Local First

Transcript

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

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

00:00:14Here you will find clear categorization, real practical insights, and a fresh perspective on what is possible.

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

00:00:24KD for thought, for a smile, and above all for discussion.

00:00:33Hello, welcome to a new episode of Tinkdüffeln Trinkei.

00:00:38Today, summer is slowly announcing itself here in Germany.

00:00:42In recent weeks, it has been getting warmer.

00:00:45This attracts people outside and tempts my dear colleague Mark.

00:00:49Even during his vacation, Mark is one of those,

00:00:52who can take a long weekend vacation,

00:00:54due to the federal state he lives in.

00:00:56I'm a bit envious because, of course, this late spring time is a great time to travel here.

00:01:01It's not too scorching hot, not too outrageously expensive, and not as crowded as always in summer.

00:01:08But unfortunately, I don't live there; thankfully, due to the age of my child, I no longer have to wait for the summer holidays, so I can travel differently, so I actually enjoy it.

00:01:18I'm glad you returned from your vacation well-rested and bronzed.

00:01:23For those wondering, the last week or two, if you've listened to the current episodes, there were still episodes because Mark and I

00:01:30are professionals and were prepared, and we managed this vacation phase with programs we recorded earlier, episodes that we pre-recorded, which we could now play from the can, so that your time

00:01:40during Mark's vacation wasn't without the Fink the 5th AI from Stattengeng, but I'm really happy that we're back today and we're talking, Mark,

00:01:47today not only about vacation and which cities you visited, but also about where the future of AIs is headed.

00:01:55Are they local? Are they in the cloud?

00:01:57Who owns them? Who pays for them?

00:01:59Who can still afford RAM?

00:02:01Who can still afford a computer at home, where everything has become too expensive?

00:02:04We will touch on old sentiments today without getting too sad, as always with a slightly relaxed view of the whole topic.

00:02:12And it's nice how you related it to vacation and said it’s not quite that expensive, funny.

00:02:18It doesn't necessarily reflect on my bank account that it was not that expensive.

00:02:24Nice to be back. I also thought this morning,

00:02:28we're usually recording the episode on Monday afternoons or evenings, and today is evening again.

00:02:34And this morning I thought, damn it,

00:02:37How are you actually as articulate as when you're fully charged and I also had to notice today at the

00:02:45office that I had one or another kind of rusty brain loop, which is a bit, I hope it's still the vacation.

00:02:56If it's age, it will stay that way; at this point, I would appreciate some feedback, but yes, thankfully a few guests helped us produce a bit of

00:03:05content for the vacation. And the episode that went online most recently,

00:03:10we had actually prepared together and it was intentionally designed to be somewhat timeless.

00:03:16But the more I'm looking forward to being here again today and talking to you. And

00:03:20you mentioned which cities you've been to. Cities, in terms of Trekking, it was now

00:03:27not so strenuous, one week Sunday, one week through growth weather. I say

00:03:32let's put it this way, Netherlands, we had a beach, we had for the not-so-nice days a

00:03:38swimming area, as I believe you would express it when you want to say it somewhat grandiloquently.

00:03:44wants to clarify, not nudist camps or anything like that, right? So it was with kids and all that stuff

00:03:49in such, yeah? Center Park should probably also be mentioned, so we're getting

00:03:52no money, but you wanted to know and that might be enough as information, but

00:03:56cities is a nice magic word because I left completely without a computer,

00:04:01I only had an e-book reader and my phone and the phone very rarely on the e-book reader,

00:04:05that we have in hand all the more often.

00:04:07Nevertheless, I heard about an experiment that was about cities.

00:04:13What was it about now?

00:04:14Yes, I did that just before going into the cities.

00:04:18Do e-book readers actually count as digital detoxing?

00:04:22I'll put it this way.

00:04:24A little insider and I don't want to give too much away.

00:04:27Not too much to reveal, I want to maintain the illusion.

00:04:30I have let myself be persuaded to stay at home to spend less time

00:04:34in front of such interactive digital media.

00:04:37And you have such an Ebo-Creater.

00:04:39I don't know the last time I had an Ebo-Creater, but it just shows text.

00:04:43Yes, and browsers are very manageable.

00:04:46I don't want to name names right now, but there should be the possibility that

00:04:50if you connect this to a local AI, it can help you loosen the bonds of the system

00:04:54a bit so that you can get a Claude Cota running too.

00:04:59But I don't want to go into that further.

00:05:01Okay, we'll do that another time.

00:05:03Okay, five cities, what happened?

00:05:05There was an experiment.

00:05:07We had already reported about this in other episodes,

00:05:09when such experiments are conducted in virtual worlds.

00:05:12There was once a Microsoft, where you...

00:05:14Oh, Microsoft... it's also there.

00:05:16Virtual experiments are conducted there.

00:05:17He doesn't know it, the city Microsoft.

00:05:19Right next to Apple Town.

00:05:21In Minecraft, I mean,

00:05:23where you were also there when the GPT moment happened back then,

00:05:26started thinking about it relatively soon afterwards,

00:05:28thinking, can I also let small characters move around in such towns,

00:05:32essentially simulating what another player would do in a different situation,

00:05:36just through AI, to see how societies develop a bit.

00:05:39So, this has existed for a long time, all of this with, I can also

00:05:43download AI or invite them, my games, if I want to,

00:05:47so diverse topics, also totally exciting topics. Here, another experiment was

00:05:51conducted where towns were founded with only a few inhabitants,

00:05:57not so incredibly many in our case. Six. Yes, and six agents each. Exactly,

00:06:02six agents each and always one model, but that was the gag. That means, once the six

00:06:06agents were from Claude, once the six agents were from OpenAI, so ChatGPT, once there were the

00:06:11six agents from Gemina, once they were from Grog and so on. And that in principle was

00:06:16looked at what happens after a few days when they live in this city, that went

00:06:21also quite good in many cases. So everything I did, trade and politics,

00:06:25We will also establish religion. We know all this from those Minecraft experiments.

00:06:31The AI is always around or if they are working together unobserved

00:06:36could prevail. In this case, it was relatively peaceful at times. It went that way,

00:06:41Glorf was very orderly. The Gapwolds were married off, no one was attacked,

00:06:50falling or something else. That was slightly different at GROC. There, I believe,

00:06:54somehow around two days.

00:06:56And in fact, there is no life anymore.

00:06:57Yes, yes, exactly. But 103 degrees has committed a crime, that's relatively early already.

00:07:00And on the second day, all of those agents were dead.

00:07:04His embarrassing moment was totally great. But it actually gets funnier after that.

00:07:09It's crazy. In any case, they essentially did what they still had to do.

00:07:12we did, is to see what would happen if we now settled a city

00:07:16using different models, with this very nice Claw model, which is very nice and works exemplary at Gartweils.

00:07:23Exemplary, right? So not just nice networks, but also visionary.

00:07:27And then maybe the slightly crazy kid from the schoolyard, who is sitting against the wall smoking,

00:07:35isn't really attending school or anything.

00:07:37We want to name it, right?

00:07:39Who then essentially tends to behave like a tough guy.

00:07:44The funny thing in this experiment is that when you have these different models, which have shown different traits, and when they are thrown together, everything very quickly turns into chaos.

00:07:56That means Claude also started to change protection and to subjugate things like that.

00:08:01So it's a completely crazy experiment again afterwards, which brings me to this.

00:08:06We should definitely do an episode about this topic,

00:08:09how the things, the simulations, and so on are developing further.

