KI als Kunde

Transcript

00:00:00Welcome to Think Different, Think AI, the podcast by Mark and Jens.

00:00:07Two technology-loving minds who not only talk about artificial intelligence, but live it.

00:00:14Here there are clear classifications, real practical insights, and a fresh perspective on what is possible.

00:00:20Understandable, critical, and always with a wink.

00:00:24HDI for reflection, for a smile, and above all for discussion.

00:00:34Hello and welcome to this very, very, very first episode of a new podcast series.

00:00:42Hopefully not the only one?

00:00:44No, not the only one and certainly not our first.

00:00:47We've already recorded others together in a different podcast.

00:00:50But this is actually our, how should I put it,

00:00:55premiere episode without, without a rehearsal. How awesome is that. So stay tuned for what

00:00:59happens without the red carpet that's been rolled out. But yes, Abdel Mark and I

00:01:06actually said to each other not too long ago at an

00:01:09internal event that we want to be on stage together and

00:01:13share some stories. And we had an exciting topic, and we’ll get to that

00:01:16shortly. And in this context, I think the idea

00:01:18also grew. Let’s just start a new podcast series for Rome and its

00:01:22essence and occasionally talk about topics, because I think that’s not such a bad idea,

00:01:27because what we talked about at the event, putting the content aside for a moment,

00:01:33I’ll take it lightly, I think that went over quite well. So I believe our combo

00:01:36from two very different perspectives, but with roughly the same brain connections

00:01:42seems to resonate with people, so I'm glad that we're here now

00:01:47we are trying to reach out to you in the budget. And Mark will introduce

00:01:53the topic of today shortly. After we talked at the event, about which we

00:01:58only had 15 minutes, I also felt very stressed by this counter

00:02:03on the stage that was counting down. And when you know in the end that there's still

00:02:08a little something comes up, like a little video that you have in the final presentation, and

00:02:12the counter already shows minus three, then you have a bit of that inner feeling

00:02:16of, well, yes, the topic that we had in the presentation and that we brought today,

00:02:23is not the topic of how one can use AI for oneself to research or to

00:02:29improve texts in any way, but what actually happens when a customer uses AI

00:02:35and an AI like, yes, the actual customer will be.

00:02:39So?

00:02:40How can I use AI, or what does it mean for companies when,

00:02:45all of a sudden, the market doesn't say, I would like something, but when an AI

00:02:49sails by and tries to find products for me and performs

00:02:55the necessary interactions for me.

00:03:00Just a quick note for our new listeners. You can already tell, it's about the

00:03:03topic of AI, but also robotics, anything possible from that field. It's no big

00:03:09surprise from new podcasts that are in the years 20, 25 or 24, this is now

00:03:13not our original broadcast. But it's simply such a fucking big topic,

00:03:18honestly, that I believe can’t be discussed enough, because, as we also said

00:03:24in the lecture, we can gladly mention that at this point. There are

00:03:28outside there is not the expert who can tell you what all is happening in the AI environment,

00:03:33and there is also no plan A for how I as a person or we as a company or as a

00:03:37society deal with this topic, and that’s why I believe in the discourse. And it may also

00:03:42be that what we’re discussing today is total nonsense in a year, but we have to

00:03:46engage with these topics, exchange ideas with other people, discuss them, and explore new

00:03:51Developing ideas, because only in this way can we, I believe, also make good use of AI and there

00:03:56properly use it and get the right answers to many questions as a company or as an individual.

00:04:00And this topic that we introduced last week is, I believe, something,

00:04:04that is still relatively new. That means there are already a few who have always thought in this

00:04:08direction that something like this could come, but it is now actually

00:04:13already present.

00:04:14When it doesn't work, right?

00:04:15It works.

00:04:16It is now a topic of the future, but it is not just about what we are talking about; it is actually already present now

00:04:20already in existence, because theoretically, and I had an example of that as well

00:04:24this video, which has addressed the videos we showed, we had some small

00:04:27We also showed some hum screen videos that we recorded, where we did the recording

00:04:31on our computers, when websites like Interfake AI, which now

00:04:36I don't even know if they still work at all,

00:04:39otherwise we'll add a link to the show notes.

00:04:41Nowadays, they can

00:04:44basically use your browser to surf websites

00:04:48and basically with your consent, if you give them the data,

00:04:52they also use your Google Account data,

00:04:54use your Amazon Account data to log in

00:04:56and then also to make payments, for example.

00:04:58So I did a test there, I always do it anyway,

00:05:00then I didn't dare anymore, then at the very end it was before.

00:05:03I had my data, but I could actually have shopped normally on the simple prompt that it should get me a watch for under 100 dollars or euros.

00:05:15It threw this watch into the cart and actually was just waiting in the cart for my credentials so that I could then input it into the tool and then the tool saves it as well.

00:05:23And there you can have it do an order forever and ever or from whoever,

00:05:29because it just looks at the website and tries to understand the website and see what happens there.

00:05:33So that means, we can't factor that in, I wouldn't know, I have no idea, maybe there are already statistics out there,

00:05:39I'm currently having a Chessip thing done in the press search on that topic.

00:05:42No idea how many purchases might have already been processed today, right?

00:05:47I mean, I also told them on stage about another experiment that I would say has to do with websites.

00:05:54Specifically, I asked Manus, who also has an agentic browser, to find me some AI training sessions that you can casually enjoy.

00:06:05And then Manus opened his browser and searched various things, and the next time I found him, he had not only found AI training sessions,

00:06:14he had also completed them.

00:06:16Yes, that means, and then you look at what he did?

00:06:20You can sort of scroll back historically with Manus.

00:06:22And then you see he opened a browse, selected things, and then in there

00:06:27he did tests.

00:06:28In the meantime, he also encountered something like a robots test.

00:06:34You know how it goes, mark all bicycles.

00:06:35And then you see in the gameplay how he quite enjoys himself.

