the very 532 00:37:43,800 --> 00:37:49,120 thorough overview of human experts. 533 00:37:50,480 --> 00:37:54,440 And do you have an experience that you yourself are currently 534 00:37:54,440 --> 00:38:02,520 in between? Yeah, I think in between. 535 00:38:02,520 --> 00:38:07,960 I think I'm I'm always in between hype and reality when it 536 00:38:07,960 --> 00:38:12,240 comes to AI because sometimes we're just thinking the whole 537 00:38:12,240 --> 00:38:16,640 world change and then in the reality it's maybe not as big as 538 00:38:16,640 --> 00:38:23,000 we think. And if we now close the session 539 00:38:23,000 --> 00:38:29,680 and finish it with just a link to take and trust topic, and if 540 00:38:29,680 --> 00:38:34,000 we if we assume that trust becomes a decisive competitive 541 00:38:34,000 --> 00:38:38,760 advantage in media, what does it mean in your role to be a trust 542 00:38:38,760 --> 00:38:42,440 builder when it comes to the future of media news? 543 00:38:44,280 --> 00:38:53,320 I think to, to build up trust is it is really important to, to be 544 00:38:53,320 --> 00:38:58,160 very transparent inside the company and also to your 545 00:38:58,160 --> 00:39:02,720 audience about how you want to use AI, how you use it, what AI 546 00:39:02,720 --> 00:39:05,400 is, what it can and what it can't. 547 00:39:06,680 --> 00:39:11,320 And to basically for me like what is really at my heart to 548 00:39:11,360 --> 00:39:14,200 everybody at Funker, to everybody who works at the 549 00:39:14,200 --> 00:39:18,400 company, give them the opportunity to understand this 550 00:39:18,400 --> 00:39:22,160 technology as good and as deep as they want, give them all the 551 00:39:22,160 --> 00:39:25,000 resources. Because I think ultimately, and 552 00:39:25,000 --> 00:39:29,040 this basically links back to what I said in the beginning, I 553 00:39:29,040 --> 00:39:33,840 think trust is built on experience and understanding of 554 00:39:33,840 --> 00:39:38,840 technology. Thank you so much for your time 555 00:39:38,880 --> 00:39:44,480 and diving so deep into how you had fungus set up, AI overall, 556 00:39:44,480 --> 00:39:47,040 but also how you view it for the industry. 557 00:39:48,640 --> 00:39:50,520 It was super valuable conversation. 558 00:39:50,520 --> 00:39:52,480 It was great having you. Thanks so much. 559 00:39:53,040 --> 00:39:54,120 Thanks Yvonne, it was fun.