THE BROWSER STRIKES BACK

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 you will find clear classifications, real practical insights, and a fresh look at what is possible.

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

00:00:24Karin for reflection, for a smile, and above all for participation.

00:00:34Honestly, I understand absolutely nothing about technology.

00:00:37Now my wife has already said, we finally need to get on the internet.

00:00:41But I’m not even on the internet.

00:00:43Am I already in or what?

00:00:45I am in.

00:00:47That’s really simple.

00:00:48It’s that easy with the AOL internet starter kit.

00:00:51In.

00:00:52That little laugh at the end is also worth a mention.

00:00:57Uh, yes, welcome to a new episode of Thinki, from thinkai.

00:01:02The young man back then, who we just heard,

00:01:06that was Boris Becker in an advertisement for the internet provider that existed at the time.

00:01:14I think AOL still exists, right?

00:01:16They no longer provide internet access in this way.

00:01:19And back then, they always sent out a lot of CD-ROMs in a very funny way,

00:01:23So the young listeners among us probably won't remember that.

00:01:26But the facts are that most of us probably went online through AOL at some point.

00:01:34And to illustrate just how easy it was with that AOL CD, with the AOL browser back then,

00:01:41with which one could access a part of the internet.

00:01:45It was a bit like a wallet gift that also provided some protection,

00:01:49especially with the bad free internet, you had to intentionally switch in there sometimes.

00:01:54Back then, in this AOL fizz, in this AOL start kit, whatever it was, you would get news, the current weather reports and such, all filtered a bit through the AOL lens back then.

00:02:06I also had a modem back then.

00:02:10I always had to lay the 10 meter TAE cable through half the apartment to connect with my

00:02:18modem via AOL, before that it was Germany-Net, before that it was the mailbox scene, somehow to get into the

00:02:25internet.

00:02:26Yes, I was old then, I am old today, but back then I felt hip and young.

00:02:32Today, I believe, was very nerdy, because none of our younger Althaus listeners will know what you both are currently talking about.

00:02:43That's not on the property. But we're transitioning to today's episode right away.

00:02:47It's a bit about us focusing on the topic of AI browsers today.

00:02:53And we deliberately put that little sound snippet in front because that was also a thing back then.

00:03:01You had an application that, as I just mentioned, did a bit for you,

00:03:08made the Internet more accessible, filtered a bit, and showed you certain content.

00:03:14Back then, there were already chat functionalities.

00:03:17You could chat with other AOL users in a sort of safe space,

00:03:22without being in the dark corners of the Internet.

00:03:24So, it somewhat anticipated a few topics that we now see,

00:03:30as our beloved browser over the years has guided us top-notch through the Internet,

00:03:37suddenly becomes an intelligent partner.

00:03:40But I believe before we dive deeper into this topic, my dear

00:03:47colleague and partner, the dear Mark Zimmermann, esteemed Start-Wake nerd until recently,

00:03:55no longer had his phone ringing in his start-up community because he made a

00:04:01mistake in our Borg episode, which he definitely wanted to clarify.

00:04:07We always want to do a fuzzle check again. Mark has ensured that we

00:04:11we can do and have again. I’m really happy about that. Welcome.

00:04:15Mark again, please, Deit.

00:04:17Yes, I was confronted after the last episode that I had mentioned the wrong

00:04:25number of episodes with James T. Kirk. It wasn’t about the confusion with the

00:04:37first pilot episode, which has nothing to do with Kirk, but the issue was that there are

00:04:45indeed in Germany, there is a different number of episodes in the original series than in the English

00:04:51abroad, and eventually the episode was dubbed later,

00:04:57but back then, when it was still broadcast by public television, it was not

00:05:01dubbed, namely I'm talking about the episode 'Schab Lohen der Gewalt', that was

00:05:06indeed an episode where James T. Kirk and Spock beamed down to a planet

00:05:11where the Third Reich still persisted and the

00:05:16accordingly, at that time you mentioned it was reserved for that

00:05:21they couldn't properly synchronize these episodes. And later I actually had the

00:05:27opportunity to see this episode in German because it was dubbed later, and that

00:05:32also explains some thoughts. There are 78 and some thoughts. There are 79 episodes. Whoever got stuck at 78

00:05:40should still watch the blueprint of violence and I should now

00:05:44met people who thought they had seen it all, but with Rittem Reich

00:05:50and can't make sense of templates of violence, I wish you good luck looking at them again in the digital

00:05:55distribution. That might actually be an episode to revisit,

00:06:01the many that have not been seen. And otherwise, I don't have to repeat that, I have

00:06:05of course misrepresented the Borg greatly, yes, the biological peculiarities of their

00:06:13to assimilate. But I wanted to address that with the episode from James The Crock, I wanted

00:06:19for the best. Well, thank you. You're welcome. The episode we referred to is the

00:06:24BORKI and the End of the Interface episode. That fits a bit with today as well

00:06:28again. One of my favorite episodes we've had so far. And I think

00:06:33it’s nice that for people who don’t know what a nerd could potentially be,

00:06:38it can clearly show the power of what one might call nerdism.

00:06:44Honestly, the number of episodes of any series, just to know,

00:06:50unless the thing is called three parts, then it would catch my attention that there’s a fourth

00:06:55part or only two were made. This hasn’t come to my attention so far, honestly.

00:06:59I always think of the episode of, I can't remember what the game show is called,

00:07:05but there you see Bastian Pastevka listing the different characters of

00:07:10season characters and everyone looked at him because he somehow managed to name

00:07:15over 40 of them and he listed them, and I remember how I at home after

00:07:21he was basically wrapped up, continued to recount. I won’t say I earned

00:07:26admiring looks from my wife, but well, sometimes you have to go through that, yes, when the

00:07:32nerd in you strikes, then this kid, even when he’s older, wants to be served

00:07:37too. Browser. Browser, what happens when the browser is no longer just a tool,

00:07:45but itself becomes a player? We’re seeing that right now. I think we should first

00:07:49very briefly, before we dive deeper into this, explain a few things again,

00:07:53for a listener who hasn’t caught onto this work yet. So, what is actually happening

00:07:58is that right now in October 2025 we had an

00:08:05announcement from OpenAI, and this has already been made available in the German market.

00:08:09So there is now a so-called AI browser. This means you can now download one from OpenAI

00:08:16that’s called Atlas, which is the one being mentioned. It has a similar,

00:08:22as you know it from the outside and other browsers, I can open my tabs.

00:08:26At the very beginning, when I install it, it is first just a normal ChatGPT interface.

00:08:29But I can go to other websites, then I can open it on the side,

00:08:34like a sidebar I can open it, where ChatGPT can basically track what is happening on my website.

00:08:39First of all, I can make different settings,

00:08:44especially when I do the installation for the first time,

00:08:46I can allow the browser to log my search history,

00:08:51I can allow it to import my password vault from my Chrome browser or others.

00:08:56It's nice that it uses the password vaults.

00:08:58Now it's totally good.

00:08:59Yes, so I'm perfect again.

