1
00:00:09,220 --> 00:00:11,540
All right, we are live.
2
00:00:12,440 --> 00:00:18,940
Thank you to everyone who is joining us during their lunch in and around the Toronto area
3
00:00:18,940 --> 00:00:26,540
or wherever else you are around the world or in the future on YouTube or on our podcast
4
00:00:26,540 --> 00:00:27,000
channels.
5
00:00:27,300 --> 00:00:28,100
Much thanks.
6
00:00:28,720 --> 00:00:31,680
This is a super special episode today.
7
00:00:31,920 --> 00:00:35,060
Olivia, I don't know if you knew this, but this is episode 50.
8
00:00:36,240 --> 00:00:43,560
We started this as just an idea to be helping and sharing with our friends in the content
9
00:00:43,560 --> 00:00:49,980
operation space and the CMS space to make themselves available to share what they're
10
00:00:49,980 --> 00:00:54,520
thinking, what they're up to, what are the trends in content operations.
11
00:00:55,800 --> 00:00:57,880
So thanks for joining everyone.
12
00:00:57,880 --> 00:00:59,540
My name is Chris.
13
00:00:59,940 --> 00:01:01,700
I'm the founder of Dot Fusion.
14
00:01:01,700 --> 00:01:07,760
We're essentially a content operations agency with most of our product being some kind
15
00:01:07,760 --> 00:01:09,860
of larger website or mobile app.
16
00:01:09,860 --> 00:01:15,660
But we really invest a lot of time in and around the management of content as it moves through
17
00:01:15,660 --> 00:01:16,580
those supply chains.
18
00:01:16,760 --> 00:01:22,720
And I'm just super fortunate to have run into a lot of thought leaders in the space.
19
00:01:22,720 --> 00:01:31,840
So for episode 50, we're like super stoked to invite Olivier de Barcaux from a variety of different
20
00:01:31,840 --> 00:01:36,860
institutions, DKD, as well as I believe you're the president of TYPO3.
21
00:01:38,160 --> 00:01:40,480
You're also staying up a little bit late.
22
00:01:40,900 --> 00:01:41,560
You're in Germany.
23
00:01:41,880 --> 00:01:45,880
So thank you for spending your evening with us.
24
00:01:46,840 --> 00:01:52,900
Maybe just before we get started, the whole idea is just some great conversation.
25
00:01:53,280 --> 00:01:56,060
I do my best to monitor the LinkedIn channels.
26
00:01:56,060 --> 00:02:01,220
If you're watching this live on LinkedIn, if you do have some questions, just let me know.
27
00:02:01,360 --> 00:02:04,740
We'll be happy to answer them or do our best to follow up with afterwards.
28
00:02:06,180 --> 00:02:09,880
But as I said before, time goes by when we've got about a half an hour.
29
00:02:10,160 --> 00:02:12,880
So maybe Olivier, do you want to introduce yourself?
30
00:02:13,260 --> 00:02:17,760
I'm kind of coming off a really great time hanging out with you in Montreal at Yanis Boy
31
00:02:17,760 --> 00:02:19,280
CMS Experts Group.
32
00:02:19,860 --> 00:02:21,660
Kudos to that organization.
33
00:02:22,720 --> 00:02:23,600
Hi, welcome.
34
00:02:25,880 --> 00:02:26,300
Thank you.
35
00:02:26,520 --> 00:02:27,580
Thank you for having me.
36
00:02:27,740 --> 00:02:28,960
Thank you for the invitation.
37
00:02:29,460 --> 00:02:32,840
I know that this invitation has been running for long.
38
00:02:32,840 --> 00:02:38,360
So it really needed us having that one moment in Montreal.
39
00:02:39,160 --> 00:02:45,680
I remember us sitting in front of this depaneur having a cold water.
40
00:02:46,360 --> 00:02:50,640
And so it was really great to meet you again.
41
00:02:50,940 --> 00:02:52,880
I think we met two years ago.
42
00:02:52,880 --> 00:03:08,280
We found out that we knew each other, you know, by one separation of, you know, and that we actually kind of ran into each other maybe even 11 years ago in your time in Cambodia.
43
00:03:08,280 --> 00:03:12,700
So I think that is like kind of, you know, the rapidity knocks sometimes.
44
00:03:13,020 --> 00:03:13,060
Yes.
45
00:03:13,060 --> 00:03:19,140
And then so this is actually a pretty nice way to talk to each other.
46
00:03:19,960 --> 00:03:25,080
Hopefully we will have something for the audience today.
47
00:03:26,200 --> 00:03:28,840
So what can I tell about myself?
48
00:03:29,120 --> 00:03:32,480
I'm 50 years, 58 years old.
49
00:03:32,780 --> 00:03:34,180
Don't feel like that.
50
00:03:34,360 --> 00:03:43,120
I feel currently being 30 again, you know, going back into the times when we created our agency.
51
00:03:43,120 --> 00:03:51,200
I think we are have the same history of, you know, working in the industry and and also being self-taught about a lot of things.
52
00:03:51,440 --> 00:04:02,920
And, you know, nowadays, I guess everyone is agreeing with that is that with the upcoming and more and more, you know, AI in our life.
53
00:04:02,920 --> 00:04:07,980
We all become more versatile on what we are doing.
54
00:04:08,200 --> 00:04:13,480
So I've started again to code and I've done a lot of stuff.
55
00:04:14,180 --> 00:04:15,900
So you're right.
56
00:04:16,060 --> 00:04:20,100
I created DKD at university.
57
00:04:21,160 --> 00:04:25,340
We went into incorporation 1990, 1998.
58
00:04:26,120 --> 00:04:28,720
So almost 30 years from now.
59
00:04:30,480 --> 00:04:34,080
Yeah, the company is now 50 employees.
60
00:04:35,060 --> 00:04:39,320
Most of what we are doing is Type 3 CMS.
61
00:04:39,640 --> 00:04:42,520
It's an open source CMS, PHP based, MySQL.
62
00:04:44,180 --> 00:04:47,640
Has a community, let's say, in the DACH area.
63
00:04:48,040 --> 00:04:49,840
So Germany, Switzerland, Austria.
64
00:04:50,140 --> 00:04:54,560
But it has also strong foundations in, you know, in Europe.
65
00:04:54,560 --> 00:05:05,760
And lately we've been going back to visit and to establish our, you know, our relations with the U.S. market.
66
00:05:06,060 --> 00:05:13,760
This is something that if you look closely, we had an opportunity many, many years ago.
67
00:05:13,760 --> 00:05:17,260
So I think at that time was not right.
68
00:05:17,400 --> 00:05:30,320
But now we find it very interesting to see that people are actually re-discovering Type 3, a very, very strong, with a lot of features CMS out of the box.
69
00:05:30,520 --> 00:05:32,380
So that is pretty nice.
70
00:05:32,380 --> 00:05:39,580
And in regards of what DKD has been doing for Type 3, we integrated Solar Search.
71
00:05:39,860 --> 00:05:46,120
That is like an open source search engine for Type 3 many, many years ago.
72
00:05:46,120 --> 00:05:48,720
We've been keeping it up to date.
73
00:05:48,720 --> 00:05:58,180
We found a way on how we can fund open source development in a group of, you know, I would call them partners or sponsoring partners.
74
00:05:58,480 --> 00:06:02,700
And this has been keeping up Solar up and running.
75
00:06:03,320 --> 00:06:15,280
And so lately we released the Solar version, the extension version 14, which is the matching version for Type 3 version 14.
76
00:06:15,280 --> 00:06:19,480
And we're seeing that this has a lot of adaption.
77
00:06:20,020 --> 00:06:29,040
If you go to packages and you look how much installations have been in the last years, I think we're speaking about one and a half million of installations.
78
00:06:30,500 --> 00:06:38,260
So that is interesting to see that search is still a thing, especially in the enterprise sector.
79
00:06:38,680 --> 00:06:43,920
So I think we will be talking today about, OK, where search is going.
80
00:06:45,280 --> 00:06:52,980
So how, let's say, keyword search is evolving into something that is more like a conversation with the website.
81
00:06:53,640 --> 00:07:01,260
We will be maybe seeing what you should do and not do if you are using that.
82
00:07:01,260 --> 00:07:13,040
But also, I think that we will see that we're all going or we're all on a journey and that there's no silver bullet yet.
83
00:07:13,620 --> 00:07:16,140
I think we've we will.
84
00:07:16,700 --> 00:07:22,960
But let's see if we have some people finding having questions about that.
85
00:07:23,620 --> 00:07:33,320
Awesome. So well, we titled the show today searches the interface, which can provoke a lot of interesting thoughts.
86
00:07:33,620 --> 00:07:43,560
But maybe give us your take on that and and why that's resonating as a phrase and how that kind of calibrates into how we're approaching search in the industry.
87
00:07:43,560 --> 00:07:51,360
Yeah, I think search is something quite very human.
88
00:07:51,600 --> 00:07:57,760
Right. It is something that is in our DNA or in our way to survive.
89
00:07:58,040 --> 00:08:09,820
Right. If we go back to the caveman times, hunting and searching for food is something that that that keeps you keeps you alive.
90
00:08:09,820 --> 00:08:20,360
Right. So I've many, many years ago, I heard a very good, you know, very good establishment of what search is.
91
00:08:20,500 --> 00:08:29,880
And actually, people started to also to speak about the scent of search actually know that you could smell search like, you know, when is the search good?
92
00:08:30,240 --> 00:08:39,280
Because people, you know, at the end, you know, if you let's go back to maybe a bear and something that happens to be in Canada, for example.
93
00:08:40,160 --> 00:08:45,260
You know, you know, the bear knows exactly where to find the food that he's craving for.
94
00:08:45,460 --> 00:08:49,160
Right. If he wants to eat berries, he knows where the berries are.
95
00:08:49,300 --> 00:08:58,620
But this is something that has happened to know that they know because they're smelling it, maybe, but also because they're remembering the places that they have been.
96
00:08:58,620 --> 00:09:13,940
So if you if you would, let's say, compare the search of a human on a website, then this human visiting the website is always expecting the search to behave like something that he has experienced before.
97
00:09:14,240 --> 00:09:16,720
Right. So this is my first hypothesis.
98
00:09:16,720 --> 00:09:27,460
I think you're also remembering Jakob Nielsen, who said we should all do things that are coping to the expectation of people.
99
00:09:27,600 --> 00:09:32,160
Right. We should never surprise people, let's say, with a super duper interface.
100
00:09:33,240 --> 00:09:39,800
I remember when we started to do web design is that we experimented a lot, you know, on, you know, concepts.
101
00:09:39,800 --> 00:09:46,160
And I don't know if you have been around with AOL dial ups.
102
00:09:46,420 --> 00:09:47,840
Yes, not in the disk days.
103
00:09:48,120 --> 00:09:52,080
There was like a thing called eWorld from Apple many, many years ago.
104
00:09:52,080 --> 00:09:55,820
You know, you will dial up and you end up like in a virtual city.
105
00:09:56,100 --> 00:10:11,880
And I think remembering the very first websites that we did was they were quite playful because we had to do also be like we had to, you know, to have like a house and have a like a post office and stuff like this where people could actually find their mail.