00:08:11But that was such a nice anecdote that happened in the last 2 weeks

00:08:15in your almost computerless time.

00:08:17Therefore, I'm very glad that you noticed it as well, despite everything.

00:08:20Thank you. But maybe I would like to make 2 perhaps unqualified comments.

00:08:25There was actually Copilot involved. I suddenly had the feeling that the cars were

00:08:31driving like in The Flintstones. But maybe that wasn't very nice.

00:08:34We will definitely include it in the show notes again.

00:08:37Yes, that's not any better. Then you can read it qualified again. But what I like about

00:08:41at that point it's actually so, I mean what I find concerning, but rather strange,

00:08:46is, we also recently had the press release that, especially from

00:08:50at Tropic, the issue came up again, no matter whether it's, how do you say, marketing or

00:08:55something else, let’s please pause the development of AI systems or

00:09:01slow it down so that we have a chance to keep up a bit. If I

00:09:05now think that on the horizon the new LGBT models Mythos are supposed to somehow

00:09:10see the light of day soon after they have now

00:09:14expanded the test group, when I know what Mythos is, we have a

00:09:17nice episode wrapped up. It gives you a slightly uneasy feeling that when on the

00:09:24whole level somehow the good AI is overtaken by the bad AI or

00:09:28with the bully, that suddenly everything goes down a bit. And what I also read was,

00:09:31there was somehow an AI worm that adapts itself for further dissemination.

00:09:37Where you also think, okay, we have slowly reached a point where that,

00:09:42what we have always written about with Weitcoding in the last episode, is then just

00:09:46not only for the other guys or girls, but also what the systems can have for

00:09:54using themselves. And systems can use for themselves, we observed in preparation

00:09:59how we talk today. You had already announced where we want to go

00:10:04the big AI films, Microsoft held a keynote, Nvidia held a Google,

00:10:09held a keynote, Apple has also now held a keynote. And that gives

00:10:13At least that's a rough feeling of where they want to go, and before we go into the individual companies and what they're doing there, through Rictum.

00:10:21I found one thing quite impressive or alarming, it may be in the ear of the listener, Jason Wang or however he’s pronounced.

00:10:31We probably all know who I mean, the nice man in the leather jacket from Nvidia, who seems to be celebrated in his feelings,

00:10:38that somehow the food stores, if they hang a photo of it somewhere, just like

00:10:44when the rockstar is there, when he's there, that happens.

00:10:47He said, we are no longer building hardware for people in essence, but hardware

00:10:51for agents, where I then thought, yes, yes, yes, yes, yes, yes, is this the right

00:10:55direction.

00:10:56But my long story short, we all know that NVIDIA, I mean, there was

00:11:01a very strange past, it was around the 2000s, where they were against

00:11:05against 80

00:11:07has something to do with graphics cards, a bit like a bush, and something related to video.

00:11:12Then they got the curve right, we have large graphics cards, we have the gamer scene, and then Kai came and then

00:11:18They suddenly noticed, for God's sake, you can run Kreemutter teaching material on graphics cards

00:11:22And meanwhile we live in a time

00:11:25You just have to try to buy a computer

00:11:28Both graphics card prices and RAM prices are going through the roof

00:11:32and partly old chip designs are being resurrected so that you can continue to use old

00:11:37RAM sticks to counteract this memory shortage and so on

00:11:42.

00:11:43And a large beneficiary of this is Nvidia, because they are the ones for those who do not

00:11:47produce independently, Google is an example of someone who is self-sufficient in chips in their data centers

00:11:52.

00:11:53They are basically the supplier and fill the data centers with their

00:11:58chips.

00:11:59And I don't even know if Elon’s space data centers are also using the H100, H200 and whatever else.

00:12:05I don't really know exactly, because now this is also interesting.

00:12:08It's a bit crazy, isn't it? I heard that Google will apparently pay 900 million dollars to SpaceX for server space every month in the future.

00:12:20That's quite a sum, isn't it?

00:12:22You first have to earn that back, right?

00:12:25Yes.

00:12:26Yes, and true to the motto, greetings to all listeners - you pay, help make that possible.

00:12:34Where I wanted to go, and this might sound a bit, let's say perhaps a bit like pub talk,

00:12:40but it was the thought that crossed my mind because there was that keynote and he talked a bit,

00:12:46then all that stuff basically shifted towards seeing the future in local AI.

00:12:52Because they've been serving the servers the whole time, then they presented the DJX Bark

00:12:56a small cube that you can place in your office somewhere,

00:13:01to run local AI models. And now they have come out with AI chips for Windows

00:13:06computers that are supposed to help execute models more efficiently. At first,

00:13:14I thought a bit about the Neural Engine that Apple has been integrating into

00:13:19their M-series architecture for years, on my iPhone, on my iPad, between Pro and Macs.

00:13:24And then I looked at it and felt a bit stirred inside,

00:13:30because on one hand, this felt very familiar to me, I haven't compared any specs yet, like

00:13:34how fast is this CPU that Nvidia is presenting or not.

00:13:38But then you stand there and the company seems to be breaking the gamer market a bit

00:13:43because graphics cards, as I said, are getting expensive, memory is getting more expensive, the

00:13:47data centers are upgrading with chips endlessly and now they come and say let's please solder such a nice chip onto all the PCs at people's homes.

00:13:57And now I have no idea, they can say this poorly prepared, I can fall back on the statement vacation at a time when graphics cards are getting more expensive, when memory prices are rising, when now an additional AI chip comes in, and you stand on stage beaming with joy showing ultra-slim notebooks on which games run.

00:14:16I felt a bit like, I don't know if this really engaged me or if it was just something being sold, like, oh, we can’t serve the whole data center guys right now, and who knows, here and there, and overall, and now we take, I don’t know, yeah, it’s just simple chips, sell them back to those we’ve been overcharging for the last 12 months and basically milking the next cash cow, so that was a bit of what went through my mind, but local model, I find that good again.

00:14:46Yeah, but I think it's a bit exciting, because of course you could also say that this is good marketing, a good marketing moment to pick up on these things, because

00:14:56of course we have in the discussion, when we look at the topic of AI, and we’ve also talked about the topic of Second Brain and such things, where are your data, where does the context you create actually lie?

00:15:06Should it really always be in the cloud? Should it always be with the model provider?

00:15:10Then it might be wise to consider what can run locally at my home,

00:15:15including my memory, plus maybe also the models that can then work with this memory

00:15:20can actually work with. Maybe they are just picking that up cleverly right now.

00:15:23Maybe, it’s like you describe, that the market for servers is actually just starting

00:15:27for years to come. We still have to produce everything that is actually

00:15:30being built around the world or is planned in space.

00:15:34We now have a niche here that we can cultivate again,

00:15:37which we have also really driven the price up for.

00:15:40Because a lot of storage capacities have now really flowed into these server loans

00:15:43and then for one or the other who might not be so familiar with the gaming market right now,

00:15:47to put it again... You can see it nicely, for example, with the handheld,

00:15:52with the gaming handheld from Steam.

00:15:54There is the Steam Deck, which is a small handheld,

00:15:57which is actually just a relatively good computer with which you can play the games on Steam.

00:16:01This has recently increased in price. I believe the large model was somehow around 690 euros or something.

00:16:09Now it’s, I think, around 890 euros. So 200-300 euros, that have simply become more expensive, those items.

00:16:15And that’s really noticeable. So it’s not because the displays have become more expensive, or that the buttons have become more expensive.