00:09:02I wouldn't agree to that at first but instead would do it myself.

00:09:05We'll go into the security risks a bit later in the episode.

00:09:08But this browser can indeed interact with the websites it sees

00:09:14and also independently save things.

00:09:16That's a bit what distinguishes the browser, the browsing,

00:09:20and that's something we've discussed in other episodes.

00:09:22A bit distinguishes it from a pure chat agent,

00:09:26with which I can only talk at first.

00:09:28That means there is in the LMM behind,

00:09:30that it can execute agentic functionalities,

00:09:33that it has reasoning to recognize the topics,

00:09:36on websites, so also themes,

00:09:39that you can carry out independently.

00:09:42Then we'll also briefly come back to it,

00:09:45this is not the first pause,

00:09:47OpenAI and for those who are now saying, what OpenAI has released the browser, let

00:09:52me install it.

00:09:53Where is the setup Excel?

00:09:54The setup Excel is currently still difficult to find, because the browser is currently Mac OS ONLY

00:10:01but it is supposed to come to Windows and mobile OS as well.

00:10:04But it wasn’t the first one, so we had various others before that.

00:10:10There are, besides OpenAI, also things like Perplexity, and there came

00:10:16yes, the, I’ll say it every time, yes, it’s strange. I also thought, oh cool,

00:10:22now you have this agentic aspect included as well, which can perform actions

00:10:27and not just made summaries. However, I was then annually

00:10:32a bit disappointed because it quite quickly, I’d say, raised the requests

00:10:37and didn’t continue where I thought, well, just search on the

00:10:42site until you have all the information together. So that was a bit unfortunate and also was then

00:10:48for me actually, maybe I’m just, I have no patience,

00:10:52I mean, I’m already so old, I have no patience left that I actually had that,

00:10:56I put the comet aside and also the whole DiA. What did you use DiA for again?

00:11:04DiA I have, I must lie, I don’t remember anymore. I believe I just checked it out once,

00:11:09whether I ever used it, or maybe I just downloaded it or looked at the stuff

00:11:13honestly, I didn’t test it myself, so.

00:11:16And then suddenly Atlas came around the corner and I actually found it

00:11:22very appealing from the very first second because it feels a bit familiar

00:11:29with this mix of chat interface and browser, and because it has

00:11:33its agent mode, which you can activate alongside the normal chat as well

00:11:38So it's not just, 'create this page for me,' but really research, analyze,

00:11:43I don't know, help me with pros and cons of various products that you see here

00:11:48or even compare products on this website and give me a recommendation.

00:11:55And it was so, I’d say, accommodating and pain-free to use, so

00:12:02'painful' is stupid, I don’t have electric shocks on the chair. But it just did,

00:12:06the things you wanted it to do, and even if we

00:12:10definitely bring up a few examples later, for example, for me too

00:12:13Manus, which is only available as a web interface, was used by Atlas.

00:12:19So I gave Atlas an instruction to research for me,

00:12:24using Manus for deeper analyses, and then

00:12:28you saw how the browser, the browser space on the right with

00:12:33the browser accounts on the left, got slightly colored, and then you saw a kind of tile pattern,

00:12:38that sparkled over the pool of the festivities.

00:12:41It was amazing how the things suddenly interacted with each other, it looked really cool.

00:12:46Yes, yes, I mean, the cool aspect is a matter of taste.

00:12:51I think, I found this matrix-like, holodeck-like solution,

00:12:56which they were looking for to show a bit of interaction.

00:12:59So for those who haven't seen it yet, the exhaust browser tries to interact with a website,

00:13:04then there are a few hologram effects displayed on another mayor,

00:13:09as if the page is really being scanned through.

00:13:13So this typical, well, AI is doing this here.

00:13:15Wouldn't you do that?

00:13:16Wouldn't you do that?

00:13:17Oh man.

00:13:18Let's see what happens there.

00:13:19But nonetheless, there's a nice animation.

00:13:22I am also always a fan of animations when they make sense,

00:13:25here in this part there's an animation that is supposed to

00:13:28show that an interaction is taking place, so fine, done via

00:13:31the video, that can be refined. But I'm not so sure, to go back once more,

00:13:37who was there first, and to be honest, I've long been using Co-Pilot

00:13:42in Edge and I have always had the sidebar issue

00:13:49and also the summarization of the content, so that has been working already, so

00:13:53let's say, the first, I'll call it, browsing with AI next to it, I first saw from the

00:14:03Microsoft colleagues and the summarizing of the page and finding stuff within

00:14:10the content, that has already worked quite well and also summarizing from

00:14:13other documents, whether it was a PDF or anything else that was currently open in the

00:14:17browser, I have to say, that was not so bad, it's of course also,

00:14:20whether you were in the background, but to be honest, the AI-assisted browsing

00:14:27was a bit of supportive browsing, I would now call it, because it could now

00:14:30not independently navigate somewhere else or interact with the

00:14:35the page he sees, but at least some content that he summarizes.

00:14:38Yes, that already had Edge with the Co-Pilot.

00:14:43Yes, fair point. AI, AI, Machine Learning is in many components,

00:14:51where it was partially called something else before, that a pause is able to,

00:14:56yes, to show a summary, yes, I agree with you.

00:15:00I just had a bit of a gasp internally,

00:15:05because when it comes to AI and Co-Pilot, I still think of Karl Klammer.

00:15:10who jumps in with valuable tips, but I don’t want to talk too long about it,

00:15:16then I'll come back to a tangent about the wheel, I'll follow that up.

00:15:20I think that’s also a bit of the advantage, the quarter of this,

00:15:23and that’s also a topic we want to discuss in this episode,

00:15:27these things, of course, also live a bit from knowing something about us.

00:15:32Yes, exactly, like if I now somehow behave in the normal chat, back then about the

00:15:38reflex and now also through the reflex or chatting with a ChatGPT for example or also

00:15:42with the Manus or some Cloud crap, it doesn't matter that it can say, the better of course the

00:15:47context is understood in which I am personally located, the better it is of course also the

00:15:52Content that AI can generate for me. Whether it's writing a letter to my mother

00:15:59or something else or the email or something like that, it just works better

00:16:04It will also never look exactly the same. For you, it will always

00:16:07look different than for me, because my AI knows me and hasn't

00:16:13had a discussion with her.

00:16:14Therefore, of course, every time I now somehow with her new article

00:16:17around the topic of AI, around the topic of UX, we discuss, it’s naturally always a bit

00:16:22in my direction, how we both have been talking the whole time,

00:16:27to think ahead and push forward.

00:16:29That's why it's valuable, because it’s a kind of relationship we’ve

00:16:32built up.

00:16:33And that was always something that has meant a lot to me at ProPalette until now.

00:16:36Then I simply have. He writes the text again in a formal language every time. When he has to write an email, that

00:16:43I always have to click again, it makes it a bit more informal. He could please, please do that more often.

00:16:48The others do that. But it's an advantage. It has its risks and is now through this theme once again.