106
00:10:12,060 --> 00:10:19,200
So I think at that time, the interfaces became something like kind of, you know, you know, how do you call it?
107
00:10:20,180 --> 00:10:23,320
You know, you had to present something that they could relate to.
108
00:10:23,440 --> 00:10:27,200
Nowadays, I think interfaces have become more and more abstract.
109
00:10:27,940 --> 00:10:30,920
We are seeing that the hamburger is there.
110
00:10:31,080 --> 00:10:34,140
So people understand, OK, this is navigation on the mobile phone.
111
00:10:34,260 --> 00:10:39,920
We know the L navigation, you know, with the top top navigation of points and the sub navigation.
112
00:10:39,940 --> 00:10:41,960
And people do understand this pretty well.
113
00:10:42,200 --> 00:10:47,160
And they see a search field and they see a search button.
114
00:10:47,160 --> 00:10:48,220
They enter them.
115
00:10:48,540 --> 00:10:49,340
They enter keywords.
116
00:10:49,880 --> 00:10:53,420
And with that, they find stuff on the website or not.
117
00:10:53,540 --> 00:11:04,840
If you have a good search engine, you will you will run into re-rankings and you run into like, OK, how much times these the keyword is in the actual document.
118
00:11:06,100 --> 00:11:15,760
There's like this algorithms called like BM25 that does that do a lot of, you know, re-ranking and sorting and relevance and so on.
119
00:11:15,760 --> 00:11:20,580
I think that is what people are still expecting nowadays.
120
00:11:21,100 --> 00:11:24,200
But the experience is changing.
121
00:11:24,440 --> 00:11:29,000
People are going now to answering machines or AI answering machines.
122
00:11:29,000 --> 00:11:31,220
They are typing in their questions.
123
00:11:31,220 --> 00:11:32,260
They are typing in their questions.
124
00:11:32,260 --> 00:11:33,380
They are receiving answers.
125
00:11:34,300 --> 00:11:47,360
And with that, yeah, they might go somewhere else and might go to might come to your website and maybe expect that your website search has it becomes more and more of a conversation.
126
00:11:47,360 --> 00:12:05,440
So, therefore, and maybe and in the future, we will only see, you know, even no longer having human visitors to the website, but just only search agents or agents that are searching for information.
127
00:12:05,440 --> 00:12:10,580
So I think we will be seeing that that that that stuff happening.
128
00:12:11,700 --> 00:12:33,420
You've seen that in the last months and weeks of the of this year, we have seen that agents have become something that end users are getting to start to work with, at least if you are in the northern American hemisphere, you're you have very fast access to that kind of stuff.
129
00:12:33,420 --> 00:12:46,120
So we in Europe are kind of, yeah, how you call it, we're left behind because, you know, we have this AI act and we want to have more security on on that kind of stuff.
130
00:12:46,260 --> 00:12:52,200
So I think that Facebook and all the guys, they all want us not to come and sue them.
131
00:12:52,200 --> 00:12:56,920
So but what does it what does it mean for search?
132
00:12:57,120 --> 00:13:16,780
I mean, we'll see we'll be seeing that our search and search infrastructure will need to change into something that will need to cope those different users or different users on the website like the human users expecting a more, you know, an interface that is more and the
133
00:13:17,640 --> 00:13:29,840
is more understanding your search intention, maybe even, you know, supercharging the search intention with the search experience of others.
134
00:13:29,840 --> 00:13:43,700
And we will see also search interfaces that will be totally programmatic, you know, and then, you know, and advertised in only agent readable formats.
135
00:13:44,360 --> 00:13:46,920
So that is that is what I'm seeing.
136
00:13:47,360 --> 00:13:49,740
But what does it mean for the website?
137
00:13:50,540 --> 00:13:50,680
Right.
138
00:13:51,820 --> 00:14:06,760
There's there's a thing that if you're if you you you work with search, there's one one saying the quality of data is also say also how you call it?
139
00:14:07,880 --> 00:14:11,300
It's influencing the search experience, right?
140
00:14:11,720 --> 00:14:24,240
You cannot search for a needle in the haystack if you also have other stuff in this haystack and actually those things also look like needles.
141
00:14:24,420 --> 00:14:30,260
Right. So you need to to be very unique to be able to disimple grade and and so on and so forth.
142
00:14:30,260 --> 00:14:33,740
So or plainly said junk in junk out.
143
00:14:33,960 --> 00:14:35,720
Right. If you you have this.
144
00:14:36,280 --> 00:14:46,320
So what I'm still seeing nowadays is that the content and the let's say the content that gets indexed into a search index.
145
00:14:48,380 --> 00:14:58,260
It's just index and people do not really care about, OK, what what kind of search am I am I'm giving to my users?
146
00:14:58,330 --> 00:15:02,950
Right. I've done a website for that.
147
00:15:04,050 --> 00:15:05,810
It's it's a test website.
148
00:15:06,090 --> 00:15:07,170
It's it's at the moment.
149
00:15:07,210 --> 00:15:07,830
It's in German.
150
00:15:08,010 --> 00:15:10,570
It's called such quality search quality.
151
00:15:11,210 --> 00:15:16,570
And you can go to the website and you can actually enter the web, the domain of the website.
152
00:15:16,570 --> 00:15:30,810
And with that, an AI is actually a bot is visiting your website, looking at your website and actually tells me what kind of questions would a user have if you visit the website.
153
00:15:31,030 --> 00:15:36,390
Right. Not just only search terms, but real questions that people could have.
154
00:15:36,390 --> 00:15:39,790
And I think this is something.
155
00:15:40,210 --> 00:15:47,730
And Chris, I can offer that to you that I send you a report on the search of your website or anyone.
156
00:15:48,070 --> 00:15:49,690
I'll post it in the link below afterwards.
157
00:15:49,890 --> 00:15:51,090
Yeah, we will do that.
158
00:15:51,290 --> 00:15:55,490
So it's at the moment, I think it's German only, but it's an experiment.
159
00:15:55,490 --> 00:16:11,590
And with that, I mean, what we're doing with that is that we're actually thinking of a transition, you know, how can keyword based search transition into something that is understanding, you know, like people and people putting in questions.
160
00:16:11,590 --> 00:16:16,070
What is the cheapest refrigerator on the website?
161
00:16:16,310 --> 00:16:20,790
I think that is something that I think in e-commerce on e-commerce platforms.
162
00:16:20,910 --> 00:16:24,570
This was a task for many, many, many years already.
163
00:16:24,810 --> 00:16:28,250
I think there's also very cool solutions for that.
164
00:16:28,630 --> 00:16:37,310
But for enterprise websites, there's the content is actually the denominator.
165
00:16:37,730 --> 00:16:40,350
Right. If you don't have sharp written content.
166
00:16:40,350 --> 00:16:54,490
Your AI search integration will not actually find anything because I mean, obviously, all those, let's say, I'm pretty sure you've heard about drag retrieval augmented generation.
167
00:16:54,490 --> 00:17:04,330
That is like a way how you your search gets interpreted by a large language model and the search results get interpreted by large language model.
168
00:17:04,470 --> 00:17:08,230
And with that, you get a presentation of, you know, the results.
169
00:17:09,030 --> 00:17:26,190
Yeah. Once again, if your haystack is, you know, not cool, you will be ending up finding a lot of stuff that is not relevant because, you know, the LLMs will do a lot of stuff to actually help you to find actually something.
170
00:17:26,190 --> 00:17:31,010
And with that, I don't know if you don't contain it and you don't ground it good, good enough.
171
00:17:31,210 --> 00:17:35,830
You end up having like a search that might even hallucinate on stuff.
172
00:17:35,990 --> 00:17:45,150
And, you know, that Canada, you know, we've heard of the Canada chatbot giving you like false policies and stuff like that.
173
00:17:45,150 --> 00:17:59,130
So I think this is also a dimension if you enter with AI search, you will really actually need to find a way to look into the groundings and look into the facts.
174
00:17:59,130 --> 00:18:09,510
And once again, going back, if the content is not holding that information, you will be having a hard time to do so.
175
00:18:09,510 --> 00:18:27,750
I think if we've met several times through Janus Boyer's CMS experts network and one thing that we've seen there for many, many years is like, you know, the machine experience, you know, how bots do, how do they interact with your website?
176
00:18:27,750 --> 00:18:37,690
And we have this special, very special friend, Mr. Cranston, who has written a lot of stuff on that.
177
00:18:37,850 --> 00:18:54,130
And I still think that our industry needs to adapt also standards on, you know, content annotation and, you know, not just only JSON-LD, but I think everything that's also around the content, not just what you're publishing or, you know, the circumstances.
178
00:18:54,130 --> 00:19:01,010
And I think we've heard also very much future foresight in Montreal together.
179
00:19:01,290 --> 00:19:14,230
I think you remember that talk that we've heard from the Forrester analyst that he was like actually saying that he's seeing that content management actually at the moment is not going to go away.
180
00:19:14,230 --> 00:19:35,190
It's going to thrive, but it has to adapt actual more layers of information, more layers of data, but not just only the content and the content model and whatever, but it's also what's around and, you know, maybe also how organizations are actually working.
181
00:19:35,190 --> 00:19:46,910
You need to put this into your content. So, so where were, what do I want to, what do I want to tell you?
182
00:19:46,910 --> 00:20:03,710
I think what I've seen from talking to customers that want to transition, let's say from keyword based search to AI search based, and I see customers coming and say, oh, we want to actually, we want to have a chat bot, right?
183
00:20:04,590 --> 00:20:20,750
What we see is that chat bots are not everybody's darling, right? If the users come to a website, they might see the bubble, they might see the pop up, and they totally feel annoyed by it. That is my experience.
184
00:20:20,750 --> 00:20:37,110
Nobody's really because also chat bot quality has been so variant over time is that people are like, oh, I can't trust the chat bot. And if I, it will not actually help me on my use case.
185
00:20:38,030 --> 00:20:51,490
Yesterday, I had an, I was listening to a phone call that my wife did calling at the doctor's appointment and they were really like, super duper, you know, you don't speak to anybody anymore.
186
00:20:51,790 --> 00:21:02,550
It's a chat bot. It was a very intricate and intelligent chat bot. But the use case that my wife had, did not really fit into the logic.
187
00:21:02,550 --> 00:21:25,510
So I think that is something also that if let's say, imagine going back to search, you have this very specialized search that you want to have, like in for example, you want a sales agent, able to find like records of, you know, customer records, you want to find technology, data, all this kind of stuff.
188
00:21:25,510 --> 00:21:40,090
I think that the search there will need to be become more intelligent than just, you know, a chat bot, right? Not something that just reads something, but you need to have like, actually like specialized skills in that search.
189
00:21:40,090 --> 00:21:49,710
And you need to have like, like a router that is actually taking care of, okay, this is the intention.
190
00:21:49,990 --> 00:21:54,910
So this person is looking for a price, then I need to go and look for this.
191
00:21:55,230 --> 00:22:09,550
And then I need to have the search agent, for example, or this is another person who is like actually looking for contact information or looking for, you know, where can I call?
192
00:22:09,550 --> 00:22:11,830
What can I call you or stuff like this?