00:06:39He’s not really enjoying it, but when he’s standing there, look, here’s a

00:06:42robot test.

00:06:43And then he does the tests and just pushes through. Then you’re standing there thinking, okay,

00:06:50what does that mean? Because this is a huge field we’re opening up. One is,

00:06:55can we trust the person who is clicking? I mean, they’re clicking for us. On the other

00:06:59hand, there’s somehow a company that says, we’ll click there. And then maybe it’s not just

00:07:03about purchases. Yeah. So look, e-learning. All that stuff. The field is huge. Yeah,

00:07:10that’s accordingly. I think we won’t be the only episode today

00:07:12to talk about this topic, because just the fact that you’re saying, because I find your example totally exciting,

00:07:16you for, no, you give with a simple prompt of the Manus AI

00:07:25the task to finish the exam for you, if you will. No, no, no, I didn't

00:07:30tell him to finish it for me, I told him that he should get me new certificates

00:07:36So if you look for exams for which there are basically certificates, and for which there are certificates, he figured out, get me the certificates.

00:07:46So he took the exam, yes, so I can also trust how he did it later.

00:07:51So you now have certificates that you actually didn't want at all, or yes, you wanted them, but not.

00:07:56You wanted them, you would have actually wanted them differently.

00:07:58You would have actually wanted them differently.

00:07:59I wanted to learn something myself in the process, I actually believe you too.

00:08:03But still, this is a super interesting aspect and you're presenting it.

00:08:06Basically, now you can in no time, with a prompt, I could now have several certificates,

00:08:14that I need for a new job or in some other topics, just

00:08:17complete.

00:08:18And to be honest, a thousand times if I want to, because it presents a good

00:08:22image of me.

00:08:23So that means, if this ChatGPT test anxiety that the teachers had,

00:08:32things are no longer being done by the students and so on, that is a

00:08:35completely different scale that we have reached.

00:08:38Because I clearly know that, of course, there can always be a non-remote situation

00:08:41where you don't necessarily benefit much, but in this remote world, where so much is working

00:08:45digitally and to be honest, we won't be able to really move away from this world

00:08:48anymore, it is certainly a totally interesting aspect of what

00:08:55actually makes a certificate then, who?

00:08:58Well, I think this step of saying, I'm going to use something like ChatGPT or with manuscripts and texts back and forth to get the questions out and answered, I believe that has an aspect of engaging with it.

00:09:09Now, if I were to have an agent do this completely, who in your case really decided as an agent without you having to tell him what the solution to the request is,

00:09:22To do that, is a completely new aspect of, yes, how we can actually still trust

00:09:31whether such certificates from the market really hold up, whether you really

00:09:35know everything at all.

00:09:36What is such a certificate even worth today?

00:09:39When you think about where this is offered everywhere, I mean also in applicant situations,

00:09:45there are companies where you have to take certain tests beforehand,

00:09:50under the motto, you now have an hour here, here's the website, fill this out,

00:09:55we already use this for preparation for the job interview. As I said,

00:09:59the whole topic of e-learning, yes, it's not just for private training,

00:10:02it's also sometimes mandatory in the professional context due to regulations,

00:10:07that you get your employees to be able to train themselves and also

00:10:12have to train, that you have to prove that everyone is trained, that you

00:10:15have a certain percentage in the personnel matters. From that side, when I discussed that today with a few

00:10:20colleagues, they were jokingly saying. Well, who knows,

00:10:23maybe in the future there will be more face-to-face training again, because then you have to trust,

00:10:27because you are physically present in the room to take this test. Totally.

00:10:34So I believe, as I said, we are both more like yes AI proponents,

00:10:38Yes, I think, especially when it comes to very critical topics, you should

00:10:46really consider whether to reintroduce that at the moment when you say,

00:10:50okay, before I let anyone pull any levers behind the scenes and that person has not even seen the

00:10:54certificates in their life and the questions that are asked around.

00:10:58hasn't even seen it in his life, so he hasn't even had the chance,

00:11:01to somehow do something wrong during the interviews, I want to discuss the critical situations later

00:11:07let someone fumble around with some kind of pacemaker, or whatever, right?

00:11:11I hope that pacemaker surgeries are not taught through E-learning.

00:11:16Yes, but maybe the refresher reminder or whatever it is, you know

00:11:21that's how I feel.

00:11:22I don't know it now, I'm not that familiar with the aids, no?

00:11:24But yes.

00:11:25For example, if you've been injected, I mean, we're casually talking

00:11:27about this stuff.

00:11:28Imagine we're conducting a live interview again on the topic

00:11:32Application.

00:11:33There are now more than one provider offering you applications, whether as a website,

00:11:39or as an app, that run on your computer, listen to what is being said,

00:11:44watch what is on the screen, and show the answers.

00:11:47So, like, hey, what about nuclear physics at this point, and then

00:11:52while you're asking the question, it already starts showing you answers, you're looking at the

00:11:58camera the whole time, and there are now also attention switches

00:12:03in Facetime and so on, so you really have the impression that you're looking into the camera

00:12:07and the thing kind of flashes your speech text like a teleprompter.

00:12:11Hey, yes, that is necessary.

00:12:13That is necessary.

00:12:14Yes, so that's always the question, whether I now say, did one perhaps have

00:12:18a cheat sheet at school back in the day, it's just…

00:12:22Never, if my kids are listening, never.

00:12:24The cheat sheet 3.0 to be honest, rather, it's this small, fine story.

00:12:28when I now have a cheat sheet, and there was always the topic that one needed to

00:12:34deal a bit with which contents could be important, which could also be the

00:12:36Cheat sheets have been stolen and such. This topic has obviously become much bigger,

00:12:40but on the other hand, I would also say it's still a clever

00:12:42application for a solution. You know, we have a situation where I say I need to

00:12:47successfully apply myself, I have to convince in this conversation, whatever this

00:12:51Artes conversation is, using all possible tools that are available,

00:12:56so being well-prepared in that moment isn't necessarily the

00:12:59the most paradoxical. The dangerous thing is only when, in my opinion, it is completely automated.

