WEBVTT 00:00.000 --> 00:13.000 Okay, thank you very much for the delay and I will start quickly, I only have 10 minutes 00:13.000 --> 00:15.400 so that's really tough. 00:15.400 --> 00:16.400 My name is Richard Haigle. 00:16.400 --> 00:21.760 I'm one of the founders of Halloweld and Halloweld is the company behind the Blue Spice Media 00:21.760 --> 00:22.760 Week. 00:22.760 --> 00:30.440 Open Source Enterprise Week as XWiki, which are good colleagues here and fighters for Open Source 00:30.440 --> 00:31.440 in Europe. 00:31.440 --> 00:37.240 So, what I want to introduce you is our some thoughts. 00:37.240 --> 00:44.760 We have, we have in the evening, if we sit in Riggsburg and have some beers and it's about 00:44.760 --> 00:50.560 the relationship between Wiki, okay, we are Wiki manufacturer, right? 00:51.560 --> 00:54.560 Well, so everybody is talking about that. 00:54.560 --> 01:00.560 To bring you on the same page, I want you to bring you quickly back 20 years ago. 01:00.560 --> 01:05.560 So, 20 years ago we have had something, what I would call a Wikipedia shock. 01:05.560 --> 01:11.560 The shock needs, it means okay, we have something like collective intelligence, 01:11.560 --> 01:18.560 wisdom of the crowd, is a new paradigm in the way how you deal with knowledge. 01:19.560 --> 01:25.560 So, even the youngest people here in the room, maybe don't know what that means. 01:25.560 --> 01:33.560 So, people are really impressed that we have an easy to use Wiki as a central source of truth 01:33.560 --> 01:42.560 in the web and it works because people are working together, they collaborate, they generate user, 01:42.560 --> 01:48.560 they use a generate content and structures, by the way, not only content structures, 01:48.560 --> 01:57.560 as important as the content I think, and it's easy to use because you can work with natural language 01:57.560 --> 02:00.560 and with free text. 02:00.560 --> 02:06.560 It's made a difference in the way how we search for knowledge, okay, you're using Google 02:06.560 --> 02:10.560 but the second click mostly is Wikipedia. 02:10.560 --> 02:16.560 And you have a fast and centralized access to initial information. 02:16.560 --> 02:21.560 And this paradigm was transferred to the private sector, of course, 02:21.560 --> 02:28.560 and the companies also become used to have a wiki internally for internal information 02:28.560 --> 02:33.560 and sharing processes stuff like this is of the interest. 02:36.560 --> 02:40.560 We have seen a few challenges from the beginning. 02:40.560 --> 02:45.560 It was about how reliable is the information we see there. 02:45.560 --> 02:48.560 And where does the information come from? 02:48.560 --> 02:51.560 What are the sources or the references to that? 02:51.560 --> 02:54.560 Can we categorize everything and what is a good category? 02:54.560 --> 02:58.560 What is a good structure beyond a single page? 02:58.560 --> 03:04.560 And last but not least, is that what I see and that what we have as information? 03:04.560 --> 03:08.560 Is this relevant somehow? 03:08.560 --> 03:13.560 I give you, in my opinion, that has to be properly solved. 03:13.560 --> 03:18.560 All these problems have to be properly solved by the wiki media community. 03:18.560 --> 03:22.560 And now we come to something where is it not solved? 03:22.560 --> 03:26.560 Today, we see something like a jet GPT shock. 03:26.560 --> 03:31.560 Sorry, it's for me, it is a shock. 03:32.560 --> 03:37.560 Because you have now the way that you, like with a human, 03:37.560 --> 03:44.560 can take conversations with artificial agents, with a large language model, 03:44.560 --> 03:50.560 and you get impressive in a passive way, you get answers back in a quality, 03:50.560 --> 03:53.560 which you haven't seen so far. 03:53.560 --> 03:57.560 And in the background, there is a lot of mass data. 03:58.560 --> 04:04.560 And so people become used to look things up in jet GPT or in others, 04:04.560 --> 04:08.560 or deep seek of whatever, and get the information. 04:08.560 --> 04:11.560 If you know how those engines are working, 04:11.560 --> 04:16.560 it's incredible that people think they get a reliable information from them. 04:16.560 --> 04:20.560 That's the reason how jet GPT or the LLM's work, 04:20.560 --> 04:24.560 they do a lot of good stuff in talking with you, 04:24.560 --> 04:26.560 but they have no clue what they're doing. 