193
00:22:12,050 --> 00:22:18,790
So you will need to have like a separation of, of, of, of, of, of, of different search for that.
194
00:22:18,890 --> 00:22:20,350
I think I had prepared.
195
00:22:21,850 --> 00:22:22,730
Let me look.
196
00:22:24,030 --> 00:22:24,810
Do you know what?
197
00:22:24,890 --> 00:22:28,150
Let me, I think I can do that just once.
198
00:22:28,150 --> 00:22:28,450
Yeah.
199
00:22:29,310 --> 00:22:30,590
There's this one.
200
00:22:30,970 --> 00:22:31,330
Yeah.
201
00:22:31,470 --> 00:22:31,470
202
00:22:31,590 --> 00:22:31,690
Okay.
203
00:22:31,790 --> 00:22:32,050
We go.
204
00:22:32,750 --> 00:22:34,330
We just skip.
205
00:22:35,370 --> 00:22:36,670
We just keep this.
206
00:22:37,270 --> 00:22:39,710
I just want to show what I'm talking about.
207
00:22:39,950 --> 00:22:50,130
I'm just actually trying to explain, you know, an architecture that, you know, can be extended on the horizon horizontal.
208
00:22:50,130 --> 00:23:11,330
You know, it's not just only a search or program finder or expert finder, or you can send audio output to it, but, you know, actually, and this is where I see that where, where software or solutions that you want to offer to your customers where those solutions need to go.
209
00:23:11,330 --> 00:23:19,430
So, not, not looking into something that solve everything, but solves one thing good.
210
00:23:19,730 --> 00:23:19,830
Right.
211
00:23:19,930 --> 00:23:27,530
So this is actually what I see as a, as a, as a, as a, as a pattern that people should have a look at.
212
00:23:29,370 --> 00:23:35,570
So we've seen, we, we, we see that yesterday, two days ago, I was in a call with a customer.
213
00:23:36,430 --> 00:23:38,770
They have this, they have thing.
214
00:23:45,250 --> 00:23:45,730
Ments.
215
00:23:46,870 --> 00:23:50,070
Oh, are we still there?
216
00:23:50,210 --> 00:23:51,730
I think we were for a moment.
217
00:23:51,950 --> 00:23:54,450
There was like a, there was like a spinning wheel.
218
00:23:54,730 --> 00:24:01,510
Um, so this customer, they've identified eight areas of interest for the users.
219
00:24:01,510 --> 00:24:01,950
Right.
220
00:24:02,070 --> 00:24:12,230
And what they did was like, actually, they sent in 350 questions that they thought about in around those, those, those areas.
221
00:24:12,530 --> 00:24:17,950
For example, I want to get a, a, a, a license for fishing.
222
00:24:18,370 --> 00:24:18,670
Right.
223
00:24:18,990 --> 00:24:24,350
Um, so it's, it's, it's, it's, it's a very nice, uh, it's a nice place in Austria.
224
00:24:24,350 --> 00:24:27,350
Um, and, and they have all kinds of questions.
225
00:24:27,930 --> 00:24:32,990
People are going to the website and asking the website, okay, how can I get this?
226
00:24:33,150 --> 00:24:36,250
How can I apply for a tourism license?
227
00:24:36,250 --> 00:24:40,110
Because I want to sublet my apartment and so on and so forth.
228
00:24:40,210 --> 00:24:43,730
But also who can help me in case of this or that.
229
00:24:43,870 --> 00:24:44,270
Right.
230
00:24:44,350 --> 00:24:46,770
And also be, be able to disambiguate.
231
00:24:47,090 --> 00:24:47,590
All right.
232
00:24:47,770 --> 00:24:49,930
We're not responsible for that information.
233
00:24:49,930 --> 00:24:53,150
We can give you the information where you can look for.
234
00:24:53,310 --> 00:24:53,430
Right.
235
00:24:53,430 --> 00:25:03,910
For example, no, um, your privacy license is issued by another, um, institution, but you can find information about this.
236
00:25:04,010 --> 00:25:16,850
So now imagine transitioning this website into a website that actually helps you find those, um, find those information on a list or maybe as a.
237
00:25:17,370 --> 00:25:17,430
Yeah.
238
00:25:18,230 --> 00:25:22,330
Like, like, even like a written answer that you can copy down and so on and so forth.
239
00:25:22,330 --> 00:25:25,350
Um, with the sources and with the links.
240
00:25:25,530 --> 00:25:29,630
So I, I think that is what I can predict is what we're going to see.
241
00:25:29,830 --> 00:25:31,290
That's pretty often now soon.
242
00:25:31,490 --> 00:25:35,270
Um, that websites will start to become something like a mini.
243
00:25:35,550 --> 00:25:41,630
A clothes or mini GPTs where people will interact, will interact with that.
244
00:25:41,630 --> 00:25:59,630
Um, so, um, um, what I see also what is happening is that, um, yeah, well, they're overspending money into, you know, trying to do this one thing.
245
00:26:00,390 --> 00:26:03,710
They try to do it with different solutions.
246
00:26:03,710 --> 00:26:13,750
They try to incorporate it or, or, or, but, you know, we, we think people should look for a more integrated solution, uh, than just a solution that is giving the.
247
00:26:13,750 --> 00:26:18,950
So, um, so what they're doing is also that we're seeing is that they're.
248
00:26:19,010 --> 00:26:25,990
They're trying to solve problems, like, especially like problems with a, uh, retrieval augmented generation.
249
00:26:25,990 --> 00:26:31,170
You know, they're trying to, to, to solve problems just very superficially, right?
250
00:26:31,330 --> 00:26:33,410
Just, okay, let's chunk everything.
251
00:26:33,670 --> 00:26:40,650
Let's have super big context, uh, windows and just then hope and pray that the search gives you, gives you an answer.
252
00:26:40,830 --> 00:26:53,350
Uh, and, and, and, um, and, and, and before, um, they, um, you know, so before they do something, they should really actually repair their content.
253
00:26:53,350 --> 00:27:05,430
Right. This is what I really tell everyone, have a look at your website, have a look at the zero results, have a look at, or for example, am I really talking the language of my visitor or am I talking my own lingo?
254
00:27:05,670 --> 00:27:23,330
Right. And this is something that I think we have, we are observing and the more a website is trying to, um, how you call it, uh, trying to, um, you know, output the organization that they are, the more you'll find a gap between the
255
00:27:23,330 --> 00:27:35,710
users and the content of the website. So I think that is something that what, but we are in, we are lucky nowadays that we can go and we can actually scrape and get all the, all the content into a format.
256
00:27:35,710 --> 00:28:05,470
We can feed this to LLM with a special prompt and we can tell it, for example, all right, please rewrite it in a way that somebody that's, that is in grade five or something. So, and so, and so can understand this. And also please, um, please write it always in the format where the question is answered in the very, very beginning, not at the very, very end. Uh, like in a movie, you know, you watch a movie and then that, you know, just until the end, uh, and so on and so forth. Um, so,
257
00:28:06,030 --> 00:28:33,070
um, that is just what's on my mind at the moment, Chris. Um, um, um, so my, uh, takeaway here and on, if you're looking, trying to find, uh, uh, some, some advice from me, it's like, please see what people are asking. What are their questions? Um, and, and also in, in, as an agency, you should actually speak to the stakeholders, you know, not just only to the IT department and have the stakeholders.
258
00:28:33,070 --> 00:28:49,970
And you also need to explain to the stakeholders that AI search is not just adding a magical powder on top of it. It's actually something that only can only, uh, work if you take care of your content, right? So this is the second fixed content and metadata.
259
00:28:51,570 --> 00:29:15,870
Even try to think of, you know, your own model, you know, what is my industry model? Um, do I have things that I'm, that, that, that is not in a large language model. So might be, you know, that you are, are the specialist of that, that valve production, and you are very specialized and, and, and, and things like that are not in Wikipedia, for example. Right.
260
00:29:15,870 --> 00:29:44,870
Right. And, and, and, and, and, and, and, um, last thing, and is first concentrate on the retrieval layer and add features later as a, as a, as a, as a, as a view, right? For example, but, you know, step-by-step and always, uh, supervise, um, what is happening with your search, log, whatever you can log it in a good way.
261
00:29:45,230 --> 00:30:00,810
Um, even ask people for, uh, feedback. How is the search? I mean, you can go like, you know, do this net score, uh, test, uh, you can do, um, um, you can do, you know, thumbs up, thumbs down for the search. You can log that and so on and so forth.
262
00:30:00,910 --> 00:30:14,810
This is stuff that we are doing for, on, on type of three, for example. Uh, we even mimicking at the moment, you know, the, uh, Google search experience people, you know, if you go to Google, you have this AI search snippet that is coming up.
263
00:30:14,870 --> 00:30:30,570
I think that is something that will works, will work way, way better than just a chatbot, right? Because people are learning at the moment that Google is offering them, them that, and they will happily use that if there are on your website.
264
00:30:30,570 --> 00:30:43,090
And this will keep people on the website, even way better than just a chatbot. And, and, and, and, you know, um, we'll see that people will start to talk to websites, um, uh, sooner or later.
265
00:30:43,090 --> 00:30:53,150
Uh, people, you know, they, nowadays, they always also even talk in public with the phone, you know, like holding a knicker bread, you know?
266
00:30:53,590 --> 00:30:54,050
Yes.
267
00:30:54,170 --> 00:31:02,490
So, uh, we will see that even more and, and, uh, yeah. So that is what I, I, I can, I can tell about, about this in our 30 minutes.
268
00:31:02,490 --> 00:31:09,990
Um, I'm, I'm, I'm, I'm, I'm happy to connect to everyone here, um, on, on, on, on, on, on, on, on actual questions and so on.
269
00:31:10,770 --> 00:31:18,950
Oh, very cool. The, the challenge here that we have is that the time goes by so quickly, because as you're talking, I'm coming up with 18 more questions.
270
00:31:18,950 --> 00:31:24,890
I'm wondering if you do have just a minute more, and this might be a nice artifact for later.
271
00:31:25,190 --> 00:31:31,190
One of your slides that we passed through at the beginning was about describing RAG.
272
00:31:31,670 --> 00:31:36,590
Do you mind just taking a moment to describe that a little bit more?
273
00:31:37,210 --> 00:31:45,850
Yeah, actually, regular search is like, for example, regular search as we see it is keyword search, right?
274
00:31:45,850 --> 00:31:48,570
You have, let's say, imagine you have a website.
275
00:31:49,270 --> 00:31:52,610
The website has a title and it has a content block, right?
276
00:31:52,850 --> 00:32:06,570
So if you index this into index, like a regular index, Apache Solar index, you end up taking the title and put it into a title field.
277
00:32:06,570 --> 00:32:10,150
You end up having the content and put it into a content field.
278
00:32:10,350 --> 00:32:27,890
And then your search is actually going with a search query and trying to find, okay, do I have, do I have, I'll call it, hits on the title or on the content, right?
279
00:32:28,010 --> 00:32:33,770
This works pretty well if you index it right.
280
00:32:33,770 --> 00:32:34,570
So, okay.