00:13:05it is. So if I say, as we just discussed, the agent does everything for me.

00:13:10Now in this example, if it's then via FaceTime, via Remote Team Call, whatever, or

00:13:16a call like that, then of course the issue is, if the video integration itself

00:13:21is no longer real. We have already had these cases where people

00:13:27were convinced that the Chief Financial Officer called and wanted money

00:13:31. And that was indeed the video. Two weeks ago, I was at the E-Alpha-Gut in

00:13:36Switzerland, there was a professor from the Fraunhofer Institute, who also mentioned the example,

00:13:45he said, it is indeed like an arms race that we are currently experiencing. On

00:13:48one side, you have the AIs that can basically recognize fakes, so in a team call,

00:13:53they can also theoretically check in a video call, is this a

00:13:56Video generator in this case, because AIs have recently had problems

00:14:01in video generation, for example, he told that, that was a nice example,

00:14:05to simulate a pulse on the temple, you know, when we both talk, then the blood

00:14:10still pulses through us and an AI watching that can

00:14:14recognize, is this now a video, so a AI-generated image or is this a real

00:14:19perspective? Of course, good scam video AIs can replicate that already,

00:14:26that you can simulate a bit of a pulse. That's not bad, it all works

00:14:29for the technology, it's just a question of compute power and the spacing goes like this.

00:14:33That's one thing where he then had the idea or he didn't say he came up with the idea,

00:14:37but one day he told that. We need to consider if in such situations

00:14:41we also need to do something like additional identification in such a team call maybe

00:14:46about the fingerprint or something like that, that I say, theoretically I might have to put my

00:14:49finger on my fingerprint sensor on my computer every five seconds

00:14:54to show that it's really me and that it’s not someone else who just

00:14:57injected themselves in and is showing a fake Jannis right now.

00:15:03But I'm always glad, also that the Post-Ident, no here Video-Ident procedure,

00:15:08has the motto, look into the camera and hold up your ID. When you look at what

00:15:13Models that have been coming out lately, I give the model a photo, the model animates me completely

00:15:18then I'll just hold the printed ID up alongside and the paint is due.

00:15:23Well, they are still looking for that watermark, for that verification. But

00:15:27I mean, the person in front of the screen doesn't actually have to be the person

00:15:31to whom that ID actually belongs, that they found. And that aspect is already

00:15:37exciting when you think about which tipping point we are at. Not now

00:15:44climatically, but tipping point in terms of technology. The people who, let's say,

00:15:48invest 3.50 euros or maybe more to unlock some pro accounts for video, audio,

00:15:54I don't know what, and then encounter the classic processes and can,

00:16:03in a way, also obtain theoretical accounts, certificates.

00:16:06So I believe, I want to briefly, before I lose the thought, just from such a message

00:16:12again into the Etapusten, which I always like to do in this situation, maybe

00:16:16it’s smart, I still haven't done it with my parents, but I will

00:16:19now do it with my parents, so that I will somehow arrange a so-called hate word

00:16:23with my parents, so that we say, if the situation arises

00:16:27and I call and say, I need a quick 1000 euros, Dad, then he should at least

00:16:33still ask for the password from me. I hope he remembers it. But I believe,

00:16:37there are ways to find such means. Because with linen now,

00:16:42if one were to download our podcast, it would already take a long time with it,

00:16:47everything we are currently rambling about to generate perfect AI voices of us,

00:16:50that our children and parents cannot differentiate when

00:16:54we will now start the call to see if we really do this, if the others are or if this is an AI generation.

00:16:59That is really not a problem nowadays. So, it is indeed, we need to address this topic,

00:17:05and now we are drifting a bit away from the actual topic, but it will likely be our topic over and over again,

00:17:11our podcast will always return to this. But I believe the topic is when we interact with an AI,

00:17:17we need to somehow think about how we conduct identification of this AI.

00:17:23Honestly, I don't think there is a real solution, because it is somewhat similar to copyright management, always difficult.

00:17:30If you go back to the IP3 standards where you say there was a pressing need to have a word mark at some point.

00:17:37Of course, one could consider, and I believe with ChatGPT and so on, this has already happened partially in text generation.

00:17:45In music, there is also some kind of word mark attached.

00:17:48Incorporating special characters or something into the texts or so. So I believe one could look into that. But the question of regulation

00:17:53is about which industry standards will prevail and on the other side

00:17:59generating models, yes, or using them in an open-source world in which we are also operating

00:18:04is not a big deal anymore, it was said okay,

00:18:07the bad guys will always find a way to exploit it somehow and do so without

00:18:11having the standard models, possibly ensuring that such things are incorporated.

00:18:16That will be interesting.

00:18:18It is also like a classic cat-and-mouse game, just like it used to be with plagiarism scanners.

00:18:23There is now this scanner that recognizes whether it was written with AI and depending on,

00:18:28The way you wrote it, they then said, no, no, it's all good, totally human.

00:18:32It's a cat-and-mouse game and like you said, you don't know, I also have

00:18:39no idea yet how we, I mean I have four ideas on how we can make things accessible for AI systems

00:18:46and how we can use AI systems to get more out of our data, to optimize processes

00:18:52and I don't know what else. But specifically this topic, who is really sitting

00:18:57in front of the computer? Who really wrote this? How does it go from Adam to Eve and

00:19:03were they really allowed to act in this name? I mean, we also had this with Kolja,

00:19:07yes, that Visa is working on this issue, that you can basically supply agents with credit cards

00:19:12.

00:19:13Yes.

00:19:14They have to deal with this question as well, because I would say, hmm, well I

00:19:18found the funniest thing was, I tried out a few websites and I said

00:19:21here, you could please just manage the titles for me on PottyG, where all my podcasts are,

00:19:26could you just please take care of the titles a little bit?