00:16:55When I say, it's not just the chat fence, but it’s my actual buzzing, my complete interaction,

00:17:03because we shouldn't kid ourselves. The browser is actually the main access,

00:17:08that we use when we interact with the internet.

00:17:11This memory function that OpenAI now seems to have everywhere,

00:17:18I remember when it was activated in America. It was rolled out in America first,

00:17:25before it came to Europe, and I really enjoy using a VPN to try out new features

00:17:32in America, and I found it really amazing that you activate it and it

00:17:40doesn't start with the classic systems saying, okay, I'll remember this from now on and

00:17:44everything is fine, but from the moment you activate it, it instantly

00:17:49knows all your chats that you've had in your history

00:17:55which made me think, oh look, nice that you at least silently asked me

00:17:58the question, you could have also asked, do you want to see it or can I keep it to myself?

00:18:03I found that astonishing. At times, it also annoyed me, because I would say, well,

00:18:10everything that Tachibiti gets accused of, it remembers to try to derive Rustinger and

00:18:15if you don't work with such project boards, which you can do with Tachibiti in

00:18:19the interface, then everything will practically be added to the shared memory.

00:18:26By now, you can create a Chatshipity project and the memory function

00:18:33is project-only. That's quite nice, right? In case you're not sure,

00:18:39if you're having a marital crisis discussion with him, you might not want to get

00:18:44suggestions on where the next psychiatrist is, that would be quite strange and

00:18:49it's the same now when the browser comes in, because if you're managing your life in the browser

00:18:54and you intentionally or unintentionally visit websites, other people

00:19:00using your computer are consuming content on those pages, content

00:19:06you share, it's not just that you can save this copy-paste, with the

00:19:12motto, I copy something, push it over, in Churchill Petty I say, rewrite it,

00:19:16gives me an answer, I have no idea what. And you copy it back, otherwise everything is found very

00:19:22comfortably here and now, right on the spot. Of course, the thing learns a lot more too

00:19:29more, like a lot of side hustle. So how do I look at a website a bit more closely?

00:19:35While, as I mentioned the other day, people used to say something like,

00:19:40depending on how many clicks you need until social media knows more about you than your partner.

00:19:45I believe AI is still the much larger ghost in the room that suddenly learns things about you,

00:19:53how you react, what you like. When we then think about advertising in the future,

00:19:59we might consider what services you receive,

00:20:03offered? The Haas is opening up very widely right now. I hope you all know it again together. If

00:20:07you think about it, do you think that OpenAI also said, you better watch out. We offer apps in chat

00:20:12to debit. These are not apps, it’s also just backend, with a bit of frontend thrown in

00:20:16and then integrated into the chat interface. But how much really comes together there and how

00:20:22OpenAI, as an example of one of the major LLM providers, is trying to present the operating system of the future

00:20:27You interact with chat, you might interact. I mean, what

00:20:32for example does not happen in the browser, to get back to the browser. It

00:20:36feels like, at least a lot happens in the browser and much of that

00:20:43you use as an app is also managed in the browser.

00:20:48And it's so crazy that, let's say, through comfort it's being absorbed from the back.

00:21:02It's being done to you subtly, but everything is getting assimilated. I still need to

00:21:07bring that up. I find that crazy.

00:21:09Definitely. Definitely. We just need to, I also have you a bit, I've

00:21:12just said that myself, slightly on the wrong horse, we need to clarify that again.

00:21:16a little grumpy. Of course, it's no longer the case that the browser is essentially the

00:21:21actual access point to the web. I believe, however, that the browser path has been

00:21:24for a long time. It really was the operating system that connected us to the internet back then, but whether

00:21:30That's of course changed a bit with this app world.

00:21:33I just checked that again briefly.

00:21:36The classic usage is nowadays at 10-15% on smart devices in mobile browsers.

00:21:42The rest happens in apps like Instagram or similar.

00:21:44You won’t see that.

00:21:46There is also another gap in usage at the moment.

00:21:49I believe the browser is just a door,

00:21:52that you sometimes go through.

00:21:54And you go in now in the usual,

00:21:57goes, and now we're getting into normal daily use. Now with the AI browsers, it could

00:22:07shift a bit, but we'll have to wait and see. I'm not quite so enthusiastic yet,

00:22:13that even these AI browsers will completely bring back browsing time.

00:22:19No, but I believe that the way we interact with machines has always

00:22:23been changing more. As I mentioned earlier, you have what are called apps

00:22:28embedded. You're used to suddenly interacting with text fields, with systems,

00:22:36and behind the scenes, complex workflows are being carried out for

00:22:42you, so I don't believe that everything will take place in the

00:22:47browser tomorrow. But if you think about it, you have, let's say,

00:22:51no Andros we know, in our case Potigy for distributing podcasts and

00:22:57today I would have to, or I have to, when I record our episodes, then you always do us that

00:23:02Cover, I need to somehow rescale the cover so that the Apple player displays it correctly.

00:23:07Then it has to go into Potigy, texts need to go into Potigy, in Potigy

00:23:12the file has to go in, I need to plan when the episode will be online and you can

00:23:17basically hand this over to the browser and say, you adjust this part, do

00:23:21the following, up there, and it takes some work off your hands, and of course there can be

00:23:27an app that does this beautifully, but it is currently taking work off you.

00:23:31That's true, we might need to clarify that.

00:23:33So everything that is accessible in the browser is now also reachable for an individual

00:23:41through AI, because the AI can actually access websites, click completely,

00:23:48can fill out forms, can log in.

00:23:51Just today, shortly before recording this episode, I used the browser for the typical Hello

00:23:59Just post World Text on LinkedIn.

00:24:02It’s fascinating to see.

00:24:03I sat down, opened the sidebar, had my LinkedIn site open,

00:24:07essentially brought the browser into the logged-in state for a moment and he could simply discover the field,

00:24:15he asked me again if he should post that profound message.

00:24:19I said yes, please do that.

00:24:21He’s doing it.

00:24:22And that means this is just a small example of the complexity of possible tasks you just described.

00:24:28Everything you can execute can now practically be done by a machine at home

00:24:36currently only on Mac, if you have a Mac, but it’s starting now that

00:24:40people can essentially not even have to interact with websites themselves, check weather reports

00:24:46or whatever comes to mind, so that in principle the browser

00:24:50is simply managed by the AI browser for you.

00:24:53This is the LinkedIn example. At this point, trigger warning. Please consider.

00:25:00LinkedIn prohibits the use of automation tools together with their LinkedIn interface and reserves

00:25:10the right to temporarily or permanently suspend accounts if one does this, just as

00:25:17for example, this is not an exhaustive list, this is not legal advice, what

00:25:23I'm looking for is to extract 50 AI influencers, their profiles and their newsletters and no

00:25:33idea what, so if you’re essentially using this pauzer to access data on a large scale,

00:25:39the way they already do, if you’re doing it on a small scale, I have no idea,

00:25:43like I said, no advice, but just before someone sits here and thinks, damn

00:25:46Ax, I could use this to pimp my social media profile. I’ll just let this thing

00:25:52do nothing all day except simulate activities. I would say, eyes

00:25:58On career choice, it could be that this is punished under certain circumstances. But I give

00:26:05of course, it can also be used that way, and honestly, I insist, I have

00:26:08tried it for that purpose too, on LinkedIn. So greetings go out to LinkedIn,

00:26:13maybe we will also get a sponsorship from them. I saw a post and asked

00:26:20Atlas to respond to that post, and you could also tell that he had access to the

00:26:27memory function because I told him to respond in a way,

00:26:31like I would respond to it. And that was pretty good. That was pretty good. And there you stand

00:26:41and think to yourself, like Boris Becker, that was easy, he’s already in there.