281
00:32:36,010 --> 00:32:42,490
But, you know, the interaction is search, query, results list.
282
00:32:42,590 --> 00:32:45,890
That is like actually the normal search pattern.
283
00:32:47,470 --> 00:32:58,390
RAG is actually something where you have, you index also your documents into a, let's say, also into an index.
284
00:32:58,390 --> 00:33:04,710
But also you index, you know, you add additional embeddings to it.
285
00:33:04,830 --> 00:33:11,890
That is like AI tokens for understanding, you know, understanding what's in the text.
286
00:33:11,890 --> 00:33:18,470
And then the search query gets also translated into those tokens.
287
00:33:19,130 --> 00:33:31,190
And with that, you can search for something and have the large language model actually, or have actually the large language model look for similar tokens.
288
00:33:31,190 --> 00:33:37,410
You know, for example, you know that a dog and a cat are animals.
289
00:33:38,050 --> 00:33:44,070
And you know that your text document is a document, right?
290
00:33:44,190 --> 00:33:47,050
And they are really, they have nothing to do with each other.
291
00:33:47,210 --> 00:33:49,830
But, you know, the text document with animals.
292
00:33:49,830 --> 00:33:55,830
But, you know, that a cat and a dog can be, are actually animals and that they live in your house.
293
00:33:55,990 --> 00:34:01,930
And if you actually would ask a search, okay, what kind of animals can I have in my house?
294
00:34:02,270 --> 00:34:16,350
Then the search would answer to you, dogs and cats are what you can, what you often humans use as animals or have animals in their, as companions.
295
00:34:16,350 --> 00:34:39,170
So, so actually the rack does, does this whole mechanics, you know, they, they, you know, it indexes, it translated into embeds, it takes big documents and cuts them down into smaller chunks because also the large language model, they don't have indefinite, indefinite room for that.
296
00:34:39,170 --> 00:34:42,990
So you cannot just ask all your documents at the same time.
297
00:34:43,130 --> 00:34:46,790
So you need to have a, come up with some, some kind of a strategy on that.
298
00:34:46,970 --> 00:34:50,830
So this is often like cutting the documents into something that is smaller.
299
00:34:51,890 --> 00:34:59,150
With that, you have often problems that the context get lost, you know, because the chunks are not interconnected.
300
00:34:59,150 --> 00:35:07,950
Well, some, some rack systems have that, that they actually start to write all kinds of stuff around it and so on and so forth.
301
00:35:09,510 --> 00:35:14,510
So, and you have two moments where the large language model is actually taking care of understanding.
302
00:35:14,510 --> 00:35:22,810
So the moment where you send us the send a query to, to the doc, to the document storage or to the retrieval.
303
00:35:23,050 --> 00:35:26,410
So this is where the large language model is doing something.
304
00:35:26,970 --> 00:35:39,110
And when the results list comes back, the large language model is actually taking care of augmenting what has been generated by the search, the search engine here.
305
00:35:39,110 --> 00:35:55,410
And, and, and with that, you get like actually, you know, proper answers, for example, you know, you could tell the agent or you can, you could actually tell the large language model to the, okay, just answer in super short questions or write, write a novel out of it.
306
00:35:55,530 --> 00:35:55,610
Right.
307
00:35:56,010 --> 00:35:59,630
So that, that is what, what people can do with rack nowadays.
308
00:36:11,790 --> 00:36:16,050
So, that, that, that, that, that, that, that, that was a good explanation.
309
00:36:16,450 --> 00:36:21,070
I'm, I, I, by no ways and IT experts in that much, but.
310
00:36:22,070 --> 00:36:23,770
Hopefully that helps me.
311
00:36:23,770 --> 00:36:32,090
People ask me and now I can refer them to, you know, go to our video at 31 minutes in and you'll hear an explanation.
312
00:36:32,890 --> 00:36:33,850
That's really great.
313
00:36:33,990 --> 00:36:35,050
I really appreciate it.
314
00:36:36,010 --> 00:36:38,130
And time has gone by quickly.
315
00:36:38,530 --> 00:36:39,210
Oh yeah.
316
00:36:39,530 --> 00:36:40,130
Waiting for dinner.
317
00:36:41,330 --> 00:36:42,690
Thank you for joining us.
318
00:36:42,770 --> 00:36:43,950
Olivia, just super cool.
319
00:36:44,250 --> 00:36:46,330
Just hanging out with you and chatting.
320
00:36:46,570 --> 00:36:50,910
And I hope folks who are listening now or in the future, get a lot of value out of this.
321
00:36:50,910 --> 00:36:51,150
Yeah.
322
00:36:51,870 --> 00:36:53,590
Olivia, how can people reach you?
323
00:36:53,790 --> 00:36:55,990
We'll, we'll post your deets down below in the comments.
324
00:36:56,050 --> 00:36:56,230
Yeah.
325
00:36:56,470 --> 00:37:02,530
I think if you're on LinkedIn, you can, you will find me actually on that webinar link here.
326
00:37:02,950 --> 00:37:05,850
You can search for Olivier Doberkau.
327
00:37:06,570 --> 00:37:07,930
My name is pretty rare.
328
00:37:08,510 --> 00:37:12,530
You will find a lot of researchers, but you will find me.
329
00:37:14,310 --> 00:37:14,790
Yeah.
330
00:37:16,910 --> 00:37:18,130
How can they reach me?
331
00:37:18,210 --> 00:37:19,410
I mean, I have an email.
332
00:37:19,410 --> 00:37:21,270
I'm still an email guy.
333
00:37:22,730 --> 00:37:29,870
You can, you can send me an email to Olivier.Dobrikau at DKD.DE.
334
00:37:31,550 --> 00:37:32,450
That's awesome.
335
00:37:33,190 --> 00:37:34,150
Do something with it.
336
00:37:35,150 --> 00:37:36,290
That's really great.
337
00:37:36,810 --> 00:37:40,270
Thanks to everyone who took some time today to listen in to us.
338
00:37:40,950 --> 00:37:44,450
A really great 50th episode anniversary.
339
00:37:44,450 --> 00:37:47,450
And it's a real pleasure of mine to host you.
340
00:37:48,130 --> 00:37:50,950
Hope to see you again soon in person somewhere in the world.
341
00:37:51,130 --> 00:37:56,410
And we'll be following you and listening to what you have to say and what you're thinking about.
342
00:37:57,130 --> 00:38:02,970
Just a little shout out to folks who are maybe curious about open source.
343
00:38:03,810 --> 00:38:05,550
It really hit me on the head.
344
00:38:05,550 --> 00:38:10,770
We've been a little bit more closed source over the last 10 years.
345
00:38:11,010 --> 00:38:13,730
But just a little story before we go.
346
00:38:13,930 --> 00:38:20,430
I was talking about, I was talking to a friend about why he's using like Claude Fable for everything he's doing.
347
00:38:20,430 --> 00:38:24,650
And he said, well, I don't know, like what's another hundred bucks?
348
00:38:25,510 --> 00:38:31,370
And we've been doing a lot of work with the open source harnesses, both OpenClaw and the Hermes agent.
349
00:38:32,770 --> 00:38:36,690
And I was chatting with my other pals in Morocco and Cambodia.
350
00:38:37,230 --> 00:38:42,610
And they're like, we don't even have like 20 bucks just kicking around to go.
351
00:38:42,850 --> 00:38:44,970
So there's this divide happening as well.
352
00:38:45,210 --> 00:38:50,590
So anyone who's out there who's kind of curious about open source, those communities should be acknowledged.
353
00:38:50,850 --> 00:38:53,290
And, you know, maybe check them out and get involved as well, too.
354
00:38:53,410 --> 00:38:55,390
I just wanted to do a shout out there.
355
00:38:55,570 --> 00:38:55,930
For sure.
356
00:38:56,890 --> 00:39:04,670
I think that could be another episode about how open source is misinterpreted in actually also in the AI world.
357
00:39:04,870 --> 00:39:07,450
That there's a lot of discussions around it.
358
00:39:07,530 --> 00:39:11,490
I mean, we in the Western world, you know, it's no problem.
359
00:39:11,710 --> 00:39:13,790
This has become something that we can actually use.
360
00:39:13,930 --> 00:39:22,970
But many, many years ago when I was in Cambodia, that's the moment when we said we need to, as Type 3 Association now,
361
00:39:22,970 --> 00:39:28,810
we need to offer a community membership to people that is reasonable.
362
00:39:29,210 --> 00:39:37,050
And, you know, it was actually the hate of a normal membership was 125 euros.
363
00:39:37,050 --> 00:39:45,650
And this is no way something that someone who's working on Type 3 in Cambodia or in Tunisia can, you know.
364
00:39:45,890 --> 00:39:51,730
And even though, I mean, now we have a membership that is nine euros a year.
365
00:39:51,730 --> 00:39:57,050
And it's still hard for those people to become members.
366
00:39:57,330 --> 00:40:02,350
And I have my girls in South Africa that I'm supporting with Type 3.
367
00:40:02,350 --> 00:40:04,910
And I'm supporting them being the mentor for them.
368
00:40:05,410 --> 00:40:13,730
And even that, I mean, you start talking about, you know, stuff that is incubated.
369
00:40:14,390 --> 00:40:17,610
No, it's keeping them from doing stuff.
370
00:40:17,610 --> 00:40:25,090
And they have no, yeah, like, you know, going to a place, they don't have the money to take the bus, right?
371
00:40:25,290 --> 00:40:28,170
They can go somewhere, access a computer to use it.
372
00:40:28,550 --> 00:40:30,710
Nevertheless, they don't have a personal computer.
373
00:40:30,970 --> 00:40:35,090
And this opens up a lot of things.
374
00:40:35,230 --> 00:40:37,170
But that is maybe a different episode.
375
00:40:37,370 --> 00:40:37,610
Okay.
376
00:40:37,750 --> 00:40:38,490
I'm looking forward to that.
377
00:40:38,490 --> 00:40:40,130
We're pretty privileged, Chris.
378
00:40:40,450 --> 00:40:41,010
It's true.
379
00:40:41,210 --> 00:40:42,850
It was really great talking to you.
380
00:40:43,030 --> 00:40:46,810
And I'm really looking forward into running into you again.
381
00:40:47,030 --> 00:40:51,350
And greetings to the folks at .fusion.
382
00:40:52,410 --> 00:40:54,950
I know that pretty nice people there.
383
00:40:55,230 --> 00:40:57,250
And you're a pretty good guy.
384
00:40:57,590 --> 00:41:01,570
So I'm sending you this or that.
385
00:41:02,410 --> 00:41:03,750
Can't go wrong with good vibes.
386
00:41:04,010 --> 00:41:06,830
So on that note, I'm going to click some buttons.
387
00:41:07,110 --> 00:41:09,150
Maybe Olivia just hang with me for a moment.
388
00:41:09,690 --> 00:41:10,910
But thanks to everyone.
389
00:41:10,910 --> 00:41:14,950
And we'll see you in the next few weeks for our next episode.
390
00:41:15,150 --> 00:41:16,510
So have a great day out there, everybody.
391
00:41:17,090 --> 00:41:17,570
Bye.
392
00:41:21,350 --> 00:41:21,390
Bye.