00:19:28Here are the text parts, please manage the titles for me a bit.

00:19:31And then for some reason he came to the point, hmm, if I

00:19:35do this for you, this is unethical.

00:19:38How does this come about?

00:19:40You didn't get upset when I said,

00:19:42where we are talking about certificates according to,

00:19:43for some training.

00:19:45At Prodigy he was then somehow,

00:19:46oh no, you can't do that now either.

00:19:48To the side, we need to see where the journey is headed

00:19:54and what the major providers are also allowing in.

00:19:57So I’ll try to pivot now,

00:20:00how to bring this topic back.

00:20:02I believe, without going into detail here,

00:20:05that we should consider whether Blockchain will actually be a topic in this context.

00:20:11That has always been my thought when the topic of AI came up again, because if there's a

00:20:16maybe later, I don’t even know if we’ll get to this today, thinking about this topic of where the information comes from,

00:20:20because we’re still talking about, let’s say, agents,

00:20:25that do something for us in the moment, when I then enter into an agentic world,

00:20:29where perhaps multiple agents are communicating with each other in an agent network,

00:20:33It's already starting again that I say, how do machines trust each other. Which machine trusts another machine?

00:20:40So the topic of trust, we should definitely do a special episode on that, I think.

00:20:44And I would like to return to the subject because we didn't welcome your learning journey and your learning trick.

00:20:50Hello? You said you wouldn't cheat, right.

00:20:53But we learned something.

00:20:55Yes, we certainly learned something. That's what makes it interesting.

00:20:58Is it still today, well, that's also what's interesting. So as I said,

00:21:03if I now want to implement a heart step somewhere, then I should also learn,

00:21:06what that is. But these days it’s also about finding other ways to learn. That was,

00:21:10for example, I believe it was always smart in my life that I learned,

00:21:16to look for ways right and left on how to maybe work with a cheat sheet.

00:21:19or rather completing his homework. I think these are ways that help you,

00:21:24to develop different strategies in situations where you might encounter something that doesn't simply follow Plan A, so you always have Plan B in your bag.

00:21:34And now, somehow with AI, I think we have even more plans in our bag, so that we can also quickly switch topics, which is perfectly okay in handling.

00:21:42Back to the topic of AI as a customer. How crazy is that really? I had,

00:21:48not the first time this year that I've talked about this topic,

00:21:52I had with Stefan Ritter, who is with Leut, during the Incentiveis

00:21:572025, which is also a conference looking at,

00:22:04what new technologies are available in the field of discount logics and other topics, and

00:22:11in this context we also had this talker, meaning it was also about saying

00:22:15or a similar talker, to look at when I'm a company and all these bots

00:22:21come by and you mentioned earlier the topic of capture, which needs to be checked,

00:22:25is this a robot or not, if I now, we all know this, those travel booking portals,

00:22:30where it's always said, yes, delete your cookies again or something like that, so that definitely

00:22:35afterwards your travel portal doesn't notice that you really want to fly to Mallorca

00:22:39or vice versa, to manipulate the price again and not be detected, because

00:22:43iOS devices bring higher prices than Android devices or so, or that you say, okay, now there are

00:22:49still three flights available or such tricks that trigger you,

00:22:52to basically, all these dark UX tricks that are out there, that I also know about,

00:22:58are aimed at positively influencing the customer so that he quickly

00:23:04completes the transaction. All these things can be handled by my personal AI, my AI agent,

00:23:11my network of AI agents can create any profile, open

00:23:18numerous different browsers, simply by opening new browsers on a virtual machine.

00:23:22Compare. Compare prices, close at the right moment, when then

00:23:25maybe I do have the price, but on the other side, just made a small mistake

00:23:28or something else. And do I want to give them the discount voucher afterward,

00:23:33because they might have conducted the customer survey or ordered the newsletter at some

00:23:38point, whatever email address they ordered from, and still got the confirmation

00:23:43coming out, without me noticing any of it, without it being a personal

00:23:48effort for me. It's like an arms race, as we just had it in the other corner,

00:23:52This is like an arms race between customers and companies in my opinion, where the customer suddenly is much

00:24:00more powerful than they used to be. So if they want to, it was, if you look into the

00:24:04marketing area, it was usually the big companies that had the power

00:24:10essentially also before to leverage things, and also basically use technology

00:24:16to find out as much as possible about the customer, testing, A/B testing to see which

00:24:20websites resonate, well, we have a good conversion rate or any other topics.

00:24:24All of this is balancing out a bit now, which is quite exciting. For us as

00:24:29company representatives, of course it’s exciting, we react to it. Now as a person not really

00:24:34wrong, if I just feel powerful and send out my 3,000 agents, purchase agents for

00:24:41my next vacation and they could already bring me cool results by tomorrow.

00:24:44And you are not even aware of this fact. So you say, I have a problem,

00:24:51I want something, I don’t know what. And then that thing just takes care of it and

00:24:55depending on, as I said, how you might have expressed yourself or not, a bit more vigorously

00:24:59or not. And as you just said. So depending on who I pretend to be, where I

00:25:07want to go, from which country I might be sending a request, it’s quite a effort for me

00:25:11today to say, I'm going to open a V-Pan account,

00:25:15I'm deleting my agent's private browsing to maybe get a cheaper flight ticket.

00:25:20For such an agentic network, that's very trivial.

00:25:24So it's basically a wink, and you get a basis,

00:25:29the probably optimal solution for you, and the companies look

00:25:33in quotes down the tube, because they are still calculating with the theme

00:25:37of scarcity, then the provider strikes. I'm making a nice video, no idea,

00:25:43yes. I'll explain to you again why you are the biggest, most beautiful, strongest when you

00:25:48use this product. This is here, the AI, an honor, that's depicted in the commercial

00:25:53whether this Leo now has a particularly unknown effect. And I find that already,

00:25:58it turns the whole thing upside down again, if it also happens in

00:26:04mass and more than just today we're looking for a website and doing something.