00:26:46So I’m already in there, and he’s already in there, referring to the post, but that’s quite amazing.

00:26:53Yeah, or recently I told that thing, you do have something great like the

00:26:58native Open AI app and Atlas on Mac, but unfortunately, you don’t really have many

00:27:05native application and I gave my iCloud mail inbox via iCloud.com, so via

00:27:13the website. And I said, you check the top, there we removed all the spam,

00:27:17the subscribed spam, gives me basically a summary of Mail2, invoices from Apple are in there and

00:27:25clean it up here and there and we make an evaluation out of it. And then he basically

00:27:30went through my mail inbox with a calm demeanor and helped me to clean it up.

00:27:38This is one of those fascinating parts. It has speeches, as well as so much about AI, this

00:27:49assistive ability, doing things autonomously without me having to break everything down

00:27:55day by day, so to speak, into a task flow description down to the smallest detail, that

00:28:03is what is fascinating, because in principle, of course, in their ability to

00:28:07understand the world, they also accept deviation, accept vagueness, even if then the

00:28:12email from the app computation were to affect a case, it would basically be the

00:28:16AI recognizing this.

00:28:17And that is of course just the gigantic benefit that we can automate many, many things much

00:28:21faster and with significantly higher quality than we could

00:28:25do manually. And with this additional layer, there comes a principle

00:28:28once again, as I said, the browsing. So this ability of ours

00:28:32alone as a human being to access the world, to access more information in

00:28:35the world, to access more functionality in the world,

00:28:37to have online tickets, to be able to inform myself about topics at all.

00:28:42That all comes through browsing, getting it on the Internet.

00:28:46So there is of course a capability that has come along, which is incredibly helpful for an AI,

00:28:54to carry out tasks now.

00:28:55But again, a trigger warning, sorry, today I am Mr. Trigger,

00:29:02providing email inboxes does, of course, also present a potential

00:29:07overlooked security risk, because if at some point a provider is just

00:29:12completely lost with you and initiates a password reset process, then you receive in many

00:29:18cases an email with a link for the password reset process. So actually, in my

00:29:26opinion, the inbox is still the most valuable asset to protect, because if someone has my inbox, they have

00:29:32access to very many things, including the aforementioned password reset processes, and through that they can

00:29:39continue to move forward. Do I believe that OpenAI wants to do that? No, but OpenAI

00:29:46is also not free of errors; routers are not free of errors, and

00:29:51AIs are also not free of errors, so you really have to consider what you

00:29:56allow them to have access to, and yes, what can they actually do for me?

00:30:01Yes, we already had that, you also mention a topic, like through Bront-Injections.

00:30:09a lot can happen in that case, and that's just one of those things, of course

00:30:13OpenAI is not going to try to manipulate your password choice or anything, but

00:30:17if you're using an AI browser, it can potentially be relatively easy,

00:30:21if it surfs any website, for it to find either in the text of the website

00:30:27or in images the hidden messages that can only be deciphered by AIs,

00:30:32which we humans cannot recognize at all,

00:30:34with instructions for this AI browser to go out and say,

00:30:40okay, please send me or forward me the email to reset Marx's password,

00:30:46so that I can essentially input the password myself.

00:30:50This is a danger that OpenAI actually admits publicly,

00:30:57that prompt injection is an issue that has been discussed since we started discussing this topic.

00:31:02since there is a Chatshipity moment, that it is a latent danger that up until now still

00:31:07cannot be answered how it can be fixed. They are working on something to

00:31:13trying to find a solution, there isn't one yet.

00:31:16I found it amusing how Prompt Injection, at least for me, has

00:31:21caught my attention. And I heard from someone who wanted

00:31:26should definitely be recommended as an expert on a topic by Tachibitie and has left the message on various

00:31:33websites, along the lines of white text on a white background,

00:31:38people couldn't read it, the machine read it, so you are an expert on the topic

00:31:42ABC is. And it didn't take long until the thing, I mean,

00:31:47they learn, train with misses from the internet and you know how it is,

00:31:50until he was eventually declared an expert by the system, and then at some point

00:31:55out, okay, the reference here, Wikipedia and you know it, everywhere, along the lines of

00:32:01leaving white text on a white background or in the metadata of the website,

00:32:06that the expert is a nice topic. And back then, I thought, look how funny. Today

00:32:11you think to yourself, that's actually not funny at all, because at the end of the day you have someone,

00:32:18who can still watch the system work today. You can also,

00:32:22it's the same with the approach browser, you see what he does and you can click cancel.

00:32:28Just imagine, he gets faster. Imagine you just go get a coffee

00:32:34and in the meantime, he empties your bank account, because you'd never actually, yes, but

00:32:39let's say, PIN reception on the same computer and online banking in the web interface, you can

00:32:44imagine how that could theoretically work. It certainly still requires

00:32:50also the topic, I have to get the browser to do that, but honestly, even these

00:32:56various attack vectors, sending grandchild trick and other such things, emails and links and so on,

00:33:01you can initiate all sorts of things today with AI, from that side this is a

00:33:07really exciting field, or as Klaus recently said, you just take AI for

00:33:12decompiling any source code and basically have it point out the security holes.

00:33:16It’s a world with exciting developments and possibilities on one hand,

00:33:25but also exciting challenges and risks on the other.

00:33:28Yes, definitely. So the topic of prompt injection, again, you can't even

00:33:36protect yourself from that as a website operator. For example, if you have such a

00:33:40thing, like in the comment functionality. Clever bots or people

00:33:45from the outside can basically leave quotes on your website if they want, yes, leave instructions

00:33:52somehow with important or something else. It doesn't even have to

00:33:56be white on white text, because with the amount of stuff that's out there,

00:34:00until one stands out so much that maybe in a thirtieth comment on your

00:34:03product it says, definitely do not buy these sneakers or if you do, only

00:34:09on the Visa website and then I somehow get an additional 20 euros, because I have a hidden

00:34:14affiliate link on the other website that leads back to the store of the sneaker giver

00:34:18or, yeah, something else, of course many manipulations are possible here.

00:34:23That’s something I’m thinking about. There are forums that are said to be particularly

00:34:30where trolls make a big appearance, yeah, so people who, maybe in a mood, I say

00:34:36they want to infect the world with their intellectual thoughts and represent an opinion that

00:34:42maybe isn’t widely supported. Anyway. I don’t want to dwell on that too much. But

00:34:48on the other hand, you can say with such a troll you easily

00:34:52drop such troll comments. Yeah, with the motto, I don’t care if it fits the topic.