00:00:09,220 --> 00:00:11,540
All right, we are live.
2
00:00:12,440 --> 00:00:18,940
Thank you to everyone who is joining us during their lunch in and around the Toronto area
3
00:00:18,940 --> 00:00:26,540
or wherever else you are around the world or in the future on YouTube or on our podcast
4
00:00:26,540 --> 00:00:27,000
channels.
5
00:00:27,300 --> 00:00:28,100
Much thanks.
6
00:00:28,720 --> 00:00:31,680
This is a super special episode today.
7
00:00:31,920 --> 00:00:35,060
Olivia, I don't know if you knew this, but this is episode 50.
8
00:00:36,240 --> 00:00:43,560
We started this as just an idea to be helping and sharing with our friends in the content
9
00:00:43,560 --> 00:00:49,980
operation space and the CMS space to make themselves available to share what they're
10
00:00:49,980 --> 00:00:54,520
thinking, what they're up to, what are the trends in content operations.
11
00:00:55,800 --> 00:00:57,880
So thanks for joining everyone.
12
00:00:57,880 --> 00:00:59,540
My name is Chris.
13
00:00:59,940 --> 00:01:01,700
I'm the founder of Dot Fusion.
14
00:01:01,700 --> 00:01:07,760
We're essentially a content operations agency with most of our product being some kind
15
00:01:07,760 --> 00:01:09,860
of larger website or mobile app.
16
00:01:09,860 --> 00:01:15,660
But we really invest a lot of time in and around the management of content as it moves through
17
00:01:15,660 --> 00:01:16,580
those supply chains.
18
00:01:16,760 --> 00:01:22,720
And I'm just super fortunate to have run into a lot of thought leaders in the space.
19
00:01:22,720 --> 00:01:31,840
So for episode 50, we're like super stoked to invite Olivier de Barcaux from a variety of different
20
00:01:31,840 --> 00:01:36,860
institutions, DKD, as well as I believe you're the president of TYPO3.
21
00:01:38,160 --> 00:01:40,480
You're also staying up a little bit late.
22
00:01:40,900 --> 00:01:41,560
You're in Germany.
23
00:01:41,880 --> 00:01:45,880
So thank you for spending your evening with us.
24
00:01:46,840 --> 00:01:52,900
Maybe just before we get started, the whole idea is just some great conversation.
25
00:01:53,280 --> 00:01:56,060
I do my best to monitor the LinkedIn channels.
26
00:01:56,060 --> 00:02:01,220
If you're watching this live on LinkedIn, if you do have some questions, just let me know.
27
00:02:01,360 --> 00:02:04,740
We'll be happy to answer them or do our best to follow up with afterwards.
28
00:02:06,180 --> 00:02:09,880
But as I said before, time goes by when we've got about a half an hour.
29
00:02:10,160 --> 00:02:12,880
So maybe Olivier, do you want to introduce yourself?
30
00:02:13,260 --> 00:02:17,760
I'm kind of coming off a really great time hanging out with you in Montreal at Yanis Boy
31
00:02:17,760 --> 00:02:19,280
CMS Experts Group.
32
00:02:19,860 --> 00:02:21,660
Kudos to that organization.
33
00:02:22,720 --> 00:02:23,600
Hi, welcome.
34
00:02:25,880 --> 00:02:26,300
Thank you.
35
00:02:26,520 --> 00:02:27,580
Thank you for having me.
36
00:02:27,740 --> 00:02:28,960
Thank you for the invitation.
37
00:02:29,460 --> 00:02:32,840
I know that this invitation has been running for long.
38
00:02:32,840 --> 00:02:38,360
So it really needed us having that one moment in Montreal.
39
00:02:39,160 --> 00:02:45,680
I remember us sitting in front of this depaneur having a cold water.
40
00:02:46,360 --> 00:02:50,640
And so it was really great to meet you again.
41
00:02:50,940 --> 00:02:52,880
I think we met two years ago.
42
00:02:52,880 --> 00:03:08,280
We found out that we knew each other, you know, by one separation of, you know, and that we actually kind of ran into each other maybe even 11 years ago in your time in Cambodia.
43
00:03:08,280 --> 00:03:12,700
So I think that is like kind of, you know, the rapidity knocks sometimes.
44
00:03:13,020 --> 00:03:13,060
Yes.
45
00:03:13,060 --> 00:03:19,140
And then so this is actually a pretty nice way to talk to each other.
46
00:03:19,960 --> 00:03:25,080
Hopefully we will have something for the audience today.
47
00:03:26,200 --> 00:03:28,840
So what can I tell about myself?
48
00:03:29,120 --> 00:03:32,480
I'm 50 years, 58 years old.
49
00:03:32,780 --> 00:03:34,180
Don't feel like that.
50
00:03:34,360 --> 00:03:43,120
I feel currently being 30 again, you know, going back into the times when we created our agency.
51
00:03:43,120 --> 00:03:51,200
I think we are have the same history of, you know, working in the industry and and also being self-taught about a lot of things.
52
00:03:51,440 --> 00:04:02,920
And, you know, nowadays, I guess everyone is agreeing with that is that with the upcoming and more and more, you know, AI in our life.
53
00:04:02,920 --> 00:04:07,980
We all become more versatile on what we are doing.
54
00:04:08,200 --> 00:04:13,480
So I've started again to code and I've done a lot of stuff.
55
00:04:14,180 --> 00:04:15,900
So you're right.
56
00:04:16,060 --> 00:04:20,100
I created DKD at university.
57
00:04:21,160 --> 00:04:25,340
We went into incorporation 1990, 1998.
58
00:04:26,120 --> 00:04:28,720
So almost 30 years from now.
59
00:04:30,480 --> 00:04:34,080
Yeah, the company is now 50 employees.
60
00:04:35,060 --> 00:04:39,320
Most of what we are doing is Type 3 CMS.
61
00:04:39,640 --> 00:04:42,520
It's an open source CMS, PHP based, MySQL.
62
00:04:44,180 --> 00:04:47,640
Has a community, let's say, in the DACH area.
63
00:04:48,040 --> 00:04:49,840
So Germany, Switzerland, Austria.
64
00:04:50,140 --> 00:04:54,560
But it has also strong foundations in, you know, in Europe.
65
00:04:54,560 --> 00:05:05,760
And lately we've been going back to visit and to establish our, you know, our relations with the U.S. market.
66
00:05:06,060 --> 00:05:13,760
This is something that if you look closely, we had an opportunity many, many years ago.
67
00:05:13,760 --> 00:05:17,260
So I think at that time was not right.
68
00:05:17,400 --> 00:05:30,320
But now we find it very interesting to see that people are actually re-discovering Type 3, a very, very strong, with a lot of features CMS out of the box.
69
00:05:30,520 --> 00:05:32,380
So that is pretty nice.
70
00:05:32,380 --> 00:05:39,580
And in regards of what DKD has been doing for Type 3, we integrated Solar Search.
71
00:05:39,860 --> 00:05:46,120
That is like an open source search engine for Type 3 many, many years ago.
72
00:05:46,120 --> 00:05:48,720
We've been keeping it up to date.
73
00:05:48,720 --> 00:05:58,180
We found a way on how we can fund open source development in a group of, you know, I would call them partners or sponsoring partners.
74
00:05:58,480 --> 00:06:02,700
And this has been keeping up Solar up and running.
75
00:06:03,320 --> 00:06:15,280
And so lately we released the Solar version, the extension version 14, which is the matching version for Type 3 version 14.
76
00:06:15,280 --> 00:06:19,480
And we're seeing that this has a lot of adaption.
77
00:06:20,020 --> 00:06:29,040
If you go to packages and you look how much installations have been in the last years, I think we're speaking about one and a half million of installations.
78
00:06:30,500 --> 00:06:38,260
So that is interesting to see that search is still a thing, especially in the enterprise sector.
79
00:06:38,680 --> 00:06:43,920
So I think we will be talking today about, OK, where search is going.
80
00:06:45,280 --> 00:06:52,980
So how, let's say, keyword search is evolving into something that is more like a conversation with the website.
81
00:06:53,640 --> 00:07:01,260
We will be maybe seeing what you should do and not do if you are using that.
82
00:07:01,260 --> 00:07:13,040
But also, I think that we will see that we're all going or we're all on a journey and that there's no silver bullet yet.
83
00:07:13,620 --> 00:07:16,140
I think we've we will.
84
00:07:16,700 --> 00:07:22,960
But let's see if we have some people finding having questions about that.
85
00:07:23,620 --> 00:07:33,320
Awesome. So well, we titled the show today searches the interface, which can provoke a lot of interesting thoughts.
86
00:07:33,620 --> 00:07:43,560
But maybe give us your take on that and and why that's resonating as a phrase and how that kind of calibrates into how we're approaching search in the industry.
87
00:07:43,560 --> 00:07:51,360
Yeah, I think search is something quite very human.
88
00:07:51,600 --> 00:07:57,760
Right. It is something that is in our DNA or in our way to survive.
89
00:07:58,040 --> 00:08:09,820
Right. If we go back to the caveman times, hunting and searching for food is something that that that keeps you keeps you alive.
90
00:08:09,820 --> 00:08:20,360
Right. So I've many, many years ago, I heard a very good, you know, very good establishment of what search is.
91
00:08:20,500 --> 00:08:29,880
And actually, people started to also to speak about the scent of search actually know that you could smell search like, you know, when is the search good?
92
00:08:30,240 --> 00:08:39,280
Because people, you know, at the end, you know, if you let's go back to maybe a bear and something that happens to be in Canada, for example.
93
00:08:40,160 --> 00:08:45,260
You know, you know, the bear knows exactly where to find the food that he's craving for.
94
00:08:45,460 --> 00:08:49,160
Right. If he wants to eat berries, he knows where the berries are.
95
00:08:49,300 --> 00:08:58,620
But this is something that has happened to know that they know because they're smelling it, maybe, but also because they're remembering the places that they have been.
96
00:08:58,620 --> 00:09:13,940
So if you if you would, let's say, compare the search of a human on a website, then this human visiting the website is always expecting the search to behave like something that he has experienced before.
97
00:09:14,240 --> 00:09:16,720
Right. So this is my first hypothesis.
98
00:09:16,720 --> 00:09:27,460
I think you're also remembering Jakob Nielsen, who said we should all do things that are coping to the expectation of people.
99
00:09:27,600 --> 00:09:32,160
Right. We should never surprise people, let's say, with a super duper interface.
100
00:09:33,240 --> 00:09:39,800
I remember when we started to do web design is that we experimented a lot, you know, on, you know, concepts.
101
00:09:39,800 --> 00:09:46,160
And I don't know if you have been around with AOL dial ups.
102
00:09:46,420 --> 00:09:47,840
Yes, not in the disk days.
103
00:09:48,120 --> 00:09:52,080
There was like a thing called eWorld from Apple many, many years ago.
104
00:09:52,080 --> 00:09:55,820
You know, you will dial up and you end up like in a virtual city.
105
00:09:56,100 --> 00:10:11,880
And I think remembering the very first websites that we did was they were quite playful because we had to do also be like we had to, you know, to have like a house and have a like a post office and stuff like this where people could actually find their mail.