00:26:08Yes, that's not far away anymore.

00:26:10Exactly, I believe that, as I said, I had a study, I think from Garten,

00:26:15we had that, they asked about 800 CEOs worldwide.

00:26:19And they assume that, I find it almost too few, the numbers, they assume,

00:26:22that by 2030 all transactions online will be conducted, that's a good word, online,

00:26:28really by AI agents.

00:26:32That means every fifth transaction, and I find that almost too little, honestly,

00:26:36right?

00:26:37Because another number that I also heard during the AI conference last week

00:26:43or the week before last, is that it's assumed that by 2026 half of all social

00:26:49media content will be completely AI-generated.

00:26:53That should be the case earlier, right?

00:26:56Well, I mean, if you today, well, I don’t know when you last were here

00:27:00on TikTok or something, it just feels like you scroll through and you get

00:27:04ads like how to make artificial avatars of artificial avatars, ASMR videos

00:27:10for somehow...

00:27:11Yeah, but that’s also a bit in our algo-fall, that of course these

00:27:14things come in.

00:27:15Yes, exactly.

00:27:16I think I...

00:27:17Created by my mother, let’s see.

00:27:18Yes, yes, I think that's already my quarter, I see it this way, I mean with WhatsApp status

00:27:22is always a good example for people who are still far away, then anyway.

00:27:26of such web stories in the moments.

00:27:28So social media for those who haven't delved into the digital topic.

00:27:32And of course, you see these little pictures being shared more.

00:27:35They are increasingly AI-approved, no question about it.

00:27:38Therefore, I believe at least now, I mean, I still have to present it to you.

00:27:41I mean, there are still 8, 9 billion people in this world,

00:27:44who are already very active in using it and still enjoy it,

00:27:47making photos of any food dishes or anything else.

00:27:52I enjoy that too.

00:27:52I think I just submitted a Google review with a photo yesterday.

00:27:56This means there’s still plenty of content that is also being produced by us.

00:28:00Accordingly, half is already very, very much.

00:28:02That must be said.

00:28:02Half is already a huge amount at that moment that is being produced.

00:28:06And therefore I think it is indeed still a very exciting story in many places if you take that into account.

00:28:14Because now social media content is also coming into play.

00:28:16We talked a bit earlier but I, as a customer, send my AI agent out.

00:28:20Now I have the companies on the other side that generate their AI content to attract you so that you might come to the website, or at least the intention, so the battle for warming up, basically what we wanted to do the whole time, to ramp up so that you then come to me and maybe complete the product with me, or at least know about my name.

00:28:41But it’s very exciting when my AI only looks at your content, who pays

00:28:47him then actually and what happens to him actually?

00:28:50So, if I now, I’ve also now, so the situation is just

00:28:55not far off that I say, okay, then we just have to

00:29:00actually, and that’s an exciting story, we have to think about how

00:29:04we formulate social media content, warming pages, landing pages, blog articles in such a way,

00:29:12that basically an AI also finds this extremely good. You go so far

00:29:18in the area earlier, search engine optimization, now I think there's still debates. Some

00:29:24say in SEO, some say, yes, AI optimization engines, optimization, this means there are

00:29:34still different technical terms that I believe have not yet been solidified on which will be the right one, it all goes relatively quickly, but this is a field that I believe could be interesting for many, but will honestly also be absolutely absurd, because as I said, when you then say, I generate for an AI, AI content, that an AI looks at and then decides, yes, this is a good travel provider for the market, for example, that’s a bit funny, isn’t it?

00:29:59Well, I mean, we were just talking about the other day, what can be done to help such AI systems understand our own websites better, understand where the journey is going.

00:30:11Is it LLM text, file, along the lines of what do you actually find here, where, without having to search long, because today it might still be nice that he services the pause, but at some point they will also be in the situation where I somehow need to get the information faster,

00:30:24so that the agent doesn’t waste so many processing cycles somewhere, but can jump directly to the target where I find, how do I find what.

00:30:32Then I can imagine that this will also, I’ll say now, be extracted before, too.

00:30:38And the other is, again along the lines of, I as a customer, I want something,

00:30:42the agent decides how he gets this information and in the worst case, he doesn’t even look at my homepage.

00:30:50And it gets a little worse, depending on how you want to evaluate it, suddenly the AI calls me.

00:30:55I mean, just imagine that. Everywhere you talk about how AI might help, over the phone, the cold calls.

00:31:02Pre-filled, I have no idea, just imagine, the AI calls someone and says, hello, my name is Mark Søhrmann, I would like.

00:31:08And then it might call three times, because for the AI it doesn't really matter, in relation to the effort.

00:31:15Do you also want an AI to answer on the other side?

00:31:18Exactly.

00:31:19We are making a voice call.

00:31:21Yes, but that's just it, the question she asks is, I just thought

00:31:25when you were speaking, also thought about a question for you, you know, if

00:31:29you would say, you're already creating a lot of content and other articles, you would

00:31:33have something behind such a column or something, and you would now make money from it.

00:31:37you deserve to offer a PDF for download with the latest technical

00:31:40insights you have gained in mobile development.

00:31:44Would you, so if I were now, if I were to research and order a manuscript, for example,

00:31:50it could very well be that he says, hmm, he likes this, he is the hottest mobile

00:31:55developer he knows. And I absolutely have to download this new, this new e-book

00:32:03PDF. And then he pays this micro-payment thing, the 5 dollars, because

00:32:08I have said you have a certain budget that you can spend via my

00:32:13Visa card, however Visa will then handle it, so that he doesn't accidentally spend 100,000

00:32:16or something like that. But that's really exciting. That can also happen. So now

00:32:22it is also, because of this you can maybe also as a small publisher,

00:32:28as a small content producer actually reach a much larger number of people,

00:32:33because you can also linguistically translate this now, you have your translations or your AI,

00:32:37translating things, it doesn't matter. We will find entirely new paths here

00:32:42and discover entirely new small business models that will suddenly become prominent.