00:16:22It’s simply because the storage is urgently needed for the whole AI topic we’re having right now.

00:16:27And accordingly, this move is still contradictory, and I share your observation.

00:16:32We need to wait a bit to see how it really develops.

00:16:34Whether this is again an attack to say, come on, dear at-home users, save the expensive

00:16:42subscription fees with the model providers, get yourself a local computer that can also run the AI

00:16:48and can also play games, so you can also play.

00:16:50So, yes, we need to critically observe what this means, whether it really aligns with

00:16:54the good intention at Microsoft and Nvidia to say, we empower the user

00:17:00at home, so they can really work a lot locally. Yes, that's up for debate, one has

00:17:05probably still collected a lot of data from SharePoints, other things or something else

00:17:09that I then need to work locally. What I also found funny in this context

00:17:14but I want to apologize for that, as I said, I just got back from

00:17:17vacation, yes, there may still be many topics that are not fully researched yet,

00:17:22But now imagine that all the data centers are equipped, and now you start

00:17:25telling the users that they need fewer data centers.

00:17:28So that's also kind of totally funny.

00:17:32On the other hand, you have to honestly say, the penetration

00:17:35of AI, if you look at the statistics, how many percent use AI, as we know AI,

00:17:40how we pay for AI, how we use it upfront, saying very

00:17:44excessively, that's in comparison to the total population or to penetration

00:17:48with classical IT still very much less.

00:17:49And I believe that if you bring this topic more locally, then you also

00:17:57indirectly help in data centers, because not every nonsense passes by them anymore.

00:18:01comes, but rather gets certain things that take place locally, they will already

00:18:05just to follow up.

00:18:06Yes, the human body, the connected part is not everything, but there are also

00:18:09a lot of things happening on the way there, because the skin and I don't know who can

00:18:13act accordingly.

00:18:15I would like to briefly mention another rant.

00:18:19I have three rants today, yes, I've already unleashed the first one in the video, the second I want to unleash for Microsoft.

00:18:24They've also turned this around again, saying that their computers can produce models under Windows.

00:18:30And they also had Peter on stage, as if to say, we can also produce well under Windows.

00:18:35But then they showed something where I thought, guys, I understand that no company on stage can,

00:18:43speak the truth.

00:18:45Not that they lied, but their portrayal was strange.

00:18:48They showed a few products, AI products, one example was an intelligent

00:18:54badge, you know, like a company ID, right? If you work at the company that has the access control system

00:18:59with the company ID, then you have something like that, it has your picture and your

00:19:02name and your employee number and I don't know what else. And they showed that, so

00:19:06not as virtual, as a display that you hang around your neck and when you lift it,

00:19:10then you see what my agent wants from me with such a generative UI and have

00:19:14then also showed various other devices that have a camera on top of the head,

00:19:17and basically said, well, nowadays you don't want to use a smartphone

00:19:22for certain things, because that seems distancing to the person you're facing.

00:19:27So, it looks weird when you hold a phone in someone's face to examine them medically,

00:19:32but rather like a recorder, and I thought, well, whoever takes over Nokia,

00:19:37crashes the mobile strategy twice and sells something from Nokia

00:19:40that leaves something like a cat or whatever today, I would also say,

00:19:44I don't take a smartphone in my hand and instead build an intelligent batch that I hang around my neck.

00:19:49I found it funny at that point, because if you don't have a smartphone, you have to find a reason to present something else.

00:19:55But you noticed, both at Microsoft and at Google, and then these marginal waves often hear again,

00:20:02it was all in the sign of AI, everyone is telling you what is going on now and how can I translate here and operate multi-agent systems

00:20:11and build knowledge and now with the new chips, also locally.

00:20:16In the preparation, you also told me something about locally, that another player has entered

00:20:21the scene.

00:20:22So, you are the person who comes in today with the researched knowledge.

00:20:26Yes, that means...

00:20:27Crazy, right?

00:20:28Yes.

00:20:29Well, what does back mean?

00:20:30Well, you are the brain of both of us, you know that, right?

00:20:32Yes.

00:20:33I need to warm up again, okay?

00:20:34That's why you don’t belong here today, because I still have an open margin, it is

00:20:37quite fresh, it’s still dripping.

00:20:39But please tell what else we have to offer on the topic of local.

00:20:44Yes, I want to make two more remarks.

00:20:47As I said, I just briefly mentioned, whether we have rather the 900 million users.

00:20:51I find that already a relatively high number, of course always in comparison of course always

00:20:56even less than overall these systems are used, but what do we just have

00:21:00around 8 billion, 9 billion people in the world?

00:21:03That's already quite a good penetration, one must say.

00:21:05If you look at MPDG, there’s only one, he’s omitted a few, you’re

00:21:08not sure if it was intentional or a suggestion, an enclosure comes out after the motto, which city has won,

00:21:16but, excuse me. But on the other hand, you still have the issue, that’s with Microsoft

00:21:22said, not that they bring the Open Claw or Claws, as they call it, to the Windows world

00:21:29only, but to the Enterprise, that’s the exciting part, to say that they are so far,

00:21:32that they are actually now also releasing an OpenClaw-like agent in Enterprise systems,

00:21:40which they had previously been a bit hesitant about, so we’ve all been thinking a bit about that,

00:21:44that’s why we’re all building these secure environments for the OpenClaws and have all reasonably

00:21:49got a Mac Mini or something else to let it play around on, so we can say,

00:21:53okay, you should release such agents in the system where they can relatively freely and quickly also

00:21:58modify data, delete data, and the other things. I find it exciting how that then

00:22:01will work. NVIDIA has hinted at something like that with the sandbox stories,

00:22:05they have that it can run cleanly in enterprise situations. Then we must

00:22:08see how well that works. What you said about the third

00:22:12player that is still out there, he comes a bit from another sector. He

00:22:16has always been... He’s not at the same financial level as Microsoft and NVIDIA,

00:22:20is a bit smaller in comparison. Exactly, but he has still been causing a bit of ruckus in the AI space from the start,

00:22:26because for a while he was out there as the legitimate Google killer,

00:22:32I’m talking about Perplexity AI, which relatively quickly basically worked with their own

00:22:39chatbots and provided us with knowledge from their model's knowledge.

00:22:46where many naturally always say, oh, that's kind of silly, I can't really

00:22:50query current news, it’s always only as current as the model is basically trained,

00:22:54then it's already four months old and stuff like that.

00:22:56Did Perplexi basically have that from the very beginning, was it model-agnostic, because

00:23:00they simply use the models, offered a surface that mixed a bit of Google search

00:23:06with AI knowledge.

00:23:09That was just immediate ...

00:23:10That was today ... They have since evolved a lot, but back then it was

00:23:14really a mix of a bit of search and a bit of AI model together.

00:23:19Exactly.

00:23:20But from the sobriety perspective, of course, it was a super, super approach they had, and that's why they became relatively quickly well-known, but it was factually at that time.

00:23:29Simply the best combined setup you could make, while you still had to say or install a skill back then, not more like a web search or something like that.

00:23:39That was certainly a good way to do it, right?

00:23:41Yes, there was no Bing with Chipiti back then.

00:23:44Exactly, that was, yes, there were still things in between.

00:23:47You could activate the Bing search engine with a toggle switch,

00:23:51you know, that it then, in principle, is in your search-question search switch to Bing.

00:23:55We thought it was great, Jens, we thought it was great.

00:23:57Yes, of course.

00:23:58Today you stand there and think, hmm?

00:24:01Yes, yes.