106
00:10:12,060 --> 00:10:19,200
So I think at that time, the interfaces became something like kind of, you know, you know, how do you call it?
107
00:10:20,180 --> 00:10:23,320
You know, you had to present something that they could relate to.
108
00:10:23,440 --> 00:10:27,200
Nowadays, I think interfaces have become more and more abstract.
109
00:10:27,940 --> 00:10:30,920
We are seeing that the hamburger is there.
110
00:10:31,080 --> 00:10:34,140
So people understand, OK, this is navigation on the mobile phone.
111
00:10:34,260 --> 00:10:39,920
We know the L navigation, you know, with the top top navigation of points and the sub navigation.
112
00:10:39,940 --> 00:10:41,960
And people do understand this pretty well.
113
00:10:42,200 --> 00:10:47,160
And they see a search field and they see a search button.
114
00:10:47,160 --> 00:10:48,220
They enter them.
115
00:10:48,540 --> 00:10:49,340
They enter keywords.
116
00:10:49,880 --> 00:10:53,420
And with that, they find stuff on the website or not.
117
00:10:53,540 --> 00:11:04,840
If you have a good search engine, you will you will run into re-rankings and you run into like, OK, how much times these the keyword is in the actual document.
118
00:11:06,100 --> 00:11:15,760
There's like this algorithms called like BM25 that does that do a lot of, you know, re-ranking and sorting and relevance and so on.
119
00:11:15,760 --> 00:11:20,580
I think that is what people are still expecting nowadays.
120
00:11:21,100 --> 00:11:24,200
But the experience is changing.
121
00:11:24,440 --> 00:11:29,000
People are going now to answering machines or AI answering machines.
122
00:11:29,000 --> 00:11:31,220
They are typing in their questions.
123
00:11:31,220 --> 00:11:32,260
They are typing in their questions.
124
00:11:32,260 --> 00:11:33,380
They are receiving answers.
125
00:11:34,300 --> 00:11:47,360
And with that, yeah, they might go somewhere else and might go to might come to your website and maybe expect that your website search has it becomes more and more of a conversation.
126
00:11:47,360 --> 00:12:05,440
So, therefore, and maybe and in the future, we will only see, you know, even no longer having human visitors to the website, but just only search agents or agents that are searching for information.
127
00:12:05,440 --> 00:12:10,580
So I think we will be seeing that that that that stuff happening.
128
00:12:11,700 --> 00:12:33,420
You've seen that in the last months and weeks of the of this year, we have seen that agents have become something that end users are getting to start to work with, at least if you are in the northern American hemisphere, you're you have very fast access to that kind of stuff.
129
00:12:33,420 --> 00:12:46,120
So we in Europe are kind of, yeah, how you call it, we're left behind because, you know, we have this AI act and we want to have more security on on that kind of stuff.
130
00:12:46,260 --> 00:12:52,200
So I think that Facebook and all the guys, they all want us not to come and sue them.
131
00:12:52,200 --> 00:12:56,920
So but what does it what does it mean for search?
132
00:12:57,120 --> 00:13:16,780
I mean, we'll see we'll be seeing that our search and search infrastructure will need to change into something that will need to cope those different users or different users on the website like the human users expecting a more, you know, an interface that is more and the
133
00:13:17,640 --> 00:13:29,840
is more understanding your search intention, maybe even, you know, supercharging the search intention with the search experience of others.
134
00:13:29,840 --> 00:13:43,700
And we will see also search interfaces that will be totally programmatic, you know, and then, you know, and advertised in only agent readable formats.
135
00:13:44,360 --> 00:13:46,920
So that is that is what I'm seeing.
136
00:13:47,360 --> 00:13:49,740
But what does it mean for the website?
137
00:13:50,540 --> 00:13:50,680
Right.
138
00:13:51,820 --> 00:14:06,760
There's there's a thing that if you're if you you you work with search, there's one one saying the quality of data is also say also how you call it?
139
00:14:07,880 --> 00:14:11,300
It's influencing the search experience, right?
140
00:14:11,720 --> 00:14:24,240
You cannot search for a needle in the haystack if you also have other stuff in this haystack and actually those things also look like needles.
141
00:14:24,420 --> 00:14:30,260
Right. So you need to to be very unique to be able to disimple grade and and so on and so forth.
142
00:14:30,260 --> 00:14:33,740
So or plainly said junk in junk out.
143
00:14:33,960 --> 00:14:35,720
Right. If you you have this.
144
00:14:36,280 --> 00:14:46,320
So what I'm still seeing nowadays is that the content and the let's say the content that gets indexed into a search index.
145
00:14:48,380 --> 00:14:58,260
It's just index and people do not really care about, OK, what what kind of search am I am I'm giving to my users?
146
00:14:58,330 --> 00:15:02,950
Right. I've done a website for that.
147
00:15:04,050 --> 00:15:05,810
It's it's a test website.
148
00:15:06,090 --> 00:15:07,170
It's it's at the moment.
149
00:15:07,210 --> 00:15:07,830
It's in German.
150
00:15:08,010 --> 00:15:10,570
It's called such quality search quality.
151
00:15:11,210 --> 00:15:16,570
And you can go to the website and you can actually enter the web, the domain of the website.
152
00:15:16,570 --> 00:15:30,810
And with that, an AI is actually a bot is visiting your website, looking at your website and actually tells me what kind of questions would a user have if you visit the website.
153
00:15:31,030 --> 00:15:36,390
Right. Not just only search terms, but real questions that people could have.
154
00:15:36,390 --> 00:15:39,790
And I think this is something.
155
00:15:40,210 --> 00:15:47,730
And Chris, I can offer that to you that I send you a report on the search of your website or anyone.
156
00:15:48,070 --> 00:15:49,690
I'll post it in the link below afterwards.
157
00:15:49,890 --> 00:15:51,090
Yeah, we will do that.
158
00:15:51,290 --> 00:15:55,490
So it's at the moment, I think it's German only, but it's an experiment.
159
00:15:55,490 --> 00:16:11,590
And with that, I mean, what we're doing with that is that we're actually thinking of a transition, you know, how can keyword based search transition into something that is understanding, you know, like people and people putting in questions.
160
00:16:11,590 --> 00:16:16,070
What is the cheapest refrigerator on the website?
161
00:16:16,310 --> 00:16:20,790
I think that is something that I think in e-commerce on e-commerce platforms.
162
00:16:20,910 --> 00:16:24,570
This was a task for many, many, many years already.
163
00:16:24,810 --> 00:16:28,250
I think there's also very cool solutions for that.
164
00:16:28,630 --> 00:16:37,310
But for enterprise websites, there's the content is actually the denominator.
165
00:16:37,730 --> 00:16:40,350
Right. If you don't have sharp written content.
166
00:16:40,350 --> 00:16:54,490
Your AI search integration will not actually find anything because I mean, obviously, all those, let's say, I'm pretty sure you've heard about drag retrieval augmented generation.
167
00:16:54,490 --> 00:17:04,330
That is like a way how you your search gets interpreted by a large language model and the search results get interpreted by large language model.
168
00:17:04,470 --> 00:17:08,230
And with that, you get a presentation of, you know, the results.
169
00:17:09,030 --> 00:17:26,190
Yeah. Once again, if your haystack is, you know, not cool, you will be ending up finding a lot of stuff that is not relevant because, you know, the LLMs will do a lot of stuff to actually help you to find actually something.
170
00:17:26,190 --> 00:17:31,010
And with that, I don't know if you don't contain it and you don't ground it good, good enough.
171
00:17:31,210 --> 00:17:35,830
You end up having like a search that might even hallucinate on stuff.
172
00:17:35,990 --> 00:17:45,150
And, you know, that Canada, you know, we've heard of the Canada chatbot giving you like false policies and stuff like that.
173
00:17:45,150 --> 00:17:59,130
So I think this is also a dimension if you enter with AI search, you will really actually need to find a way to look into the groundings and look into the facts.
174
00:17:59,130 --> 00:18:09,510
And once again, going back, if the content is not holding that information, you will be having a hard time to do so.
175
00:18:09,510 --> 00:18:27,750
I think if we've met several times through Janus Boyer's CMS experts network and one thing that we've seen there for many, many years is like, you know, the machine experience, you know, how bots do, how do they interact with your website?
176
00:18:27,750 --> 00:18:37,690
And we have this special, very special friend, Mr. Cranston, who has written a lot of stuff on that.
177
00:18:37,850 --> 00:18:54,130
And I still think that our industry needs to adapt also standards on, you know, content annotation and, you know, not just only JSON-LD, but I think everything that's also around the content, not just what you're publishing or, you know, the circumstances.
178
00:18:54,130 --> 00:19:01,010
And I think we've heard also very much future foresight in Montreal together.
179
00:19:01,290 --> 00:19:14,230
I think you remember that talk that we've heard from the Forrester analyst that he was like actually saying that he's seeing that content management actually at the moment is not going to go away.
180
00:19:14,230 --> 00:19:35,190
It's going to thrive, but it has to adapt actual more layers of information, more layers of data, but not just only the content and the content model and whatever, but it's also what's around and, you know, maybe also how organizations are actually working.
181
00:19:35,190 --> 00:19:46,910
You need to put this into your content. So, so where were, what do I want to, what do I want to tell you?
182
00:19:46,910 --> 00:20:03,710
I think what I've seen from talking to customers that want to transition, let's say from keyword based search to AI search based, and I see customers coming and say, oh, we want to actually, we want to have a chat bot, right?
183
00:20:04,590 --> 00:20:20,750
What we see is that chat bots are not everybody's darling, right? If the users come to a website, they might see the bubble, they might see the pop up, and they totally feel annoyed by it. That is my experience.
184
00:20:20,750 --> 00:20:37,110
Nobody's really because also chat bot quality has been so variant over time is that people are like, oh, I can't trust the chat bot. And if I, it will not actually help me on my use case.
185
00:20:38,030 --> 00:20:51,490
Yesterday, I had an, I was listening to a phone call that my wife did calling at the doctor's appointment and they were really like, super duper, you know, you don't speak to anybody anymore.
186
00:20:51,790 --> 00:21:02,550
It's a chat bot. It was a very intricate and intelligent chat bot. But the use case that my wife had, did not really fit into the logic.
187
00:21:02,550 --> 00:21:25,510
So I think that is something also that if let's say, imagine going back to search, you have this very specialized search that you want to have, like in for example, you want a sales agent, able to find like records of, you know, customer records, you want to find technology, data, all this kind of stuff.
188
00:21:25,510 --> 00:21:40,090
I think that the search there will need to be become more intelligent than just, you know, a chat bot, right? Not something that just reads something, but you need to have like, actually like specialized skills in that search.
189
00:21:40,090 --> 00:21:49,710
And you need to have like, like a router that is actually taking care of, okay, this is the intention.
190
00:21:49,990 --> 00:21:54,910
So this person is looking for a price, then I need to go and look for this.
191
00:21:55,230 --> 00:22:09,550
And then I need to have the search agent, for example, or this is another person who is like actually looking for contact information or looking for, you know, where can I call?
192
00:22:09,550 --> 00:22:11,830
What can I call you or stuff like this?