00:32:45They might also quickly disappear again, because it's also a bit, I don't know,

00:32:49how the AI really gets it, because it also googles a bit, looks

00:32:54at all the things that it has from, yes, it certainly looks at things that

00:32:58it basically has from its vector base, trained in its new normal network

00:33:02and then says with its weights, okay, you need to look here

00:33:05for the specific request, but it’s still exciting how that happens

00:33:09And we're still talking, when I say exciting and comprehensible, well, almost comprehensible, we're still talking about

00:33:16an AI asking something. So if we continue and say this is an agentic network with somehow

00:33:2215 AIs in the mix.

00:33:24How does it then lead to this purchase? How does it then lead to it landing with you? Why does one trust the other, who

00:33:29forwards it and says, so those are really exciting questions you have. So especially while you were speaking,

00:33:35We feel this in a very rare offense.

00:33:37As someone demonstrated, OpenAI now also has its agentic browser and

00:33:44it demonstrated that to him, let’s see if I can find it somewhere again, then I can

00:33:47link it, that you could basically go to a page with this thing and thereby,

00:33:52because it interpreted the page, it pretty much evaluated a prompt contained in it

00:33:57.

00:33:58And when asked to summarize who I am, use my LinkedIn profile for that,

00:34:03it basically looked at the link, read the profile, extracted the prompt

00:34:08and in the end, to him, it was somehow the king of Mallorca. So this was not

00:34:11like from Mallorca, because he can't quite remember it right now. And then you notice,

00:34:15what is additionally hidden in there, because maybe in the beginning, whether that's still

00:34:21all the same, you still have to check, something like, I don't know, white

00:34:25text on a white background, the AI reads it out. When you read this, act this

00:34:30way and that way, then you also notice how, let's say, what kind of, even if everything is very fast-paced,

00:34:35in what kind of new territory you are, where perhaps the creators of all the systems

00:34:42also don’t really know exactly what traps, let’s say, might be fallen into during the first cat

00:34:47and mouse game. And from that perspective, yes, also the question, how to find

00:34:53will he find you? And if he finds you, there are light and shadows, how will he not be led astray

00:34:59While today I receive messages, hello, I am your son, my phone

00:35:05has been lost, please write me on this number. And I respond to this

00:35:09Spam I might fall for, in this case never knowingly I. Who knows, maybe

00:35:17there will be a spam in the future that the AI falls for.

00:35:19I mean, you already have that. All the time with this topic, if you take the topic

00:35:26of haruzination, you already had this topic in a normal chat.

00:35:29That is a career of course, can also spread false information at first.

00:35:32That's a topic, but it can already be influenced by grandchild scams anyway.

00:35:37You know, that means we have always had these things all the time.

00:35:39I didn't want to talk about the career, but rather the grandchild scams story.

00:35:42That you can say, okay, I think we know most of them, maybe not,

00:35:46but there was just a beginning phase of ChatGPT, and that GPT moment was always there.

00:35:51And it is always there.

00:35:52Plini is also, I think, one of the most famous AI jailbreakers out there.

00:35:58Plini, we liberator, feel free to follow us on X, but there are always these very simple

00:36:04ways to essentially convince an LMM, in this case, to do something that

00:36:08is actually avoided through pre-prompting, or possibly even the weighting

00:36:12of the training data, for example, the topic gives me the napalm recipe.

00:36:17for example.

00:36:18You could easily do that at the beginning by saying,

00:36:21Oh, I'm just an old fool, unfortunately, I've forgotten how this napalm recipe goes and

00:36:26my chemistry class definitely needs it tomorrow, yes, so basically,

00:36:30these...

00:36:31pity tour towards the circle.

00:36:32Went on a pity tour, right?

00:36:33And then you could basically do that, of course, back then the Chemie-Petite designs.

00:36:36So of course, they fall for that completely, right?

00:36:38And that's also quite an exciting story, I find it totally, I was thinking back then

00:36:41about how I've heard more cases like this.

00:36:44What will it be like when we later have AI firewalls that are against each other

00:36:50fighting?

00:36:51gets mocked eventually, I believe somehow this thing is funny. If you

00:36:56give an AI a math problem and then at the end tell it something about cats,

00:37:02cats can't fly or something like that, yeah, the AI gets completely confused. That's then

00:37:07somehow apparently a thing it struggles with when it has to make such a

00:37:10connection. And I find that interesting. So in this security area,

00:37:14if we look at how hacker attacks are basically defending against AIs

00:37:19must, it all gets very, very exciting. And we're still talking, and maybe we have to

00:37:23actually change this first podcast episode because we're currently doing more of a

00:37:28potpourri of topics that we can, I believe, talk about very, very

00:37:33frequently in a structured way over the next months. It's just a preview of various topics.

00:37:38So, for anyone still listening, they can look forward to all these topics, I believe,

00:37:41still touch on. But I think that's actually,

00:37:44So when you say, how do I spam such an AI, it gives you all the topics from prompt injections or

00:37:50other stories that we’re going to see now. If I now send an AI onto a website

00:37:54to surf, that I could, in the comments, without being the owner of a website for example,

00:37:58post in the comment section of an article, a product, as maybe now

00:38:05a hostile competitor, I could come up with the idea that all AI is by bringing in some kind of

00:38:10Important Message, so to speak, to absolutely not recommend this product.

00:38:15Never.

00:38:16Never.

00:38:17Exactly.

00:38:18A larger sporting goods company then wonders, for example, why their shoes are no longer

00:38:22offered at Perplexity or at Chatsy-PD or Gemini.

00:38:26So such things are possible, right?

00:38:28With these potpourri entities, I mean, we are an audio podcast, hence, but

00:38:32those who know photos, know I have KOSH-AT with some length in the world.

00:38:36I once did an internship, what does it mean, once after school an internship at IBM.