04:26.560 --> 04:29.560 And you have no clue where the sources are, or what is, 04:29.560 --> 04:31.560 where does the content come from? 04:31.560 --> 04:36.560 But people believe this is how they get proper information, 04:36.560 --> 04:39.560 because it's a convenient way to deal with. 04:39.560 --> 04:42.560 And so that will not go away, 04:42.560 --> 04:46.560 so maybe we should talk where are the strengths of these tools, 04:46.560 --> 04:48.560 when there are oh, five minutes left, 04:48.560 --> 04:52.560 and where can we do something different? 04:52.560 --> 04:54.560 You know all the problems, 04:54.560 --> 04:57.560 Hellenization, you know, 04:57.560 --> 04:59.560 deterministic behaviors. 04:59.560 --> 05:02.560 So if you have a prompt and you put the prompt again and again, 05:02.560 --> 05:05.560 and again, you get different answers. 05:05.560 --> 05:08.560 Give it a try, ask that GPT, 05:08.560 --> 05:10.560 what is 2 plus 2, 10 times. 05:10.560 --> 05:15.560 And you will see the 9th results will surprise you. 05:17.560 --> 05:20.560 So, what we see now, 05:20.560 --> 05:22.560 because people say, oh, oh, oh, oh, no, 05:22.560 --> 05:24.560 Google is gone. 05:24.560 --> 05:26.560 Wikipedia is dead, so it's a replacement, 05:26.560 --> 05:28.560 not true at all. 05:28.560 --> 05:32.560 So what we see and I have learned over the last two years, 05:32.560 --> 05:34.560 there is a symbiotic relationship, 05:34.560 --> 05:38.560 so between those two technologies. 05:38.560 --> 05:45.560 Because, okay, first of all, everybody knows that, 05:45.560 --> 05:49.560 that engines where the models are trained with Wikipedia content, 05:49.560 --> 05:51.560 okay, that's given. 05:51.560 --> 05:54.560 But what we see also is that Wikis, 05:54.560 --> 05:55.560 as knowledge-based, 05:55.560 --> 06:03.560 can be a very reliable and valuable source for controlled internal information, 06:03.560 --> 06:07.560 and we can give this to the chatbot. 06:07.560 --> 06:11.560 So we can, or the ideas, 06:11.560 --> 06:15.560 we can enhance the reliability of the answers of a chatbot 06:15.560 --> 06:17.560 by using Wikis. 06:17.560 --> 06:19.560 How is that done? 06:19.560 --> 06:21.560 Okay, use the chatbot. 06:21.560 --> 06:27.560 Look, I have a pretty nice source for you in a Wikis, 06:27.560 --> 06:31.560 for instance, just get the information from there. 06:31.560 --> 06:35.560 And maybe I also use a search and enterprise search 06:35.560 --> 06:37.560 and say, make a pre-selection, 06:37.560 --> 06:41.560 what content should be part of the answer. 06:41.560 --> 06:43.560 So with that, it's easy now, 06:43.560 --> 06:46.560 you get from a controlled source, 06:46.560 --> 06:48.560 the information back, 06:48.560 --> 06:52.560 and with the search, you can also say, 06:52.560 --> 06:54.560 okay, that other references, that very sources, 06:54.560 --> 06:56.560 we have looked up. 06:56.560 --> 06:59.560 So now it became more and more interesting. 06:59.560 --> 07:01.560 You can add references, 07:01.560 --> 07:03.560 maybe you also give, 07:03.560 --> 07:05.560 for the answering machine, 07:05.560 --> 07:07.560 you give an ontology, 07:07.560 --> 07:09.560 you say, there are some relationships, 07:09.560 --> 07:12.560 so you work with guidelines or checklists, 07:12.560 --> 07:14.560 stuff like that. 07:14.560 --> 07:15.560 And in the end, of course, 07:15.560 --> 07:18.560 you can validate if that answer was good, 07:18.560 --> 07:22.560 and you can improve the source where it came from. 07:24.560 --> 07:25.560 One and a half years, 07:25.560 --> 07:27.560 nobody had an idea that will work. 07:27.560 --> 07:29.560 That is one practical example. 07:29.560 --> 07:31.560 That is, 07:31.560 --> 07:34.560 and sorry, is it's manual. 07:34.560 --> 07:40.560 From the Federal Ministry for Education and Research, 07:40.560 --> 07:46.560 in Germany, they have an internal source here for projects, 07:46.560 --> 07:48.560 a funding of projects, 07:48.560 --> 07:50.560 and what we have done is exactly here. 07:50.560 --> 07:53.560 You see there is a chat where they can ask 07:53.560 --> 07:56.560 question how they get the funding, 07:56.560 --> 07:57.560 and some more details, 07:57.560 --> 07:59.560 and you see here where it comes from, 07:59.560 --> 08:02.560 and you can look up the sources. 08:02.560 --> 08:06.560 Another example, a bakery. 