193
00:22:12,050 --> 00:22:18,790
So you will need to have like a separation of, of, of, of, of, of, of different search for that.
194
00:22:18,890 --> 00:22:20,350
I think I had prepared.
195
00:22:21,850 --> 00:22:22,730
Let me look.
196
00:22:24,030 --> 00:22:24,810
Do you know what?
197
00:22:24,890 --> 00:22:28,150
Let me, I think I can do that just once.
198
00:22:28,150 --> 00:22:28,450
Yeah.
199
00:22:29,310 --> 00:22:30,590
There's this one.
200
00:22:30,970 --> 00:22:31,330
Yeah.
201
00:22:31,470 --> 00:22:31,470
202
00:22:31,590 --> 00:22:31,690
Okay.
203
00:22:31,790 --> 00:22:32,050
We go.
204
00:22:32,750 --> 00:22:34,330
We just skip.
205
00:22:35,370 --> 00:22:36,670
We just keep this.
206
00:22:37,270 --> 00:22:39,710
I just want to show what I'm talking about.
207
00:22:39,950 --> 00:22:50,130
I'm just actually trying to explain, you know, an architecture that, you know, can be extended on the horizon horizontal.
208
00:22:50,130 --> 00:23:11,330
You know, it's not just only a search or program finder or expert finder, or you can send audio output to it, but, you know, actually, and this is where I see that where, where software or solutions that you want to offer to your customers where those solutions need to go.
209
00:23:11,330 --> 00:23:19,430
So, not, not looking into something that solve everything, but solves one thing good.
210
00:23:19,730 --> 00:23:19,830
Right.
211
00:23:19,930 --> 00:23:27,530
So this is actually what I see as a, as a, as a, as a, as a pattern that people should have a look at.
212
00:23:29,370 --> 00:23:35,570
So we've seen, we, we, we see that yesterday, two days ago, I was in a call with a customer.
213
00:23:36,430 --> 00:23:38,770
They have this, they have thing.
214
00:23:45,250 --> 00:23:45,730
Ments.
215
00:23:46,870 --> 00:23:50,070
Oh, are we still there?
216
00:23:50,210 --> 00:23:51,730
I think we were for a moment.
217
00:23:51,950 --> 00:23:54,450
There was like a, there was like a spinning wheel.
218
00:23:54,730 --> 00:24:01,510
Um, so this customer, they've identified eight areas of interest for the users.
219
00:24:01,510 --> 00:24:01,950
Right.
220
00:24:02,070 --> 00:24:12,230
And what they did was like, actually, they sent in 350 questions that they thought about in around those, those, those areas.
221
00:24:12,530 --> 00:24:17,950
For example, I want to get a, a, a, a license for fishing.
222
00:24:18,370 --> 00:24:18,670
Right.
223
00:24:18,990 --> 00:24:24,350
Um, so it's, it's, it's, it's, it's a very nice, uh, it's a nice place in Austria.
224
00:24:24,350 --> 00:24:27,350
Um, and, and they have all kinds of questions.
225
00:24:27,930 --> 00:24:32,990
People are going to the website and asking the website, okay, how can I get this?
226
00:24:33,150 --> 00:24:36,250
How can I apply for a tourism license?
227
00:24:36,250 --> 00:24:40,110
Because I want to sublet my apartment and so on and so forth.
228
00:24:40,210 --> 00:24:43,730
But also who can help me in case of this or that.
229
00:24:43,870 --> 00:24:44,270
Right.
230
00:24:44,350 --> 00:24:46,770
And also be, be able to disambiguate.
231
00:24:47,090 --> 00:24:47,590
All right.
232
00:24:47,770 --> 00:24:49,930
We're not responsible for that information.
233
00:24:49,930 --> 00:24:53,150
We can give you the information where you can look for.
234
00:24:53,310 --> 00:24:53,430
Right.
235
00:24:53,430 --> 00:25:03,910
For example, no, um, your privacy license is issued by another, um, institution, but you can find information about this.
236
00:25:04,010 --> 00:25:16,850
So now imagine transitioning this website into a website that actually helps you find those, um, find those information on a list or maybe as a.
237
00:25:17,370 --> 00:25:17,430
Yeah.
238
00:25:18,230 --> 00:25:22,330
Like, like, even like a written answer that you can copy down and so on and so forth.
239
00:25:22,330 --> 00:25:25,350
Um, with the sources and with the links.
240
00:25:25,530 --> 00:25:29,630
So I, I think that is what I can predict is what we're going to see.
241
00:25:29,830 --> 00:25:31,290
That's pretty often now soon.
242
00:25:31,490 --> 00:25:35,270
Um, that websites will start to become something like a mini.
243
00:25:35,550 --> 00:25:41,630
A clothes or mini GPTs where people will interact, will interact with that.
244
00:25:41,630 --> 00:25:59,630
Um, so, um, um, what I see also what is happening is that, um, yeah, well, they're overspending money into, you know, trying to do this one thing.
245
00:26:00,390 --> 00:26:03,710
They try to do it with different solutions.
246
00:26:03,710 --> 00:26:13,750
They try to incorporate it or, or, or, but, you know, we, we think people should look for a more integrated solution, uh, than just a solution that is giving the.
247
00:26:13,750 --> 00:26:18,950
So, um, so what they're doing is also that we're seeing is that they're.
248
00:26:19,010 --> 00:26:25,990
They're trying to solve problems, like, especially like problems with a, uh, retrieval augmented generation.
249
00:26:25,990 --> 00:26:31,170
You know, they're trying to, to, to solve problems just very superficially, right?
250
00:26:31,330 --> 00:26:33,410
Just, okay, let's chunk everything.
251
00:26:33,670 --> 00:26:40,650
Let's have super big context, uh, windows and just then hope and pray that the search gives you, gives you an answer.
252
00:26:40,830 --> 00:26:53,350
Uh, and, and, and, um, and, and, and before, um, they, um, you know, so before they do something, they should really actually repair their content.
253
00:26:53,350 --> 00:27:05,430
Right. This is what I really tell everyone, have a look at your website, have a look at the zero results, have a look at, or for example, am I really talking the language of my visitor or am I talking my own lingo?
254
00:27:05,670 --> 00:27:23,330
Right. And this is something that I think we have, we are observing and the more a website is trying to, um, how you call it, uh, trying to, um, you know, output the organization that they are, the more you'll find a gap between the
255
00:27:23,330 --> 00:27:35,710
users and the content of the website. So I think that is something that what, but we are in, we are lucky nowadays that we can go and we can actually scrape and get all the, all the content into a format.
256
00:27:35,710 --> 00:28:05,470
We can feed this to LLM with a special prompt and we can tell it, for example, all right, please rewrite it in a way that somebody that's, that is in grade five or something. So, and so, and so can understand this. And also please, um, please write it always in the format where the question is answered in the very, very beginning, not at the very, very end. Uh, like in a movie, you know, you watch a movie and then that, you know, just until the end, uh, and so on and so forth. Um, so,
257
00:28:06,030 --> 00:28:33,070
um, that is just what's on my mind at the moment, Chris. Um, um, um, so my, uh, takeaway here and on, if you're looking, trying to find, uh, uh, some, some advice from me, it's like, please see what people are asking. What are their questions? Um, and, and also in, in, as an agency, you should actually speak to the stakeholders, you know, not just only to the IT department and have the stakeholders.
258
00:28:33,070 --> 00:28:49,970
And you also need to explain to the stakeholders that AI search is not just adding a magical powder on top of it. It's actually something that only can only, uh, work if you take care of your content, right? So this is the second fixed content and metadata.
259
00:28:51,570 --> 00:29:15,870
Even try to think of, you know, your own model, you know, what is my industry model? Um, do I have things that I'm, that, that, that is not in a large language model. So might be, you know, that you are, are the specialist of that, that valve production, and you are very specialized and, and, and, and things like that are not in Wikipedia, for example. Right.
260
00:29:15,870 --> 00:29:44,870
Right. And, and, and, and, and, and, and, um, last thing, and is first concentrate on the retrieval layer and add features later as a, as a, as a, as a, as a view, right? For example, but, you know, step-by-step and always, uh, supervise, um, what is happening with your search, log, whatever you can log it in a good way.
261
00:29:45,230 --> 00:30:00,810
Um, even ask people for, uh, feedback. How is the search? I mean, you can go like, you know, do this net score, uh, test, uh, you can do, um, um, you can do, you know, thumbs up, thumbs down for the search. You can log that and so on and so forth.
262
00:30:00,910 --> 00:30:14,810
This is stuff that we are doing for, on, on type of three, for example. Uh, we even mimicking at the moment, you know, the, uh, Google search experience people, you know, if you go to Google, you have this AI search snippet that is coming up.
263
00:30:14,870 --> 00:30:30,570
I think that is something that will works, will work way, way better than just a chatbot, right? Because people are learning at the moment that Google is offering them, them that, and they will happily use that if there are on your website.
264
00:30:30,570 --> 00:30:43,090
And this will keep people on the website, even way better than just a chatbot. And, and, and, and, you know, um, we'll see that people will start to talk to websites, um, uh, sooner or later.
265
00:30:43,090 --> 00:30:53,150
Uh, people, you know, they, nowadays, they always also even talk in public with the phone, you know, like holding a knicker bread, you know?
266
00:30:53,590 --> 00:30:54,050
Yes.
267
00:30:54,170 --> 00:31:02,490
So, uh, we will see that even more and, and, uh, yeah. So that is what I, I, I can, I can tell about, about this in our 30 minutes.
268
00:31:02,490 --> 00:31:09,990
Um, I'm, I'm, I'm, I'm, I'm happy to connect to everyone here, um, on, on, on, on, on, on, on, on actual questions and so on.
269
00:31:10,770 --> 00:31:18,950
Oh, very cool. The, the challenge here that we have is that the time goes by so quickly, because as you're talking, I'm coming up with 18 more questions.
270
00:31:18,950 --> 00:31:24,890
I'm wondering if you do have just a minute more, and this might be a nice artifact for later.
271
00:31:25,190 --> 00:31:31,190
One of your slides that we passed through at the beginning was about describing RAG.
272
00:31:31,670 --> 00:31:36,590
Do you mind just taking a moment to describe that a little bit more?
273
00:31:37,210 --> 00:31:45,850
Yeah, actually, regular search is like, for example, regular search as we see it is keyword search, right?
274
00:31:45,850 --> 00:31:48,570
You have, let's say, imagine you have a website.
275
00:31:49,270 --> 00:31:52,610
The website has a title and it has a content block, right?
276
00:31:52,850 --> 00:32:06,570
So if you index this into index, like a regular index, Apache Solar index, you end up taking the title and put it into a title field.
277
00:32:06,570 --> 00:32:10,150
You end up having the content and put it into a content field.
278
00:32:10,350 --> 00:32:27,890
And then your search is actually going with a search query and trying to find, okay, do I have, do I have, I'll call it, hits on the title or on the content, right?
279
00:32:28,010 --> 00:32:33,770
This works pretty well if you index it right.
280
00:32:33,770 --> 00:32:34,570
So, okay.