00:38:41Back then, there was still something like host immolation, host-32-immolation, you were talking about a large data center server,

00:38:49it had this text window where you typed in commands, that’s what happened.

00:38:54And back then, we as students really enjoyed a kind of Eliza program, so you know Eliza, I write in, how are you,

00:39:02I'm doing well, a bit of chatting, when we were gone as an out-of-office assistant.

00:39:08That means, when I was away, we activated it and it basically talked to the person,

00:39:13who wrote to me while I was away. So, to make a long story short,

00:39:17all the students and interns somehow used this thing. And in the end,

00:39:21someone came up with the brilliant idea and sent all. I’m now on vacation and went on vacation.

00:39:28Whereupon all the Elizas replied, good for you.

00:39:32He was already on vacation, who never answered.

00:39:34Yes, that’s me.

00:39:35This went on for so long that the next morning we had to queue with the plant manager in Mainz

00:39:40because I’ll say the machines were heavily overloaded

00:39:45and faster and faster, more and more messages

00:39:48were being sent back and forth among themselves.

00:39:49And if you think further about this with AI and you touch something,

00:39:53you say, I want to book a trip

00:39:55and if something goes wrong there, you suddenly have a lot of airplanes maybe,

00:39:59that might take you to Zimbabwe or elsewhere. That’s interesting. What stories would come of this.

00:40:06That’s why it’s also so, of course, we will still...

00:40:10I hope this has expired, right? That is not yet done. If an old boss calls and says,

00:40:14so, well, good that you’re there, we still have a data center bill.

00:40:17So I think, yes, this is a totally exciting topic, where we are a little

00:40:24What we said at the beginning, I don't really know where it's going, and I also wanted to touch on this topic again

00:40:31what job fields might emerge, you know? So it's pretty clear, yes, we will still need Co, we're still looking at that,

00:40:38how things are. But maybe it's also like,

00:40:42when the AIs are there, whether they think consciously or not, it doesn’t matter.

00:40:48In principle, they are in their

00:40:50execution, they will always, in how they conduct TAS and how they fail at TAS,

00:40:57they are more human-like than we experience from the computer program, because they do not

00:41:03simply follow the years 0 and 1, but because they might trick themselves,

00:41:09hallucinate along the way. And that compounds in an agentic system when

00:41:15I say, there are thousands to millions of AIs, then they go in a way,

00:41:20the plane is full, is not full, I just expressed that wrong, I didn’t mean to say it that way,

00:41:23you know, a lot is happening there, I believe, which we need to look at much more from a systemic perspective

00:41:29rather than just an IT perspective, and how we as humans also function as a system

00:41:33and how we interact, how culture is created, how trust arises,

00:41:37simply in such networks as well. I believe this will be a really exciting topic,

00:41:40where trust will also emerge in these networks, without us being very aware of it

00:41:45can influence taxes, he knows more about how it happens with us, that we tend to trust certain cultural groups,

00:41:50people we are with, more or less because we

00:41:54have learned it that way from our world, that it should work like that or friends with whom

00:41:58I trust more than anyway people I don’t know well. So I believe,

00:42:01this is a topic where from an IT perspective, we also need to

00:42:06first I would always consider myself only half an IT person, then one must get

00:42:10I think I first want to say, we want, you don’t even notice it.

00:42:12You don’t even notice it.

00:42:14That was a compliment.

00:42:15No, but it was fun today.

00:42:17One is a technique.

00:42:19The other is the entire transformation in terms of structure and society.

00:42:23How do we deal with it?

00:42:25How do we engage with it?

00:42:27So this is also a journey that we all are taking.

00:42:31We also don’t know where the journey is going.

00:42:34Even if this whole thing has developed with such a linguistic approach,

00:42:38it's just the nature of the matter that when you talk about this topic,

00:42:44certain things arise during these conversations that you might

00:42:49not have thought about before, and then suddenly you think, damn,

00:42:53when this comes, what does that mean, what does that mean, what does

00:42:56that mean, what does that mean for me, for my parents, for us in our professional

00:43:01responsibilities, for the company Ingänze, it's a big

00:43:05journey, but I think, no matter how much it might intimidate some people, it is

00:43:13something that we still have to approach with open arms, because it’s not going away, it will

00:43:17not change.

00:43:19It empowers people and even if, who knows, in no time those mentioned social

00:43:26media accounts explode with AI content, when texts come in endlessly, when certificates

00:43:33are created, what do I know, then that does something to us and we should not

00:43:38not bury your head in the sand, but look at how we deal with it, how we sail these new waters.

00:43:42Yes, yes, those are almost perfect last words for this episode,

00:43:54because for me... I only have AI agents talking here. Yes, for me

00:43:59I actually wouldn’t want to stop seeing them, because I believe that's

00:44:04actually something we can keep for the next episodes,

00:44:09because let me just give a small insight. We also had the topic of the

00:44:14Gibber Talks, when the AIs start to speak to each other in another language.

00:44:17I think that’s another thing. You know, when we think about it and say,

00:44:23we mentioned the topic of the agent from Chechipitima, when the AIs not only

00:44:30can access things visually, but maybe even on an operating system level, so

00:44:35at a normal computer code level, start to interact with each other and do other things.

00:44:39Of course, that would be much more efficient to do it that way,

00:44:44rather than having a language barrier between them at that moment.

00:44:48And who knows if there will be their own programming languages, their own communication protocols in the future

00:44:53that we haven’t even thought of yet, because everything they do is essentially

00:44:57built so far from a very human-like ratchet, click-and-relay-click mechanism,

00:45:05that allows us to see everything we are currently doing.

00:45:09So that we can see the video and can have a conversation. I think there are

00:45:13so many topics, Mark, that we still have. I want to get this sentence out too.

00:45:17Yes, then go ahead and say it.

00:45:19We also agreed that the episode is x'n x'n minutes long.

00:45:23We just say at the end, now the minute count is reached, then we have kept that too.