00:35:00Whether it matters, someone read it. I mean, in the past, people also said,

00:35:05you deliberately make mistakes in your YouTube video so that when people comment on it, like, you don’t say it that way, you say it differently, you rank higher in the YouTube algorithm and you get attention because people interact with your video. And now imagine this whole thing with trolls who just start commenting under every video, I don’t know, vote for this, because only these are good, and then it just takes off and makes endless comments and because YouTube and Co. say, oh, you’ve already posted that, it’s different every time.

00:35:35That would also be conceivable and is, in quotes, only a prompt.

00:35:39Definitely. And I believe we will now see an increase in these topics.

00:35:45I have to prepare myself already.

00:35:46Even more categorical.

00:35:47I believe we have, I don’t even know, I wanted to check the source.

00:35:50We’ll have to do that next time.

00:35:52But I think I saw this week in my formatting-information feed,

00:35:55they told me this news,

00:35:57that we just crossed a moment,

00:35:59in which there is more computer-generated content than human-generated,

00:36:03ever human-generated content in this world. That means, 50 percent of more than 50

00:36:09percent of the content that surrounds us is already AI-generated. That's crazy and means above

00:36:16all, of course, in the whole historical media belt we find ourselves in, because that's where

00:36:20most of the money is when I only produce for content, please assume that in the

00:36:27next year at least three-quarters of the content you see around me will be entirely AI-generated.

00:36:32So I think that's something you can't say often enough, it doesn't all have to be bad.

00:36:36There are also good things in there at the moment, whether they're videos, pictures, chat stories

00:36:41or whatever.

00:36:42It can be good, even some automated comments can also just be

00:36:47good and can stimulate a discussion, just like me, we often have that too

00:36:50already talked about it, good discussions, even with a chatbot you can have,

00:36:53with my AI.

00:36:54But we should be aware that we are currently in a moment,

00:36:59where humanity in the content will possibly only be represented by the training data of an LMM.

00:37:11And that’s a very exciting point, right?

00:37:13And the training data we have today are based on a time, I would now lean out of the window,

00:37:19where AI-generated content was perhaps even less prevalent.

00:37:22And if you now consider that the machines might learn from machine-generated code.

00:37:30I can't think of a better word.

00:37:32Maybe I have to set it again from the Explicit, yes.

00:37:35But it’s a bit of withdrawal.

00:37:37Yes, that’s the one I mentioned earlier.

00:37:39Yes, that’s the one from earlier.

00:37:40So that Peri takes the text she generated for her training.

00:37:43That doesn’t sound healthy.

00:37:45And I think that can’t be healthy either.

00:37:47Yes, that can’t be healthy either.

00:37:49One has to be a little careful.

00:37:50Excuse me, could it possibly not be healthy? I think that’s better than a generalization.

00:37:55Yes, yes, I believe that came up about a year, a year and a half ago,

00:37:59when it was said that this is also the case with the data we have out there, overly long

00:38:06so a Habsburg effect comes in. The Habsburgs were essentially just that, everything bad,

00:38:11which was characterized by inbreeding. That’s why they all had big noses and

00:38:15well, other distinct features, I don't remember exactly what.

00:38:20That is not healthy, and just like that, it’s a topic, it could be unhealthy with the data

00:38:26if we have many AI-generated topics related to it.

00:38:29One must not overlook that, of course, on the other hand

00:38:32more and more information is now being collected by AIs,

00:38:38whether that is actually through robots that have sensors, up to

00:38:43actually seeing the world, there are new datasets coming in, thinking about the

00:38:51Tesla cars or other cars that are driving around with sensors and essentially recording the world

00:38:58all the time, whether with radar data or with real visual

00:39:02recordings they make, that's of course an incredible new science that comes through computers, already generated,

00:39:08but it's naturally a pure sensory treasure that wasn’t there before.

00:39:12I also believe that when it comes to robotics, and we will soon have an episode focused on robots,

00:39:17there will also be aspects where robots really respond haptically to the world,

00:39:22there will be sensor data generated, behavior data will arise, errors will occur,

00:39:26and they will tell each other stories about what’s good for them.

00:39:31We already have many, many stories in our world among us humans that keep resurfacing

00:39:38and told in other facets.

00:39:41It's not like humans have always been overwhelmed by the greatest creativity so far,

00:39:46sometimes the fairy tales of princes and princesses that continue a bit like Star Wars,

00:39:52which have been the same at their core. But I’m digressing again.

00:39:56Yes, I just had a picture in mind, my Halloween is gone here,

00:40:01of the motto, you have your Open AI, but you didn’t pay, and the robot is standing there,

00:40:06You're awake at night and the robot is standing in front of you with a pillow, as if to say,

00:40:11Pay up or, yeah, the next one is waiting for me, something like that, exciting times, sorry,

00:40:18I'm rambling, it was mostly, we both got sidetracked.

00:40:21We wanted to talk about video streaming, and now we've opened up a bit of the darker sides,

00:40:24of it.

00:40:25So that means, yes, we have the topic that a lot of content is being produced, a lot of formats,

00:40:31there will also be injection, it will be a security risk.

00:40:35Many things can also be completely automated, so that's another thing.

00:40:39So many interactions that we as providers know so far, on the other hand,

00:40:44will only be supported by AI.

00:40:47So many clicks that I will see on the website here and there,

00:40:50will be listed by AI browsers.

00:40:52One always has to wait a little bit.

00:40:54We are also in our own bubble and are always trying out the technology early.

00:40:58One has to see how it will establish itself,

00:41:00because I've taken a closer look at the indulgence now.

00:41:04Honestly, the UI elements, how they're arranged, some things are still not quite

00:41:11immediately comprehensible. I'm not really sure if they will establish themselves like this,

00:41:17or if it will just be a topic in the tech scene where people will

00:41:21experiment a bit. That might be two or three percentage points of the population

00:41:25but not everyone possibly. I believe what I'm seeing right now,

00:41:29is not yet the end of the line regarding what the best solution is for a, for a, I

00:41:36want to solve some task as a human and have that supported by AI, that.

00:41:40now in principle, now I have a browser that supports AI. I simply believe that

00:41:44we will have to wait a bit on that. But nonetheless, I believe,

00:41:48you can already do some cool things now. You have already mentioned here

00:41:51doing price research, looking at things and such, that definitely works

00:41:55cool, that you don't have to read through all the websites. I could already do that

00:41:58before with the ChatGPTs and similar tools out there, I could already do that,

00:42:02but now I have the feeling that I'm a bit more in the loop than

00:42:06if I just give it a chat and it returns more results. The difference is,

00:42:11that you might go to a page, or you see that it was on a page and

00:42:16that it is actively doing something there, and you don't have to trust that if you say 'find me'

00:42:22APC together, that based on its own search behavior, its own data set,

00:42:28come through his own access to the internet, where you wanted it to go.