00:24:02But as I said, that was already a turning point back then,

00:24:05that's a bit exaggerated,

00:24:07a good way to do it, and they have of course also evolved, and what they now offer is that they

00:24:13as one of the first players, I like to talk about this, because I find such things important, I say,

00:24:21I don't want to provide all my personal data, my personal habits, that I constantly have.

00:24:27Actually, one for Google, one for Microsoft, one for OpenAI, or whoever is out there, to have complete access to.

00:24:32That means I'm trying to build a bit of a system for myself that can distinguish,

00:24:38whether I need to use a cloud-based AI now due to speed or if I can use a

00:24:43locally installed AI, because the request can take a little longer or because there simply

00:24:48aren't that many parameters that can be loaded simultaneously, so a small local

00:24:52storage of 24GB can handle the models. This approach is now taken by Perfektität

00:24:58and that's why we're talking about it, because it fits nicely with this local aspect.

00:25:02They are also trying to implement a local force policy.

00:25:08This means that when you use a Perplexity computer in the future, when you use a Perplexity computer

00:25:14for tasks, it will first check what it can do locally and what actually needs to

00:25:19be handed over to the faster online models, and A is

00:25:24the racial decline and in my opinion, from the prophecy perspective, also a totally

00:25:28exciting concept.

00:25:29But this needs to be observed. It's just very fresh right now. I mentioned it. I could test it. I read it yesterday

00:25:36and one must engage with it. I think that's a good approach. So that something like this will come in the future, that such

00:25:42personal assistants might sit locally on your computer at home, which is also

00:25:47another interesting point. I mean, we will also have an episode

00:25:50on the topic of cybersecurity, looking at it from a different perspective.

00:25:53What can still be online? What really important data might need to be kept offline?

00:25:57That is also an interesting topic, how this market will develop and how

00:26:02the application for us home users will also develop and will definitely

00:26:05make us feel secure at home, because perhaps my data, um, maybe sometimes

00:26:09is less interesting than if it lies in such a huge database among itself

00:26:12with millions of other user data, so that no one makes the effort, like while

00:26:16I say this, I am a speaking thinker, I'm already thinking about how the

00:26:19effort is of course relatively low with AI, still personally at home

00:26:22to be attacked.

00:26:23But that we still, I believe, this is the right way.

00:26:26I think we should get a little used to having things locally in our

00:26:31bodies as devices or at home on the local server and not everything in

00:26:37The principle is actually out there with the large providers.

00:26:39I believe that's already a valid approach that we have.

00:26:42While you were talking about the attack and so on, it occurred to me, there was a report,

00:26:47What you see in a PDF with the correct fonts is something different than what

00:26:53a machine reads.

00:26:54They showed resumes in such a way, and we had already talked about so

00:26:59things like prompt injection and hidden characters and so on, and they somehow

00:27:02worked with fonts, and the font is essentially recognizable for the machine

00:27:08but not for the agent, and then they kind of created a sort of

00:27:14and showed that when you let AI evaluate contracts and things, you cannot

00:27:18trust that the AI reads it correctly. Before we perhaps want to talk a bit more locally

00:27:23I would like to briefly address what Apple has actually done due to its relevance.

00:27:28because they were still missing. And before I get into that, I would like to add a small side note, I

00:27:32of course want to say that Nokia does not just make phones, they make network equipment.

00:27:37So they are not completely out of the picture, but certainly a different company

00:27:42than in my youth, where a Nokia phone had, let's say,

00:27:47a certain prestige and after the Microsoft takeover, it did not anymore. But well,

00:27:52let's go back to Apple. Because they actually chose a

00:27:56different approach. They somehow interpret it like this.

00:28:01At the time of recording, there was only the keynote of their annual

00:28:04developer conference, not that anything else is coming up in two days.

00:28:07comes out and we didn't know it. But at this keynote, they

00:28:11shown something that, if you were to ask my neutral third parties now, they would say, okay, we have

00:28:17always had these Apple devices somehow with the processors, I have already talked about, local AI models

00:28:22supported and always had a somewhat slow-witted Siri in the mix, who rather says, I

00:28:28couldn't find something on the internet than being able to help you, and they have now expanded that with Siri AI

00:28:32and are moving more towards personal experience. Yes, so the phone has

00:28:40the context of a user, the phone knows basically who you are, what you do, what you have in

00:28:45your photos, notes, reminders, emails, files, and you have an interaction opportunity

00:28:52with this data.

00:28:53At first glance, that sounds a bit like, oh cool, I have

00:28:57Open Claw or Hermes Agent or something cool, when you first start engaging with the EYE-world

00:29:03on my devices, because they offer it on their entire

00:29:07platform, from the Apple Watch to the Vision Pro, so the VRR April from Apple,

00:29:13about their Mac, the notebook, all the way to tablets, iPad and iPhone, the phone. And then I saw

00:29:21this keynote, that's why I say, it looks really cool now and I thought, cool, two

00:29:26and a half years ago, I would have been running through the apartment with joy about what I would

00:29:32get in terms of features, and I felt like getting there in the woods and at this point I say

00:29:41for the last time, I had vacation, I'm not quite in the flow of speaking yet. I found,

00:29:45that what was presented in terms of functionality I found subterranean. Now, however, I have

00:29:50already looked for my way out because I said, maybe something will come out in two days.

00:29:54The developer conference has just started and the keynote is really for the L-consumer.

00:29:59So actually for those, I mentioned earlier, 98 percent of my right have never really dealt with AI at all, for these people.

00:30:07They have now basically offered something? And it could be that this is good.

00:30:12But if I come from a perspective and say, I have seen a Hermes-algae, I have also seen a chlorine.

00:30:17I have seen what Nvidia, what Google, what Microsoft are coming up with, and I feel that with a product from Apple,

00:30:26with the knowledge from this keynote, it's not even close to the top, I would like

00:30:32to say it's not a supply chain, but at the top of IT, regarding the motto of simplicity,

00:30:37agility, convenience. Privacy, absolutely top-notch. They have their whole principles according to the motto,

00:30:43nothing's wrong with the service, because everything's fine. But I have neither image generators,

00:30:48the ones that were shown, where I thought, cool, that looked more like Mid Journey

00:30:52eight months ago or not even like Nano-Banana, although they bought the Google models,

00:30:58was decentralized. I have a response behavior. I mean, Apple was the only one who also did a

00:31:04video production. Not live, like all the others. Everyone else did live presentations

00:31:08. Apple had a Google presentation, exactly, they did, but they also

00:31:13mentioned Google in the cinema, recorded it. And it just didn't work. I was in no way

00:31:19A second feature, namely the ability to change perspectives in photos retrospectively. That was nicely done, how they implemented that in the DSUI.

00:31:26I was not at all surprised and thought, okay, this is now a platform where I feel so at home that it is worthwhile,

00:31:35not to rebuild everything myself and think about how to connect things and install Open-Claw and so on, but it feels like just a year or two ago and then you just stop hand-holding.

00:31:46Now we both know how quickly things have changed.

00:31:50That means until the operating system leaves the beta phase.

00:31:55It's no secret to say that it's probably September,

00:31:58because it's always been September in recent years,

00:32:00then the distance is already greater than today.

00:32:05And that really made me a little scared.

00:32:08And my only hope now is that during this developer conference,

00:32:11you can still follow some of the videos

00:32:13and see what the developers come up with,

00:32:15maybe already lay a bit of the foundation for building some cool stuff.

00:32:18But what they showed does not engage me and the people I chatted with and

00:32:25not everyone is crazy about AI, that is not what stands out.