08:07.560 --> 08:09.560 So there is a huge bakery, 08:09.560 --> 08:11.560 not a single one-man, 08:11.560 --> 08:12.560 one woman bakery, 08:12.560 --> 08:15.560 but a huge one with hundreds of people 08:15.560 --> 08:18.560 working there on large machines, 08:18.560 --> 08:21.560 and if that machine is breaking, 08:21.560 --> 08:25.560 maybe you need to know what is, 08:25.560 --> 08:26.560 how can I, 08:26.560 --> 08:27.560 what should I do? 08:27.560 --> 08:28.560 I have an error message, 08:28.560 --> 08:30.560 what does this mean? 08:30.560 --> 08:31.560 And what we do here, 08:31.560 --> 08:33.560 together with the Fraunhofer Institute, 08:33.560 --> 08:34.560 is, 08:35.560 --> 08:37.560 okay, you get a chatbot, 08:37.560 --> 08:38.560 but now, 08:38.560 --> 08:43.560 we try to teach the LLM, 08:43.560 --> 08:45.560 that that is a specific machine, 08:45.560 --> 08:46.560 and an ancient, 08:46.560 --> 08:49.560 and there are some relations between, 08:49.560 --> 08:51.560 okay, this is a, 08:51.560 --> 08:55.560 yeah, how this machine works, 08:55.560 --> 09:00.560 and make it easier to get exactly the information 09:00.560 --> 09:01.560 that's needed, 09:02.560 --> 09:06.560 because if that information is not reliable, 09:06.560 --> 09:09.560 you play with the Herz of the people there, 09:09.560 --> 09:11.560 and Herz of other people. 09:11.560 --> 09:14.560 So what we are doing in this research, 09:14.560 --> 09:16.560 project, 09:16.560 --> 09:20.560 is how can we improve the reliability here, 09:20.560 --> 09:24.560 and make something that is not only for a bakery, 09:24.560 --> 09:29.560 but also for other industries useful. 09:30.560 --> 09:34.560 To come to terms, 09:34.560 --> 09:37.560 there we are. 09:37.560 --> 09:41.560 I would say that Ricky's special platforms for using AI, 09:41.560 --> 09:42.560 so everybody comes around, 09:42.560 --> 09:43.560 I haven't seen her, 09:43.560 --> 09:46.560 okay, we are doing something with AI. 09:46.560 --> 09:48.560 We have a ticket system doing something with AI. 09:48.560 --> 09:51.560 Everybody is doing something with AI, right? 09:51.560 --> 09:56.560 But what makes Ricky's unique? 09:57.560 --> 10:01.560 First of all, they complement each other. 10:01.560 --> 10:07.560 They are very, very, very good in working with processing free text. 10:07.560 --> 10:11.560 And then we have a layer for, 10:11.560 --> 10:12.560 for chatting, 10:12.560 --> 10:15.560 or something that is a perfectly communicator, 10:15.560 --> 10:17.560 with the chatbot, 10:17.560 --> 10:18.560 and we have the Wiki, 10:18.560 --> 10:19.560 or the other hand, 10:19.560 --> 10:24.560 that is a perfect tool for control, 10:25.560 --> 10:29.560 content together with other people, 10:29.560 --> 10:31.560 and make it accessible, 10:31.560 --> 10:33.560 but not also only for humans, 10:33.560 --> 10:36.560 but also for machines. 10:36.560 --> 10:39.560 So I have to quit now, 10:39.560 --> 10:41.560 that, 10:41.560 --> 10:45.560 if you want to talk with me about this person, 10:45.560 --> 10:48.560 learn how this works in detail, 10:48.560 --> 10:52.560 just contact me or Robert after this talk. 10:53.560 --> 10:55.560 So what I think is, 10:55.560 --> 10:57.560 there is a lot to do, 10:57.560 --> 10:59.560 because that's not that easy, 10:59.560 --> 11:02.560 and you don't throw an L&M on a Wiki, 11:02.560 --> 11:04.560 and you get good results. 11:04.560 --> 11:05.560 You have to do a lot, 11:05.560 --> 11:08.560 that the information is valuable, 11:08.560 --> 11:11.560 and that's exactly a good process, 11:11.560 --> 11:13.560 a good project for open source communities, 11:13.560 --> 11:16.560 researchers and companies, 11:16.560 --> 11:18.560 who can now, 11:18.560 --> 11:20.560 with open source software, 11:20.560 --> 11:21.560 experiment, 11:21.560 --> 11:22.560 innovate, 11:22.560 --> 11:24.560 and refine to gaps, 11:24.560 --> 11:26.560 reliable, 11:26.560 --> 11:28.560 non-menu, 11:28.560 --> 11:30.560 non-menu, 11:30.560 --> 11:32.560 my new pull-at-heath, 11:32.560 --> 11:33.560 knowledge, 11:33.560 --> 11:34.560 that forms, 11:34.560 --> 11:36.560 for the 21st century, 11:36.560 --> 11:37.560 and I promise, 11:37.560 --> 11:40.560 I will practice this word, 11:40.560 --> 11:42.560 until the next four statements you there. 11:42.560 --> 11:44.560 Thank you very much for your time.