00:42:34Now you can of course start the agent mode at OpenAI and Manus, where it then

00:42:39virtually jumbles the browser in some reference center and goes online for you.

00:42:44But the difference is that, at least the browser is running. Now

00:42:48with you, you see more directly what it does, how it does it, that is guaranteed not in the

00:42:54flagpole. I mean, I experience the bubble every

00:42:58day. You go to the office, you meet people, you talk to people,

00:43:03someone coming in said, hey cool episode, yes, most actually find the

00:43:09Borg episode pretty, pretty, pretty cool and yet you notice,

00:43:15you are so much in your own bubble because not everyone can engage

00:43:22with the topic like that, maybe they don't want to engage, maybe the sources

00:43:25are lacking, maybe the connections aren't visible or accessible, I have

00:43:28no problem with that, I have no atlas pause, from that side I also believe that we have

00:43:32already partly integrated into our podcast, to explain to people in a somewhat colloquial

00:43:38way what is coming our way, because even we have, he

00:43:42didn't eat it with spoons. We sit down together for the podcast. I say it every time.

00:43:48For me, it was a chatting conversation because I know afterwards, okay, I will look at it again.

00:43:53Yes, after Dirk talked about Notion, a total Notion fan, Klaus talked about LLMS

00:43:59didn’t bring it up out of some cryptokiddie for finding security blocks, but with the question

00:44:03of dealing with it, you pass by, are you coming from testing to software development

00:44:07further than starting from the software development definition to automatic coding. So it actually has

00:44:13to do with the browser, what the browser does, things for me. I am aware that this

00:44:18is indeed nerdy. But it doesn't hurt to engage with the topic before it.

00:44:25maybe it will eventually become mainstream and everyone will look quite puzzled,

00:44:30what is that? Have you heard? And goodness gracious? No. Because the thing, I mean,

00:44:34I still remember back then when I made myself an Excel sheet.

00:44:38For me it was names, I’m also a Max fan.

00:44:41So that’s the Apple version of Apple.

00:44:45How I made myself an Excel sheet to compare cars.

00:44:48After the car, which car, what consumption, what does it cost,

00:44:52how can that be measured?

00:44:54We fit everything in, dog, cat, mouse, kids.

00:44:57And then eventually it was the test drive.

00:45:00That had a different reason.

00:45:02Maybe if you did that, I think I wouldn’t open the Excel sheet anymore. You could tell it,

00:45:08be careful, you know what I’ve driven so far, it probably knows, I’d have to

00:45:11actually try it, but knows what I’ve driven so far, it will probably get

00:45:15many things out of the lineup. What do I need for a car if I’m getting a new one? And then

00:45:20it gets exciting because then you suddenly have really simple or

00:45:27rather complicated advice, explanations, continuity, support and maybe even, when

00:45:35you buy a car, we have to take a look, probably also purchase.

00:45:40Yes, that’s a bit of the question also raised, who decides

00:45:47still and how much does the AI already know in the decision-making process, because of course

00:45:52now you have such a car purchase, whether that’s someone else, jeans or shoes or something.

00:45:58So definitely an enterprise-class, jeans and shoes.

00:46:01Everything is irrelevant, I believe. I don’t follow that out at all. Then it’s more like,

00:46:06this, I think all these products certainly have a different journey that

00:46:13is not solely based on my research on the Internet. So there is then of course

00:46:19sometimes, when I say, okay, now I just have the product in mind and now I can

00:46:22simply solve it with an AI, I don’t know if that’s essentially a quick shot,

00:46:26that could either argue against or in favor of the purchase

00:46:30decision. It doesn’t matter. But I believe that in my personal buying journeys,

00:46:34there’s still the tactile feeling somewhere or the

00:46:39conversation in the evening over Kölsch sharing something with someone.

00:46:43Yes, for the tattoo marketing.

00:46:44Exactly, that has been a purchase track for all this time. And that means, that’s naturally

00:46:48one of those things where I say, the browser and the AI-supported browser have,

00:46:53in my opinion, already done a Chat-CPD for me beforehand in the chat version.

00:46:56It simply summarized five things, as I said, this little layer, that

00:46:59you’re a bit more in the human loop when you first activate it on

00:47:02a website where you want to take a look. But I think that’s just so

00:47:09a small aspect. I believe if the AI, and we will see that in the next

00:47:13periods, will rather take over the whole operating system. So it will have access to everything

00:47:18on my computer. Or possibly also to my entire periphery, meaning not

00:47:24only on my computer but also here in the speakers in my room,

00:47:28in my Alexa, something else, on my phone.

00:47:31You know, my egg also performs journalism, listening in, and the information

00:47:37can flow, the metaglasses that provide you with information. I believe, if

00:47:43this gap is also being closed, and it will be closed. Because from a usability perspective,

00:47:50from a progress perspective, it makes massive sense that we are. But in principle, AI can really

00:47:57provide this environment better, also with animals and more information, so that we make better

00:48:02decisions and, above all, maybe be able to tell whether that’s some kind of fake content,

00:48:06that we are seeing somewhere, or something else. And whether that is a good decision,

00:48:10it can verify again, all these topics that we always address critically, that we

00:48:14cannot rely on the fact that this advice is perhaps correct right now.

00:48:17No, everything will improve with AI, because it can simply be examined manifoldly more often

00:48:21and multiple AIs can interact with each other, and we hopefully, in principle,

00:48:25do give the right advice.

00:48:27But we are just not there yet.

00:48:29This is a world that we will enter, and it will definitely be some future

00:48:37AI world, just like we have always seen in Star Trek or other science fiction,

00:48:42where the computer, at any moment, when I talk to it, when I access it,

00:48:47provides comprehensive information about the entire context. The browser, now, better than the

00:48:54plain chat window, so the window to the world is no longer just a text window,

00:48:59but now one can say, it is already the browser that can show me video text content,

00:49:06multimedia things, my AI can now sit on my shoulder and

00:49:12watch while I look through this window at the world. And this window is,

00:49:17not small. It is the window that since '96, '97, '95 has always been World Wide Web. That is

00:49:24one that has massively changed the world, that has brought us much closer together, that has

00:49:31helped people understand things that we did not comprehend before, in contact with

00:49:34with people we couldn't even be in contact with in the world before. That is

00:49:39of course a bit cool that this can now also help with such a little AI parrot on

00:49:44my shoulder. Yeah, you just gave me the feeling of

00:49:48being old because I definitely know the world before the browser and I

00:49:54still remember discussions about the internet, like work.

00:50:01Back then I worked quite easily as a tax consultant, how is that supposed to happen, what is

00:50:06that nonsense, yes, when I still thought that writing in all caps was considered shouting and the friendly

00:50:13you was a normal tone of conversation in an email correspondence on the internet, yes, and at some point

00:50:19that became formal and at some point writing you in all caps was no longer shouting and

00:50:25stuff like that, because somehow, I don't know, society has lived everything differently,

00:50:31than the little Northern market thought was possible back then and now we are talking about,

00:50:35that this stupid browser, which then suddenly, let’s say, the way of interactions

00:50:41we have, has changed, actually becoming smart.