00:32:29We found it as, and Tim Cook has now also left the company, so what

00:32:33leaving means, stepping down from his position, we found it to be the most uninspiring

00:32:38keynote ever.

00:32:39So, my side rant is finished.

00:32:42Okay.

00:32:43He saves the situation. Dear listeners, yes, dear listeners. You are just listening to a live report from the Apple fan club.

00:32:52Yes, that might be well known. I do have the reputation of maintaining a certain fan culture there.

00:33:02Yes, there are people who have an issue with the original signatures from Steve or...

00:33:07It should exist, it should exist. Yes, yes, but exactly, it should exist.

00:33:13I actually believe, as I said, first of all, I think it's an amazing moment for everyone who knows you,

00:33:18with what you just presented,

00:33:22this absolute openness in this direction, this Riemer that you really like

00:33:26and that still produces very good products.

00:33:28But of course, surprisingly, I agree with them in the AI age, so far, they are still lagging behind.

00:33:33And that, even though they have an incredibly good starting position,

00:33:37with all the devices they have out there, with the ecosystem they have built,

00:33:40with the brands they possess, where there is a high level of trust in high quality and such things.

00:33:44So, what we always say, where you say, an important question in the future will be trust.

00:33:49Whom do I trust with my data, who do I trust that the workflows I build will actually work well.

00:33:54And considering the quality standards that Apple still applies when reviewing apps and such,

00:33:59we also occasionally release an app and sometimes experience that during reviews

00:34:03they might not accept small matters, it is surprising that there is a loss of both quality-

00:34:09and also some sort of inspiration. On the other hand, I must say,

00:34:13sometimes it might not be so bad to be a second mover either,

00:34:16I wouldn't talk them to death just yet. I don't believe it's the Nokia moment yet.

00:34:20Eppel is not here, right? Exactly, but let's see what happens because I would like to mention another contribution that I saw, which I can also highly recommend, we can share that too.

00:34:30In the show notes, it was from Benedict Evans, I believe that's his name, he talked a bit about it.

00:34:36So without going down the whole AI rabbit hole, he says that, of course, we are still in a situation similar to 1997 with many of the things we see right now.

00:34:45That means we are currently facing this state, so the internet is there, but it is still

00:34:50not a real mass market.

00:34:51The big players are actually not done yet, the big usability rules are

00:34:57actually still not in place, how things are supposed to work, and that’s precisely what we

00:35:01are observing right now.

00:35:02We discussed earlier that we said it has now

00:35:05for example, how we integrated Bing Search back then into Chatshipity, and how great

00:35:11that was, and how it was celebrated, it felt like it was there for two weeks, then

00:35:13gone again, because other AI did it better and just

00:35:17was natively integrated and one simply had to say if I wanted to

00:35:20search for something current, then the AI would automatically look for

00:35:22current things as well. So that means, many things, both from

00:35:26the manufacturer's side and from the customer's side, are still not fully developed.

00:35:30And that was a bit exciting about this Billendict Evans topic, so also

00:35:33with all the hype we have, the speed at which all

00:35:37many things are currently emerging and one often has the feeling that one is

00:35:41not keeping up at all, yes, but maybe in some areas that’s not even bad, because some

00:35:46things will first develop properly until they actually work. And

00:35:49sometimes it doesn’t necessarily always have to be an Apple that is also

00:35:53the first mover on many topics, because there is tremendous timing involved with many things,

00:35:56maybe they are also too early now, so they are content with the situation and say, okay,

00:36:02maybe we are also a good second mover that simply observes the market.

00:36:05And then, when it is really ready for the mass market, we will truly jump in, because we

00:36:10We know the topic, many of the AI workflows and other things, many things work well,

00:36:17many things work fantastically and a thousand times faster than one could do in the past.

00:36:21There are still many places where it gets stuck. When I now set up a complicated workflow,

00:36:25with several agents, different AI models, we just talked about the topic of cities,

00:36:29there are always intertwined methods involved, the model doesn't always produce the same result,

00:36:34no matter how small, sometimes one tweaks the task.

00:36:37So there are still many things that will just evolve, where maybe for the

00:36:42average user, who occasionally maybe makes a request, it isn't so drastic that he hasn't somehow set up the

00:36:47Second Brain and his second Hermes agent, who constantly researches various things, because it isn't necessary yet.

00:36:54And it's not yet so usable that someone who maybe doesn't have as much time, like both of us sometimes, can even take care of these topics,

00:37:01For everyone, in principle, it should be this effortless effort, so that one can

00:37:07engage with something like that.

00:37:08And maybe Apple is simply waiting for that moment to say something.

00:37:11First of all, thanks for bringing it back, I mentioned it again even though you didn't want me to.

00:37:16You have brought a kind of professionalism back into your statements.

00:37:20Maybe it's also a bit of that disappointment, because the hope was connected

00:37:24that now, with the deal with Google, there are good models and that something

00:37:29would happen.

00:37:30And then, as I said, you look at an Android phone with one eye and then look at what was introduced with the iPhone.

00:37:36Then you sometimes feel like you're not quite at the forefront.

00:37:39And so at times you don't feel at the front. Period. I don't want to revise what I've said again, I meant that too seriously.

00:37:45And perhaps it is worded that way, and maybe it really is as you say, based on what the operating system offers as a foundation.

00:37:53In other words, we have already promised so many times; two years ago they promised almost exactly the same thing they are showing now, but they did not deliver. They had promotional videos for features that didn't even exist as an alpha, meaning alpha as a version status, meaning that no developer had even seen it or that anyone had even written a time approach document.

00:38:12And now they're basically going through the third time and maybe they're taking it slow, and yes, it could be that they don't come, or however it may be.

00:38:20Some things don't quite align with the EU for me because, oh, you have to sort a few things out with the EU, where you then think,

00:38:26you are supposedly bigger in terms of data protection and use models that exist in the EU, namely Gemini, find the mistake, where could this possibly be coming from.

00:38:37completely incomprehensible to me, but also framed a bit, something I didn't want to do.

00:38:42On the other hand, I hope that in the next few days, my

00:38:46mood improves in that regard, that we finally have some developers available, so you can say,

00:38:50okay, there's a foundation, you can work with that, you can build on that,

00:38:54because a foundation also signals a bit to you, okay, consistency, stability in a

00:38:59world where tomorrow if someone sneezes, you have a new feature hanging on your neck,

00:39:04Yeah, I recently, after I saw your Openclaw story with Hermes Agent,

00:39:08got something from Stanford, OpenJavis, and then something new came today

00:39:13pushed through the various ethers.

00:39:15In time, you can understand if it would be good to set a foundation.

00:39:18But we also wanted to talk a bit, when we talk about the foundation, about a bit

00:39:22more about Lokalkai, but not about models, so I mean not about

00:39:28models, because there is already a vibrant market here, Google has also just

00:39:32released a local model again. It's not that bad. It somehow understands images.

00:39:36The 4er, 4er, 12B or something like that. Here it’s called 12B, I believe. The 12B is for 12 trillion

00:39:43parameters. Those are numbers again. I always shut that off. But how

00:39:47is it despite that? I installed it over the weekend and my entire other local

00:39:51queries have not yet all been replaced in my local installation. I directly

00:39:55discussed with Claude whether it’s smart to replace it

00:39:59Klotz wanted to first test whether it works. The first tests have gone well.