281
00:32:36,010 --> 00:32:42,490
But, you know, the interaction is search, query, results list.
282
00:32:42,590 --> 00:32:45,890
That is like actually the normal search pattern.
283
00:32:47,470 --> 00:32:58,390
RAG is actually something where you have, you index also your documents into a, let's say, also into an index.
284
00:32:58,390 --> 00:33:04,710
But also you index, you know, you add additional embeddings to it.
285
00:33:04,830 --> 00:33:11,890
That is like AI tokens for understanding, you know, understanding what's in the text.
286
00:33:11,890 --> 00:33:18,470
And then the search query gets also translated into those tokens.
287
00:33:19,130 --> 00:33:31,190
And with that, you can search for something and have the large language model actually, or have actually the large language model look for similar tokens.
288
00:33:31,190 --> 00:33:37,410
You know, for example, you know that a dog and a cat are animals.
289
00:33:38,050 --> 00:33:44,070
And you know that your text document is a document, right?
290
00:33:44,190 --> 00:33:47,050
And they are really, they have nothing to do with each other.
291
00:33:47,210 --> 00:33:49,830
But, you know, the text document with animals.
292
00:33:49,830 --> 00:33:55,830
But, you know, that a cat and a dog can be, are actually animals and that they live in your house.
293
00:33:55,990 --> 00:34:01,930
And if you actually would ask a search, okay, what kind of animals can I have in my house?
294
00:34:02,270 --> 00:34:16,350
Then the search would answer to you, dogs and cats are what you can, what you often humans use as animals or have animals in their, as companions.
295
00:34:16,350 --> 00:34:39,170
So, so actually the rack does, does this whole mechanics, you know, they, they, you know, it indexes, it translated into embeds, it takes big documents and cuts them down into smaller chunks because also the large language model, they don't have indefinite, indefinite room for that.
296
00:34:39,170 --> 00:34:42,990
So you cannot just ask all your documents at the same time.
297
00:34:43,130 --> 00:34:46,790
So you need to have a, come up with some, some kind of a strategy on that.
298
00:34:46,970 --> 00:34:50,830
So this is often like cutting the documents into something that is smaller.
299
00:34:51,890 --> 00:34:59,150
With that, you have often problems that the context get lost, you know, because the chunks are not interconnected.
300
00:34:59,150 --> 00:35:07,950
Well, some, some rack systems have that, that they actually start to write all kinds of stuff around it and so on and so forth.
301
00:35:09,510 --> 00:35:14,510
So, and you have two moments where the large language model is actually taking care of understanding.
302
00:35:14,510 --> 00:35:22,810
So the moment where you send us the send a query to, to the doc, to the document storage or to the retrieval.
303
00:35:23,050 --> 00:35:26,410
So this is where the large language model is doing something.
304
00:35:26,970 --> 00:35:39,110
And when the results list comes back, the large language model is actually taking care of augmenting what has been generated by the search, the search engine here.
305
00:35:39,110 --> 00:35:55,410
And, and, and with that, you get like actually, you know, proper answers, for example, you know, you could tell the agent or you can, you could actually tell the large language model to the, okay, just answer in super short questions or write, write a novel out of it.
306
00:35:55,530 --> 00:35:55,610
Right.
307
00:35:56,010 --> 00:35:59,630
So that, that is what, what people can do with rack nowadays.
308
00:36:11,790 --> 00:36:16,050
So, that, that, that, that, that, that, that, that was a good explanation.
309
00:36:16,450 --> 00:36:21,070
I'm, I, I, by no ways and IT experts in that much, but.
310
00:36:22,070 --> 00:36:23,770
Hopefully that helps me.
311
00:36:23,770 --> 00:36:32,090
People ask me and now I can refer them to, you know, go to our video at 31 minutes in and you'll hear an explanation.
312
00:36:32,890 --> 00:36:33,850
That's really great.
313
00:36:33,990 --> 00:36:35,050
I really appreciate it.
314
00:36:36,010 --> 00:36:38,130
And time has gone by quickly.
315
00:36:38,530 --> 00:36:39,210
Oh yeah.
316
00:36:39,530 --> 00:36:40,130
Waiting for dinner.
317
00:36:41,330 --> 00:36:42,690
Thank you for joining us.
318
00:36:42,770 --> 00:36:43,950
Olivia, just super cool.
319
00:36:44,250 --> 00:36:46,330
Just hanging out with you and chatting.
320
00:36:46,570 --> 00:36:50,910
And I hope folks who are listening now or in the future, get a lot of value out of this.
321
00:36:50,910 --> 00:36:51,150
Yeah.
322
00:36:51,870 --> 00:36:53,590
Olivia, how can people reach you?
323
00:36:53,790 --> 00:36:55,990
We'll, we'll post your deets down below in the comments.
324
00:36:56,050 --> 00:36:56,230
Yeah.
325
00:36:56,470 --> 00:37:02,530
I think if you're on LinkedIn, you can, you will find me actually on that webinar link here.
326
00:37:02,950 --> 00:37:05,850
You can search for Olivier Doberkau.
327
00:37:06,570 --> 00:37:07,930
My name is pretty rare.
328
00:37:08,510 --> 00:37:12,530
You will find a lot of researchers, but you will find me.
329
00:37:14,310 --> 00:37:14,790
Yeah.
330
00:37:16,910 --> 00:37:18,130
How can they reach me?
331
00:37:18,210 --> 00:37:19,410
I mean, I have an email.
332
00:37:19,410 --> 00:37:21,270
I'm still an email guy.
333
00:37:22,730 --> 00:37:29,870
You can, you can send me an email to Olivier.Dobrikau at DKD.DE.
334
00:37:31,550 --> 00:37:32,450
That's awesome.
335
00:37:33,190 --> 00:37:34,150
Do something with it.
336
00:37:35,150 --> 00:37:36,290
That's really great.
337
00:37:36,810 --> 00:37:40,270
Thanks to everyone who took some time today to listen in to us.
338
00:37:40,950 --> 00:37:44,450
A really great 50th episode anniversary.
339
00:37:44,450 --> 00:37:47,450
And it's a real pleasure of mine to host you.
340
00:37:48,130 --> 00:37:50,950
Hope to see you again soon in person somewhere in the world.
341
00:37:51,130 --> 00:37:56,410
And we'll be following you and listening to what you have to say and what you're thinking about.
342
00:37:57,130 --> 00:38:02,970
Just a little shout out to folks who are maybe curious about open source.
343
00:38:03,810 --> 00:38:05,550
It really hit me on the head.
344
00:38:05,550 --> 00:38:10,770
We've been a little bit more closed source over the last 10 years.
345
00:38:11,010 --> 00:38:13,730
But just a little story before we go.
346
00:38:13,930 --> 00:38:20,430
I was talking about, I was talking to a friend about why he's using like Claude Fable for everything he's doing.
347
00:38:20,430 --> 00:38:24,650
And he said, well, I don't know, like what's another hundred bucks?
348
00:38:25,510 --> 00:38:31,370
And we've been doing a lot of work with the open source harnesses, both OpenClaw and the Hermes agent.
349
00:38:32,770 --> 00:38:36,690
And I was chatting with my other pals in Morocco and Cambodia.
350
00:38:37,230 --> 00:38:42,610
And they're like, we don't even have like 20 bucks just kicking around to go.
351
00:38:42,850 --> 00:38:44,970
So there's this divide happening as well.
352
00:38:45,210 --> 00:38:50,590
So anyone who's out there who's kind of curious about open source, those communities should be acknowledged.
353
00:38:50,850 --> 00:38:53,290
And, you know, maybe check them out and get involved as well, too.
354
00:38:53,410 --> 00:38:55,390
I just wanted to do a shout out there.
355
00:38:55,570 --> 00:38:55,930
For sure.
356
00:38:56,890 --> 00:39:04,670
I think that could be another episode about how open source is misinterpreted in actually also in the AI world.
357
00:39:04,870 --> 00:39:07,450
That there's a lot of discussions around it.
358
00:39:07,530 --> 00:39:11,490
I mean, we in the Western world, you know, it's no problem.
359
00:39:11,710 --> 00:39:13,790
This has become something that we can actually use.
360
00:39:13,930 --> 00:39:22,970
But many, many years ago when I was in Cambodia, that's the moment when we said we need to, as Type 3 Association now,
361
00:39:22,970 --> 00:39:28,810
we need to offer a community membership to people that is reasonable.
362
00:39:29,210 --> 00:39:37,050
And, you know, it was actually the hate of a normal membership was 125 euros.
363
00:39:37,050 --> 00:39:45,650
And this is no way something that someone who's working on Type 3 in Cambodia or in Tunisia can, you know.
364
00:39:45,890 --> 00:39:51,730
And even though, I mean, now we have a membership that is nine euros a year.
365
00:39:51,730 --> 00:39:57,050
And it's still hard for those people to become members.
366
00:39:57,330 --> 00:40:02,350
And I have my girls in South Africa that I'm supporting with Type 3.
367
00:40:02,350 --> 00:40:04,910
And I'm supporting them being the mentor for them.
368
00:40:05,410 --> 00:40:13,730
And even that, I mean, you start talking about, you know, stuff that is incubated.
369
00:40:14,390 --> 00:40:17,610
No, it's keeping them from doing stuff.
370
00:40:17,610 --> 00:40:25,090
And they have no, yeah, like, you know, going to a place, they don't have the money to take the bus, right?
371
00:40:25,290 --> 00:40:28,170
They can go somewhere, access a computer to use it.
372
00:40:28,550 --> 00:40:30,710
Nevertheless, they don't have a personal computer.
373
00:40:30,970 --> 00:40:35,090
And this opens up a lot of things.
374
00:40:35,230 --> 00:40:37,170
But that is maybe a different episode.
375
00:40:37,370 --> 00:40:37,610
Okay.
376
00:40:37,750 --> 00:40:38,490
I'm looking forward to that.
377
00:40:38,490 --> 00:40:40,130
We're pretty privileged, Chris.
378
00:40:40,450 --> 00:40:41,010
It's true.
379
00:40:41,210 --> 00:40:42,850
It was really great talking to you.
380
00:40:43,030 --> 00:40:46,810
And I'm really looking forward into running into you again.
381
00:40:47,030 --> 00:40:51,350
And greetings to the folks at .fusion.
382
00:40:52,410 --> 00:40:54,950
I know that pretty nice people there.
383
00:40:55,230 --> 00:40:57,250
And you're a pretty good guy.
384
00:40:57,590 --> 00:41:01,570
So I'm sending you this or that.
385
00:41:02,410 --> 00:41:03,750
Can't go wrong with good vibes.
386
00:41:04,010 --> 00:41:06,830
So on that note, I'm going to click some buttons.
387
00:41:07,110 --> 00:41:09,150
Maybe Olivia just hang with me for a moment.
388
00:41:09,690 --> 00:41:10,910
But thanks to everyone.
389
00:41:10,910 --> 00:41:14,950
And we'll see you in the next few weeks for our next episode.
390
00:41:15,150 --> 00:41:16,510
So have a great day out there, everybody.
391
00:41:17,090 --> 00:41:17,570
Bye.
392
00:41:21,350 --> 00:41:21,390
Bye.