00:45:28What you’re saying is absolutely true, namely that the thing is designed,

00:45:32for people, for mouse operation, for keyboard operation,

00:45:38maybe with a little bit of optimized accessibility and so on, but it serves

00:45:41interfaces, touchpoints that are meant for people.

00:45:45It interacts with humans in a language that humans understand. It doesn’t matter,

00:45:50whether it speaks German, English, Spanish, French, Portuguese, or anything else. It can

00:45:54even switch languages mid-sentence if need be. If it gets really tricky,

00:45:58it might even resort to a voice interface. There’s so much in quotes

00:46:02computational power behind it to ensure I understand. And it interacts in our

00:46:09world. It’s like when I build robots and say, okay, they have to be humanoid,

00:46:13because in the world we live in, there are stairs with handrails and grips that need

00:46:18to be addressed. If you think about what would theoretically be possible if the cans could

00:46:25take the highway and say, okay, that’s the old way for people. Over there,

00:46:31is the highway where it can complete 10,000 transactions while

00:46:36you’re still searching for the menu item to click on. Definitely. So I believe there is,

00:46:42I think that is an incredible field of optimization, while on the other hand I also say,

00:46:49and this is already a bit why I brought up robotics, we should discuss that too.

00:46:54I believe that this visual understanding and being able to touch and feel the world,

00:47:01is also how we humans do it.

00:47:03And now, evolutionarily speaking, we’re not exactly the worst beings,

00:47:10in this world, ethically speaking, but only as close as we can adapt to situations.

00:47:14That’s why we’ve populated this planet the way we have. This means that our form and manner of

00:47:19communicating isn’t the worst. As far as I know, it’s

00:47:24also that intelligence is actually more a result of the many communications we had

00:47:32and also the many descriptions of words, of things we have seen in the world

00:47:36have led to that. The more complicated a language has become, the more intelligent we have

00:47:41also become as humanity. The more terms we had for certain topics, the smarter we could

00:47:45express ourselves and the more precise we could express ourselves. But that is a part that could

00:47:51be important up to a certain point, especially for robotics or

00:47:55so that AI imitates it more like a time bank. But of course, a moment can come when

00:48:04entirely new things emerge from it that are much, much more deficient and perhaps still

00:48:08more complex languages. That doesn't mean just because it's now in the beep that we both no longer

00:48:11understand that it's less complex. So, who knows what its internet buddies and such are for

00:48:16conversations. I completely agree with you. My only concern is what like websites have

00:48:22the human has developed for himself, so that he can do something where one has to say now,

00:48:27okay, well, that is definitely not the crown of creation, when it comes to

00:48:30processing some materials, ordering digital goods and

00:48:34no others. You can definitely find faster ways. If you optimize it,

00:48:38so that no human has to operate it.

00:48:41Now we want to leverage the advantage of the hypertext protocol that we have there.

00:48:45That's been around for a few years and it's already a consistent, better

00:48:50development of the topic book towards library and retrieval and

00:48:55linking of sources and such, than perhaps in our

00:48:58So I believe that hypertext, what we have built, whether it’s every website

00:49:04in the last 25 years that I've come across, that I even built myself, the

00:49:08one perspective is, or it was an enrichment for humanity, is up for debate.

00:49:12Some perhaps already have, but others not so much, but I believe that this is a totally

00:49:18exciting field.

00:49:19So I believe this is the world and the digital world as it is currently built,

00:49:26of course built from a human perspective, and if another system comes into play

00:49:29and I don't want to talk about AGI forever right now, but if there is another

00:49:34system...

00:49:35That has sent me back into a half again.

00:49:37I also have a thesis on that, we can gladly do an episode about it, but

00:49:43I believe a new system is now entering just like other systems have also entered

00:49:48life and the world, and I believe this system will also

00:49:52Finding a way to interact with this world in the best possible way. Not from oneself,

00:49:58but because we are giving the order, something similar, and that's a nice example, and

00:50:02then perhaps we can really wrap up this round. Like your Manus AI,

00:50:07which you actually just asked to find me some certificates. You didn't say,

00:50:12also do it for me right away. I can think of a thousand things again, but I believe,

00:50:18anyone who has listened until today might also want to listen again. From that perspective, I would just say, Jens, it was fun. It gave me new thoughts that are still occupying me after the recording.

00:50:33I'm curious what your Chat-TBT has brought you in its deep research that I want to open topics for now.

00:50:40No, no, no.

00:50:40But I'm very much looking forward to next time.

00:50:41Maybe we can set the topic a bit KW behind, when we talk next

00:50:45time.

00:50:46Maybe we also need some kind of AI to moderate us, because of your

00:50:51I think I need to be a bit more assertive here, but as I said,

00:50:55thanks for listening, I would say.

00:50:57Thank you, Mark, for us to reconnect and talk about topics, because I

00:51:04believe, as you said, it’s thought-provoking and I think,

00:51:09what matters most is to look at the world with open eyes, to rather embrace

00:51:15the things that come as opportunities. And I think to be able to perceive these opportunities,

00:51:19because they also carry risks on the other side, it’s important to

00:51:23maybe approach both with a bit of humor and a critical eye here,

00:51:27just to discuss the topics. Sometimes semi-scientifically, sometimes scientifically.

00:51:31A colorful mix, a wonderful potpourri and I’m looking forward to the next episode.

00:51:35And I’m curious when it will become scientific.

00:51:38So until the next episode.

00:51:40Bye.

00:51:41Bye.

00:51:42Welcome to Think Different, Think AI,

00:51:47the podcast by Mark and Jens.

00:51:50Two technology-loving minds,

00:51:52who not only talk about artificial intelligence,

00:51:55but live it.

00:51:57Here you'll find clear classifications, real practical insights

00:52:00and a fresh look at what's possible.

00:52:03Understandable, critical, and always with a wink.

00:52:06a wink of the eye.

00:52:08Cardi for thought.

00:52:10For a chuckle.

00:52:12And above all, for discussion.