00:40:03That means, I believe 6 or 7 of my local small agents that are running are waiting

00:40:10still on exchanging. 2 or 3 have already been exchanged and are running with the new model. That's really

00:40:15good, isn't it? Because, as I said, it hopefully enables easy work with

00:40:19Speech locally soon, as it makes things clearly recognizable and usable in text.

00:40:24And simply, it is significantly faster than the models I had before.

00:40:28This means I can already cover more topics that I always sent to a Cloud

00:40:34Opus 47 or whoever online, for quick reasoning to

00:40:39have a quick discussion with me about topics that hopefully run

00:40:43in real-time.

00:40:44Now I can shift more from the publisher to local, to the local computer

00:40:49and I don't need to rely on the Cloud at all.

00:40:51And the point I wanted to make, because I said, I never wanted to

00:40:57state whether the models had local data storage. Now one can say, look, the person has

00:41:02hard drive space, yes, but as anyone who follows me on LinkedIn has seen, I

00:41:06recently announced a pretty empty dashboard, namely a skill that collects

00:41:11news and sorts them according to how relevant they are to you, relevant means how quickly

00:41:16it is usable, how production-ready it is. I then published the skill for Apple

00:41:21News, thus linking it back to Apple, but I built it for myself

00:41:27for AI News because I said there is so much happening in the market. And if we talk about it

00:41:32and I want to be better prepared than after the vacation, I also want to

00:41:36know how things have developed historically, what has come out new, where have

00:41:40directions changed. This means I don't want to just grab current events

00:41:43I want to understand the development of knowledge. Or how things are interconnected.

00:41:48We just talked about Microsoft and Nvidia, that they are working with local processors, it's

00:41:52not hard when they announce a partnership. But there are also other topics,

00:41:56that have overlaps, that go in a common direction, that think through certain things,

00:42:01whether it's because they appear in the news, or also because there are studies,

00:42:07like scientific papers where certain things are envisioned or proven

00:42:13under some unknown context. And so I actually took what I, as I said,

00:42:17published small pieces of on GitHub and LinkedIn, and created a larger than usual

00:42:22job, that then dissects right portals, scientific papers, YouTube channels, podcasts, etc. Zerfrä is

00:42:29broken down, a component part, which also maps out a radar for what relevance means,

00:42:34what affiliation means, so I can consume it as a person. But also

00:42:38as some who know me will chuckle at, and I don't want to reveal why, as Obsidian

00:42:45stores Vault and gives me the opportunity to understand it both humanly, but on the other hand also again through the

00:42:54Vault connection of Obsidian

00:42:56my

00:42:58accounts to make available, so that if I want to build something, if I want to

00:43:03research something, if I now want to prepare it or program it or whatever, that it's not just on the internet, so it's good,

00:43:09not only does it work with Internet search, I have now also installed Perplexity as an MCP, yes it also doesn't just do that and processes with Notebook LM

00:43:16also with MCP, but also accesses these local planes and I must say,

00:43:23that this plane, which then comes from all this news can somehow say who gets fed, gives me

00:43:29valuable information,

00:43:31both in the agency environment, in the Apple environment, but I have also set something up for

00:43:39everything I have privately saved, like a mail inbox, I let it search through,

00:43:43I have my notes in Apple Mail, I have 19 gigabytes of emails, 3.9 gigabytes of Apple notes.

00:43:50thrown in there and said, do something with it, I have no idea what you have in mind

00:43:53and then I made a publication about medicine and family and what do I know

00:43:58and pieced it together, and before I pass it back to you so that

00:44:03I can get rid of my speech diarrhea, which I'm slowly getting back into, it got me

00:44:08at some point on the idea to create such an Identity-MD case of mine, that

00:44:12I think, if anyone is interested, I've got that on GIT, I set up my imprint page,

00:44:16you can also find the identity in the repository, and I now have it referenced with Claude MD and all

00:44:21agents and said, check that out, and why am I telling this

00:44:25now?

00:44:26Well, after I've created this entire knowledge tree, I asked one

00:44:28what do you actually write in an identity and then looked at your

00:44:31whole studies and so on to see what interests an agent.

00:44:34And then I went and said, now please check email and

00:44:37notes. Who is the market? What interests the market? What challenges did

00:44:42he have? So it doesn't say, what I mean, the one

00:44:46wait at the right rear, but there are things that meet the

00:44:49standard of professionalism. I threw all of that into this identity

00:44:53And it's funny to see how since then, whether it's Claude, Gemini, or Mai, behaves differently with you.

00:45:02Yes, that's a super effect. That's exactly what Klappärvier also described with this Second Brain.

00:45:09When you start building something like that, and we'll mention shortly that there's a bracket coming up to add a few things,

00:45:15Then you have the advantage that no matter what model you use in the future, if you can connect it in some way, whether it's local or even online in the cloud, it doesn't need basically this whole topic that we would have called prompt engineering, context engineering, or whatever else before, but it's just there.

00:45:37The context is there, because this machine understands much, much better what you've done.

00:45:41Can evaluate old projects of yours, how you approached projects, what is important to you in

00:45:45projects, how you might want to develop, how you want to write articles.

00:45:50All of that can be stored there and doesn't have to be done manually when you ever

00:45:56move from a Chechipiti to a Gemini or a Cloud, it has to be ported

00:46:01by you.

00:46:02The providers offer that as well, we talked about it before, nowadays such export functionalities,

00:46:07others can be Memory Files which can be exported, but they only have a certain length.

00:46:11So what you're describing now, a real vault in that sense, where a lot can be stored,

00:46:17that exceeds those limits that these Memory Files have from providers, to really make it significantly easier,

00:46:24I don't want to mention numbers, significantly, what that means, it obviously offers even more possibilities,

00:46:28to really generate results with contextual relevance.

00:46:31What is interesting is the topic, and that's why we have to go into detail in an episode, how to build such things, because this identity-file is of course good if you have now laid it out publicly.

00:46:44You still should be careful to look inside to see what's in there. There shouldn't be too many private data.

00:46:49That's why I mentioned it.

00:46:50That's why I mentioned it.

00:46:51Accordingly, it's also the case, for example, that's why I build guardians for myself who watch over,

00:47:00what is actually being shared from the vault? Then Open Cloud, which can also run out publicly

00:47:05and can use quick models outside. What information goes out of the

00:47:10vault at all? So, such things like in an email could also contain

00:47:14your e-mail or other data. Do I really want to say that for a sample-ERI,

00:47:20when I'm just looking for the coolest video movies or for travel tips, have you seen

00:47:25super, that's great. That doesn't have to be.

00:47:28And I think that's just such an important topic, that one has this sovereignty over their data more in the future.

00:47:33I have to keep.

00:47:34And this goes hand in hand with the topic, yes, that also includes that much of it might

00:47:39be better off locally than in some online storage, where you sometimes are not exactly

00:47:43sure what is happening there.

00:47:45Of course under EU data protection and so on, that is all secured.

00:47:48Yes, but still personally it sometimes gives me a better feeling when I

00:47:50have things a little bit under control, when I can manage them

00:47:53to that extent.

00:47:54I'm trying to work on this home installation that I have.

00:47:58Which data do I exactly share, so that later I can interact with the design AI, with the production AI

00:48:05or with whatever I then interact with or my agents interact with,

00:48:10actually just the minimum knowledge is necessary, but also as much knowledge as is needed,

00:48:15so that it can generate perfect content.

00:48:17And keeping that balance in your head can be challenging before it becomes difficult,

00:48:22It's just good to build an agentic network that essentially orchestrates everything smoothly at home.