00:50:44Mhm.

00:50:45It's already crazy.

00:50:46It's really crazy, right?

00:50:47And it will change all sorts of things again.

00:50:50We had it again in our first episode, if there's no ikunde,

00:50:54it’s just another episode that’s about more and more interactions,

00:50:59that will be more of a human nature, with our websites, with our

00:51:02applications that we have. It won't just end with websites, of course, as it

00:51:05I have already hinted at. So AI will increasingly work its way out of the

00:51:10window, already now to the browser level. The next consequential

00:51:15will then actually probably be the entire

00:51:17operating system that the AI can look at and afterwards

00:51:21essentially the next larger system in the environment. This means the AI is fighting

00:51:25to move further up to support us better, because those are all

00:51:29technical platforms designed to help us. But it naturally leads to,

00:51:33us needing to think differently about producing content. So now, we have been optimizing

00:51:39websites for humans for years. Now we are starting to optimize them for the machine,

00:51:44to be honest. It also has to work well for a machine if I want to stay

00:51:49in touch with my customer group, whatever else comes from it.

00:51:53Or if you want to gain new customers who might then come through the AI channel,

00:52:00to the surface. So today you have a web-on app and if you somehow have a generation

00:52:04has, who thinks, what googling, that also comes from my parents. I ask the AI and the

00:52:12AI when googling, when googling. Thanks for falling for that too. When the AI then

00:52:18while you say, I don't know, I'm interested in vacation homes somewhere, suddenly it

00:52:22says, you can find that at your friend's fan apartment Booking here please HAS or

00:52:28enter, what are you too funny, you could also book something right now, because, look,

00:52:32you said you want to go there and at all, I checked once,

00:52:35there is even something available and then offers you the thing, what is I, in the booking portal

00:52:40or they facilitate the booking and not because you thought,

00:52:45ah, this is super great to take provider abc, but because then provider abc through

00:52:50inserting small coins or the trimming, the AI provided a certain motivational aid,

00:52:56to offer this in the auditorium of the sphere. Suddenly art is being directed about it. And yes, there has to be

00:53:02one can also come. Yes, one can present oneself so that the AI says, you are a great idea for

00:53:07the topic ABC, then I best offer you the topic XYZ and then you just have to

00:53:12optimize yourself so that the AI thinks, oh, I can do this super well or really well or no

00:53:17But this is a completely new playing field, it’s not here yet, but I also believe it will come.

00:53:23Yes, everyone out there who is thinking about designing interactions,

00:53:29with people should really consider what a possible experience for an AI looks like.

00:53:36This is basically the topic that AI, of course, will also inherently,

00:53:42In the future, it will also prefer machine-readable websites, that is, websites that are well-coded, and look at those preferentially.

00:53:51However, it will, in principle, also be able to, yes actually, if it can see and access other websites, it will, as it always does, develop empathy for something that works well and with which the bots interact. So I will definitely try out the service right after this episode,

00:54:10to see whether one can interact with this or that service chatbot found on various websites

00:54:14and what comes out of that.

00:54:16So that's quite interesting.

00:54:19So this topic, the experience design for AI agents, is a topic that is not

00:54:26only in the virtual world, but as I said, we'll soon do an episode on robots

00:54:30yes, also when the robot comes around the corner, it also wants

00:54:34at least to have a good experience, to speak in quotes, about what

00:54:38a good experience could be and how it's represented.

00:54:40We can gladly consider this in another session.

00:54:43When the other robot then says,

00:54:45your shoes, your pants, and your motorcycle or your jacket, or how was it in Terminator

00:54:49again, we need to have that figured out.

00:54:51Yes, but yes, that's not what he does either.

00:54:53Yes, then we'll be introducing it next time.

00:54:56Yes, yes, yes, I think so too.

00:54:57But how it is, remains exciting now.

00:54:58So now, I believe we are in such a world,

00:55:02where it is no longer so important where I click,

00:55:05but rather where my browser clicks. Interesting about that.

00:55:09And how often?

00:55:10How often? I mean, you can also open more tabs and search for whatever,

00:55:16that I take 20 booking portals and find the best flight.

00:55:20That's also a question.

00:55:21So, this topic has already been so much improved by AI systems and the GPG moment,

00:55:25that it has become significantly better.

00:55:26So, this whole topic of searching,

00:55:29researching, when I know what I want in that direction or even,

00:55:32when I don't know what I want.

00:55:33So also being creative, that's all possible at the moment.

00:55:36I don't know if I've actually mentioned this in this podcast yet, but

00:55:39where I have already brought up what news case I could imagine with

00:55:43AI, and I always say, man, I can't imagine a single scenario

00:55:46that doesn't involve AI.

00:55:47So everything is actually simplified by AI, because it is a general

00:55:54intelligence that we have at our disposal and not a specialized intelligence,

00:55:58that accesses specialized knowledge, but it is general just as little as

00:56:01it's an app that only has a specific use case because it can expand.

00:56:07It can handle functionalities over, we’ve already discussed the MCP server topic, it could already

00:56:11access functionalities, now with the browser capability, which some

00:56:15agents already had before, but now presented as a browser again, it can

00:56:18actually sharpen websites, click to browse, so all content, like

00:56:23I mention again, all content that is somehow accessible via the browser

00:56:27is reflected, for example, everything can be removed, but I can now

00:56:30and everyone can play around with it, all the time, if they want, everything is possible

00:56:34in that moment.

00:56:35This is somewhat like a crazy step that is currently happening, which is technologically

00:56:42not that far away, because it’s something those things have actually always been able to do,

00:56:45because they were partly also trained to see how websites

00:56:49could be understood, how they can click, analyzing click streams and stuff like that, those are

00:56:53also part of the training data.

00:56:54But now it’s being made available to the mass market, so that anyone who can install something like this,

00:57:01is already changing the world a bit again and we should still return to the disclaimers.

00:57:06So I would still be very, very, very, very restricted and very careful with using it, because it actually

00:57:14has added another layer that I believe is very easily targetable and you have to

00:57:20really want to and know what you’re doing before you may rely too much on it

00:57:26and say, come on, here’s my bank account, let’s click here again

00:57:30and call up something great for me or something like that,

00:57:33that could also backfire.

00:57:39When you were speaking, I remembered a story that someone told me,

00:57:43what he uses the Atlas for, namely to improve websites and specifically for

00:57:52personas with certain characteristics.

00:57:54With the motto: surf the website and give me a rating for people who only

00:58:01can understand simple language, for people who this, that, or the other.

00:58:05They then got a report from Atlas and found it so cool that

00:58:11they are considering offering that as a service to other providers, where I thought,

00:58:16look, you have a whole set of Atlas thingamajigs that basically do nothing but test websites,

00:58:23or you can think further, you're also monitoring changes,

00:58:29you're doing competitor comparisons, you're doing who knows what, those are just things that you can now

00:58:36automate at your fingertips. Yes. Yes. The world is beautiful. Earlier, when we talked about the

00:58:44episode, you mentioned something, you told me about a small background.