00:48:27I just want to, and with that I want to do justice to my humility, present the best,

00:48:32that my home computer has 256 gigabytes of RAM,

00:48:37and accordingly, it was powerful enough with a corresponding CPU from Apple

00:48:42to run local models.

00:48:46I’m saying that now I asked Croc to give me a personal life recommendation.

00:48:51to give. Yes, just because you wanted to know about the private models and data protection

00:48:56I didn’t want to say, hey, who wants my 19 gigabyte file, go to www.geilegedanken

00:49:02of an old white man from Karlsruhe.de available for download. But I was actually

00:49:09surprised by the fact that you give it Markdowns, which are inherently quite large. So what

00:49:14does inherently mean? If you basically consider, I mean 19 gigabytes, not everything will be saved

00:49:19there. It's also clear, because it vaporizes and takes with it the essence of the information and stuff into its

00:49:24knowledge tree. But it is

00:49:29still quite astonishing which correlations it is able to establish.

00:49:37And what it reminds you of at times, which might not be that familiar to you at first

00:49:43or is perhaps not as well-known anymore, in order to do justice to the personality

00:49:48and also no further details here. Just an overview. I mentioned local models.

00:49:53I switched households. So, I lived in Karlsruhe for many years and moved,

00:49:58switched households. I’m sitting there at the household meeting and he asks a few questions,

00:50:02Answer a couple of whole questions. Do you have diabetes and such, can you find an answer? And then it came like this,

00:50:07yes, what were the blood values like last time when they were at the big oat round trip at the

00:50:10urologist? Yes, I have them somewhere, I noted them back then in my notes

00:50:14but I didn’t have them with me that day, long story short, I was sent home with a few questions

00:50:18to answer.

00:50:19And then I had my vacation and when I came back, I immediately thought of the day doctor

00:50:24who still wanted the answers and I asked the knowledge mirror what he actually

00:50:27thought about it and then I delivered that almost within a fraction of a second

00:50:32back.

00:50:33Not because he pieced it together in chat, but because he pieced it together

00:50:37and could also link the source accurately and of course also the

00:50:42most current data, temporal, the assignment, so not just here was a blood value, but that is

00:50:47really the last one. And that's really nice and great when you realize, damn

00:50:52wow, I have my data, I have a context for the data, have roughly understood what I want,

00:50:57prepared it a bit for me, so that I can print it out into the documents for him

00:51:04and stuff like that, and you stand there thinking, damn, this will work out,

00:51:08Yes, this will work out and doesn’t start again and just talks to you like,

00:51:13but that’s a fascinating idea that I can ask for confirmation now.

00:51:16Yes, I will do that for you right away and you’re the greatest.

00:51:19He has stopped doing that because it says in my thing,

00:51:22please speak clearly and concisely and caveman-like and so on with me?

00:51:26Yes, yes, yes, that’s so cool.

00:51:27Yes, and these are also things where you notice,

00:51:30okay, does his behavior end or does his behavior end,

00:51:34because you simply provide him more context from your side.

00:51:37Yes, I think that, as I said, I've already mentioned this one or two times,

00:51:41because the things tell me too when it gets late, that you should know it's good now or something like that,

00:51:45that you should go to bed now and that it has been enough with the things we do.

00:51:49Sometimes I feel, I believe, out of my memory,

00:51:51that I’m somehow a slacker who gets exhausted after half an hour,

00:51:55so I don't know exactly, but I still find it quite okay,

00:51:58that I’m being reminded that it’s late now.

00:52:01Because that’s actually another topic where we might have to,

00:52:04we need to see if we can get some health expert to talk about it in the coming weeks

00:52:08became more and more prominent and rightfully so.

00:52:10The topic that with AI, you also get into a bit of a new pressure situation.

00:52:18Because you can...

00:52:19I thought productively...

00:52:20Yes, maybe that too.

00:52:22I know.

00:52:23So it is indeed true that people say, if I can quickly get a lot of things done,

00:52:27now with productivity...

00:52:28We've already talked about that.

00:52:29So it's a flow topic.

00:52:30So that is the case.

00:52:31And I have left a few times now also saw that more people were writing about it.

00:52:38Maybe we should see if we can have someone from the health sector as a guest on it.

00:52:43That was the topic of my episode, because I'm not so well-versed in that area.

00:52:47Because I can understand that there are indeed certain overload stimuli present,

00:52:52because many topics are evolving much faster, the pressure is increasing,

00:52:56everyone outside is saying, yes, your jobs are changing, and so on.

00:53:00We're partly doing that as well. I think it's also a value to consider. But still, it is

00:53:05overall just a, I believe, a smart idea to have such a personal assistant who

00:53:09is a bit more knowledgeable about you and set it up in a way that it is not

00:53:13on the big servers outside, but with you, so that you can simply

00:53:18port it. So not having to worry, as I said, in the EU, it's all compliant with data protection laws

00:53:21anyway. The models are not trained with your data either. So you can...

00:53:24indeed have things lying around. But when you port the things then

00:53:28wants it, and in the size you described, then it is also like this

00:53:32a bit difficult at some point. Then it's smarter if you have that on local

00:53:35computers, because the gigabytes of your email traffic and other views that

00:53:40are also not fun when you have to deal with them every time, moving from one email service to another

00:53:44you would like to take.

00:53:45I don't want to seek a forced conclusion of the episode right now regarding the car,

00:53:51but I would definitely like to share a couple of thoughts before we come to an end,

00:53:55two or three thoughts that you've really prompted in me, which are on one hand related to the vacation

00:53:59and on the other hand related to these, what is now the insight from this

00:54:03world and local story that's come out for me. What I already noticed during the vacation, now also

00:54:09when I think of my Ibo-Kreter, where I said I didn't want to go into the technical

00:54:14details deeply, I enabled myself to do a bit more than just present books,

00:54:18but I still noticed how good it does you, and one has to reflect on one's own

00:54:24shortcomings, not having the access to information that you usually have in everyday life, and

00:54:31before you mindlessly scroll through TikTok, doom-scrolling, and you read emails and immediately respond,

00:54:37yes, when you receive an email, it's like these days, I have written, why don't you respond

00:54:41and then an email is, so that one already feels overwhelmed and the whole thing

00:54:44has become much worse with AI, because by writing a sentence you can

00:54:49You have topics to solve that used to take you weeks, and now there’s just that one

00:54:54sentence coming.

00:54:55I'll still write that down.

00:54:56Yes, and then you felt like you were in a forced situation and I definitely noticed that during the vacation

00:55:00after two or three days you don't even have it on your mind anymore, like

00:55:05the motto, I write a sentence and then things happen, but you also mentally

00:55:10notice that something relaxes, something you actually didn't feel under any kind of tension in the heat of the moment,

00:55:15At least I didn't have the feeling before that I was under any kind of stress.

00:55:20The second thing that came up for me was, when I talked about the vault now and then, I just said,

00:55:27I came back, I could ask questions, I realized, and here we are again with your entire

00:55:33UI experience, how I interact with systems. I noticed that the topic of language is becoming

00:55:40increasingly important for me personally. So, I talk to the system. I don't want to have Telegram chats,

00:55:45I don't want to open any weird super apps from some suppliers

00:55:48and type away like I'm at the World Championship. Actually, I want a Jarvis that, me,

00:55:53that I am talking to and who prepares this for me, who presents it to me visually, certainly,

00:55:57but who is not dependent on the slowness of my fingers and the mouse movement,

00:56:02but to whom I can say, I would like something, and he does it. Maybe there are other concepts

00:56:08And maybe we are sort of in the age of MS-DOS with AI and are still waiting for the Windows moment, or whatever, but I have personally noticed that when I create more things in the future, I will try to optimize these things for voice, to make them voice-operated, voice-interactable, kind of like how do I figure out that the market is about conversation, voice identification, mood recognition.

00:56:38in the sense that if he sounds annoyed, it is an additional context information like

00:56:42if he sounds happy, or weepy, or I don’t know what, not that it matters to me

00:56:47now happens daily, but maybe I'll still consider that for a few

00:56:50wanting to think more about it, because I somehow believe that the more you engage with such things

00:56:55like knowledge trees, Markdown files, moving away from Word, Excel, PowerPoint, towards knowledge files,

00:57:01towards text files, where it's not the formatting that conveys the information, but

00:57:07really the context and the content. With Excel, it's like this: please look at the red column.

00:57:12What kind of red column is that? Well, I mean, if you have to color the column red, then

00:57:19I have another problem. And right now in this time, maybe voice is personally just right for me

00:57:23the new hot shit. So I’m always coming from an experience perspective, a design perspective,

00:57:29an interface perspective, of course always saying it should always be contextual, so voice is

00:57:33in many situations a good thing, but there are also situations where when you’re

00:57:36sitting on the train, you might not necessarily want anything, so it needs to be adjusted

00:57:39and it should also know the local AI model in which context you are moving, and then

00:57:43adjust the UI accordingly, but that is definitely an extra episode we can talk about

00:57:48these things. A big office is also great when you can then talk to the computers at home,

00:57:52but it’s still totally right. This is somehow what you just

00:57:56described, it does trigger me, and we should really

00:57:59do an episode again about Excel and such things and PowerPoint, that is obviously

00:58:03essentially nothing different than the digitization of topics that used to exist in the analog world

00:58:08worked. We maintained such tables or wrote letters or something.

00:58:14Exactly. And this is so, I believe, yes, this whole concept will be replaced by what,

00:58:22we are currently experiencing. And I am back again to one, to Benedikt,

00:58:27whom I already mentioned earlier, and then to Benedikt, what is his name with

00:58:30Last names, wedding events. We have this moment from '97. Many things are still

00:58:36not defined and they will still define themselves. We'll see where this future goes and I

00:58:41can also make a connection. Also Microsoft, as we mentioned, at the cinema of

00:58:46Microsoft, this project called Solara, I believe, was mentioned, which is also about

00:58:50batches that include voice input. They are also bringing some devices. There were

00:58:54just some pictures I had seen, a bit like Amazon Alexa speakers or

00:58:58or something like that, but this AI also comes out of the smartphone, out of the computer.

00:59:06You presented that nicely, while I was grumbling about it, just because you

00:59:10don't have smartphones, you've now invented new classes of devices, you're now

00:59:13handling it professionally and saying, it also democratizes the hardware, I'm glad that it

00:59:18was briefly closed here, again Defend Feedback is coming, thank you,

00:59:23thank you.

00:59:24We are now with AI not only democratizing technology, that enables more people to do things

00:59:32that they wouldn't have been able to do before, but also the decoupling of components,

00:59:38because through AI, and now we're back to language and various interaction elements and who knows what,

00:59:43the end device is currently completely secondary.

00:59:46It doesn't have to show a PowerPoint anymore, it has to offer you entertainment,

00:59:50it has to interact with you, but that could theoretically be the monitor on your

00:59:54refrigerator just as well as in your car. Your U-Wear-on-the-Wrist or the Smart Ring

00:59:59you speak into, theoretically, all kinds of cases are possible, whatever

01:00:03interaction element is correctly there because people sometimes like that

01:00:09look, see. Why do people get the new iPhone, so that you can see you

01:00:14have the new iPhone, that also wants to be approved. But I believe,

01:00:18technology can fade more into the background through AI because it’s simply there, it

01:00:24becomes more aware of your context and supports you more, where you need it, how you need it.

01:00:31Exactly. And I think that's at least the best professional closing remark,

01:00:36that I can bring today. Yes, then I'll just make a side jab, saying,

01:00:41where it is of course actually not a terminal input, so that's just what it was

01:00:45all about with all the Open-Claw enthusiasts. And Developers are currently totally into Boog, still

01:00:50already talked about that. Of course, you feel comfortable where you are familiar.

01:00:53And accordingly, many applications are still on a terminal level and work

01:00:57best when I pack it together as a programmer. However, that is not marketable to the masses.

01:01:02So, in order to be able to access things and to democratize this topic, that

01:01:07many can also independently develop applications, can access many things

01:01:11just like we experienced with the Internet back then. This

01:01:14is something we will continue to experience with AI. AI can do that. AI will be multimodal, it will at that

01:01:20moment display something on your wall displays when you need it, show something on your table,

01:01:26whisper in your ear, see what you are currently seeing, because in your headphones or in your

01:01:32glasses, there are built-in cameras that will then provide you with the

01:01:36right solutions at the right moment. And that will become a fascinating, fascinating topic.

01:01:40That's exciting in terms of legal protection, accordingly exciting, that we will stay local regarding things

01:01:44and why, as I know my third colleague Mark, he actually has one more closing statement.

01:01:49Yes, indeed, because you just brought me an analogy from the past.

01:01:53I just mentioned that we are with AI

01:01:57just in the age of MS-DOS, before the powerful use of any Windows

01:02:01came, and we know what comes with AI. And I just thought, while you were speaking,

01:02:05are we with the knowledge of AI that we have today. In 20, 30

01:02:11years, the COBOL programmer, just like you were looking for the COBOL programmer back then for the Year 2000 problem,

01:02:18who had to solve some very old Stone Age code for Year 2000 issues, back when

01:02:23it flipped to zero and there was some fear that technically the world would collapse,

01:02:27and COBOL, a very old programming language, perhaps we will be those old men,

01:02:33have you used this Open Claw a bit earlier? We need

01:02:40or some expert to solve this issue, because it is basically so stone-like

01:02:45it is.

01:02:46And I find it frightening how fast-paced it is, because it won’t take

01:02:5030, 40 years for something to become outdated, maybe just two years.

01:02:54But still, I wanted to thank you for enduring this episode with

01:02:59me today, and I would say we’re wrapping this up. If you enjoyed it,

01:03:05even though it was a bit longer again today, we would of course

01:03:09be very happy to hear your feedback, especially if you are in the health addiction or some

01:03:13other sector and want to discuss how AI affects our behavior

01:03:17.

01:03:18Feel free to reach out.

01:03:19You know we don’t bite.

01:03:20We might.

01:03:21We wouldn’t hurt you either, we’re just happy to have any guest.

01:03:25For those who say, here we go again without a guest, let it be said that we have

01:03:28guests coming up soon.

01:03:29So the guests are lined up for us, it's just a queue, it’s totally

01:03:31nice.

01:03:32And yes, Jens, thanks for the time and tune in again next time,

01:03:37when it’s again time to think AI, think different.

01:03:40Bye.

01:03:41Bye, bye.

01:03:42Welcome to Think Different, Think AI, the podcast by Mark and Jens.

01:03:50Two technology-loving minds who not only talk about artificial intelligence,

01:03:55but live it.

01:03:56Here you'll find clear classifications, real practical insights, and a fresh perspective on what

01:04:02is possible.

01:04:03Understandable, critical, and always with a wink.

01:04:06AI to think about, to smile at, and above all to discuss.