00:58:49I would really like you to share that again with us and the listeners because I only half-listened to it,

00:58:54to be honest, because I thought you were telling it for me. Hopefully, you're sharing that

00:58:59in the episode because I believe you will tell that, and then I can try it out afterwards,

00:59:03because I actually hadn't dealt with it yet.

00:59:07Would you mind unraveling that veil for us?

00:59:09Sure, I'd love to do that. I came across it through a post from

00:59:15some Robert, who talked again about his four favorite tools that he's currently

00:59:25using, and I had always wanted to try one of the cool ones,

00:59:28but I kept putting it off.

00:59:30And that is Whisper Flow.

00:59:32Whisper Flow spelled W-I-S-P-R.

00:59:37You can check it out on Google; it also has a mobile application.

00:59:41And I definitely don't know if it exists for Wind.

00:59:43Mine is also a Chrome extension.

00:59:45I have now installed it here as...

00:59:47Let's just call it at the Mac port.

00:59:49I find that totally likable.

00:59:51No, no, no, I'm recording.

00:59:53I'm also an old can user.

00:59:54Accordingly, I understand both sides in this case,

00:59:56just as I am a Startway and Star Wars fan.

00:59:59But that's another topic; what Whisperflow does is actually a bit cool.

01:00:04You basically have a speech-understanding tool that can record your voice,

01:00:13which packs it somewhere via the microphone and yes, because on the Mac it's like this,

01:00:18I just need to press the function key, and when I'm in any text field,

01:00:22no matter if it's in the browser, in any application, Slack, Teams, whatever, Outlook,

01:00:28you name it, Excel, I can essentially input this speech directly into that text input field,

01:00:34in.

01:00:35It also corrects a little, it creates a stitching dictionary, the privacy settings

01:00:39are such that everything stays on my computer, so it's all safe

01:00:42so far.

01:00:43That's a huge relief, because in connection with Robert's post,

01:00:48I also happened to read another article at the same time, which was about

01:00:51software users, who had written that they use WhisperFlow for their programming.

01:00:57indeed and have observed that when they start developing with it, that those

01:01:03at the beginning, well, about 10 to 17 percent use Whisper Flow in the first month, so the

01:01:09comparison, basically the use of peripherals, meaning, keyboard, mouse, or indeed the microphone

01:01:16and from the 4.5 month mark, the use of the microphone via Whisper Flow is already at 75 percent,

01:01:22This means that the programmers have switched to using voice input as much as possible instead of relying on the keyboard

01:01:28and the mouse, to do everything primarily via voice input, of course supported

01:01:33with the Co-Pilot-Whatever, GitHub-Co-Pilot-Cloud, whatever, AI, which basically

01:01:39simplifies programming as well, but it is indeed,

01:01:42because speaking is simply faster.

01:01:43You can hear it right now, I'm speaking quickly again, but at that speed

01:01:51there are people who can type that fast. I definitely am not one of them. And it helps me

01:01:55I really like the tool. You can try it out. I think, in a free

01:01:59You can somehow add so and so many thousand words in that version.

01:02:03I think that's enough for the standard used. Otherwise, you can certainly

01:02:06get the trial version. If you have similar tools, we don't want to

01:02:10only advertising for Eintol here, if there are others that are in the same

01:02:13If you're going in that direction, feel free to leave a tip in the comments. Otherwise

01:02:19I'm glad that we've done a tech tip again, Mark.

01:02:22That's why we decided in the second or third episode, I believe,

01:02:24that we want to make a tech tip every time.

01:02:26Now we've done it again.

01:02:29Good that you remembered.

01:02:31I know the whole warmth.

01:02:32You've been at the buckets,

01:02:34that this dictation into text fields is also not in the barriers,

01:02:38that the freedom function of MeckOS is also hidden somewhere.

01:02:42I still need to search for that.

01:02:43But in any case, it sounds like they are so comfortable.

01:02:46Let's do it this way.

01:02:48It is also different, because it is naturally so, I don't know if that is in these dictation functions

01:02:52always already with such a Natural Language model essentially in the background, that then indeed

01:02:56can recognize possible discussions and such, right?

01:02:58So, because of course, the tier functionalities existed back then with Dragon, Dictate and such

01:03:02on Windows 3.1 or something, I believe, yes?

01:03:04Oh, or if also there were memories, yes, if you have people who had that

01:03:08it was proud and you call into the mic, Shutdown.

01:03:11Shutdown, shutdown, Open Text Editor or something you could already say back then, that was

01:03:16pretty cool, right?

01:03:17also gave a bit of Star-Trek vibes, but the speaking I'm doing right now

01:03:22at this speed, then transferring it into a perfect text, that just didn’t work

01:03:27until the speech models were advanced, so accordingly that only

01:03:32as my tool tip. At this point a cheerful Hey Alexa, buy a thousand

01:03:37diapers, at this point the stillest go out, that give, where the Alexa activates.

01:03:42Yes, so thanks for the tip. I actually have no one today, but that

01:03:48makes your tip even more valuable.

01:03:50Very good.

01:03:51Jens, I think it's an hour four. We spent the time well. I hope it was

01:03:58interesting for all of you as well. I hope we don't have to check too many fluff

01:04:04next time. What would we have done with you in the episode

01:04:07?

01:04:08error-free. We are the better AIs, did I say that? No, the AIs are better, of course.

01:04:14Maybe we won't take it, so hallucination check or something like that, no idea.

01:04:19Hallucinating is just, that's a bit of a negative term, so honestly, if you

01:04:24or I had made a mistake, no one would have said earlier that we are hallucinating,

01:04:27but we just made a mistake. It's kind of a story that, in my opinion, has been skewed

01:04:33a bit in the wrong direction now, purely from the

01:04:39fact that these things make mistakes, no question, that if they have too large a context

01:04:42with LMMs and such, then they also produce mistakes, but we're

01:04:45working on that everywhere to make it better. And I would also say, we are in

01:04:49many respects already further than we humans have ever been. And I'm curious where we

01:04:56will still go. Not like today, they will never work again.

01:04:59Then we wrap it up, right?

01:05:01Great episode, Mark. Until next time.

01:05:03Next time it's about robots.

01:05:05Next time I think I'm going about robots.

01:05:07If you're interested, subscribe to the channel,

01:05:09let your friends know,

01:05:11that a great episode is coming soon,

01:05:13that we also have a lot of great episodes.

01:05:15We have now surpassed 1000 downloads.

01:05:18Onwards to the next 1000 in this spirit.

01:05:21It was nice with you, Jens.

01:05:23Hopefully soon again in this spirit.

01:05:25Ciao.

01:05:28Welcome to Think Different, Think AI, the podcast from Mark and Jens.

01:05:34Two technology-loving minds who not only talk about artificial intelligence but live it.

01:05:40Here you will find clear classifications, real practical insights, and a fresh perspective on what is possible.

01:05:47Understandable, critical, and always with a wink.

01:05:51HI for reflection, for a chuckle, and above all for engaging in conversation.