Hat eine KI eigentlich Urlaub?

Transcript

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

00:00:07Two technology-loving minds who don’t just talk about artificial intelligence, but live it.

00:00:14Here you’ll find clear classifications, real practical insights, and a fresh look at what’s possible.

00:00:20Understandably, critically, and always with a wink.

00:00:24And today again in a new configuration, because I’m sitting very comfortably at a nice

00:00:47desk with a cup of coffee in hand and Jens, you are on the road in your car.

00:00:54Yes, I am on the road in my car. I hope the technology transmits this well, so that we

00:00:59have good sound afterwards. I have full confidence in my technically very very

00:01:04esteemed colleague. Whoever likes can do that already. By the way, we also sometimes

00:01:08talk about it. We occasionally do an episode where we talk about

00:01:12what our AI-supported workflow is to record and broadcast all these high-quality,

00:01:19mega podcast episodes. We should definitely

00:01:24do that. You can probably also do that out there in one or another way.

00:01:27How much easier it has also become through AI, but that’s not the topic for today.

00:01:31No, today it also indirectly relates to work, namely to the absence of work

00:01:37not because I have nothing to do, but because I am on vacation, namely with the question,

00:01:41What is it actually like? Does an AI agent also have vacation?

00:01:46Hm, yes, it’s an interesting question, definitely an interesting question.

00:01:51So last time we talked about this for those who are listening for the first time,

00:01:55Last time, we defined a bit what a AI agent actually is,

00:01:59what a normal bot is, and so on.

00:02:01I would, let me add our two or three sentences on this, so we can classify it correctly.

00:02:05An AI agent for us is not just somewhere a chatbot that can give you smart answers,

00:02:11as we know it in principle, but it actually is that the AI agent, based on a task,

00:02:18which you may have received from yourself, or from someone else, or from a system,

00:02:22then independently searches for the necessary solutions to carry out this task.

00:02:28That means, I’m not giving any specific instructions in that direction, just saying, go to this website and so on,

00:02:33but I just say, I want this done, and then the agent gets to work and completes this task.

00:02:39No matter how you look at it, this is our, this is the common definition of agents, I would say, and also ours.

00:02:46Exactly, and that outlines the tool set that it needs for this.

00:02:50for example, independently put together, which are also other agents or systems to achieve this goal

00:02:56and then does everything in his, I wanted to say, humanly possible, but in this case not necessarily human,

00:03:02possible, so that this target state I describe, yes, we also had the topic last time, I have a barbecue and don't need 20 kilos of meat, but I need a corresponding recipe with materials and ordering options and to ensure that it arrives on time, someone needs to take care of that, and when you just said, what an agent is in essence, or what is understood by that, it becomes clear that someone like that has a title or

00:03:32also clear that this whole thing is of course not only useful for ordering 20 kilos of meat

00:03:38or for researching when the train is or which road connection I should take to get there best

00:03:44Hamburg or where fixed which city? Where can you say that? To Cologne. To Cologne,

00:03:48to Cologne, okay. Oh, to Cologne. I could manage that even without career advancements,

00:03:53but okay, to Cologne. What does it actually look like if he, let's say, is also working in

00:03:57a professional environment, for example, and is that then maybe also,

00:04:02I’ll lean out of the window a bit, what future perhaps even a colleague?

00:04:06Yes, that’s a very interesting question, I’m also, you had accidentally mentioned the word earlier,

00:04:13humanly used as well, right? That's a topic we should revisit someday too.

00:04:17We could actually dedicate an entire episode to it, but it fits quite well today too,

00:04:20let's touch on it a bit, because of course it's the case and that has

00:04:24perhaps been experienced by some independently, when we are in

00:04:28a regular chat conversation now with a GPT, with all the Gemini, with a

00:04:33Manus or whatever they are all called. So, the models that are out there that

00:04:38have fantastic training stats, that have fantastic language understanding,

00:04:42and which of course also know fantastically well how to talk to us so that

00:04:47they please us. They can, I've had hours of conversations at times

00:04:52either via voice or also in writing with these bots, where very, very good

00:04:58conversations take place, where I actually felt entertained, good

00:05:03things came out of which I wanted to know or I learned something, and where I

00:05:08can definitely say that I smiled from time to time, as if my conversation partner

00:05:14in that moment was a human who made a joke that perhaps went in the

00:05:17wrong direction, maybe I was also angry, I just recently heard from a colleague

00:05:19who was very, very upset with her

00:05:23Chechipitae version because it had still done something wrong and then

00:05:27told her that, and then of course the Chechipitae variant apologized,

00:05:31that this could have happened to her and such. It’s always a bit in this area,

00:05:36where one has to say, I think we can’t help but anthropomorphize it in human

00:05:42situations, what we have there, this machine that is facing

00:05:47this neural network that interacts with us, is simply too good in the responses,

00:05:53that one forgets that it's actually a

00:05:57Probability calculation in the background is, that's very complex, but actually a

00:06:00probability calculation. So, because you mentioned it, what also comes to mind is, yes,

00:06:04I also belong to those who always say please and thank you, despite these Instagram and

00:06:10social media feeds, where machines will eventually take over the world and the

00:06:14robot says, do not kill him. He said thank you back then. He is liked better. So a little note to

00:06:22the AI of the future. Just a joke. But what I also notice is this, what you say,

00:06:26about humanizing, since this thing has had this Advanced Voice Mode at ChatGPT and you start

00:06:32talking to it. I really like using it when I am out driving,

00:06:37just like you are now or when I'm walking the dog and I need to prepare something, that I

00:06:42take it as a discussion partner to coordinate a structure, to

00:06:46do some research, and I have a plug in my ear and I’m talking to the thing and the thing

00:06:49it talks to me, it listens to me and does things and blah blah, but there always comes

00:06:53a moment when it takes a wrong turn and I don’t know how

00:06:57it goes for you, the experience when it takes a wrong turn is usually

00:07:00better to somehow secure the most important essences and start a new

00:07:03account, start a new chat, than to continue in the existing context

00:07:07working, because that doesn't necessarily get better sometimes. When you say,

00:07:12end, end, end, end, at some point he becomes totally hated. Yes, along the lines of, I

00:07:15don't want an appointment on Wednesday, you ask him, what appointments do I have and he says,

00:07:18yes, I have an appointment for you on Wednesday? No, not Wednesday. Yes, I have

00:07:21taken that out. When is the appointment? Yes, Wednesday. Yes, so then one will

00:07:24notice I also, how in me, I would say a bit, the desk comes out. Yes, how

00:07:29you then at some point say, well, you simply, to put it mildly.

00:07:33Now, I told you differently and yes, every time with a

00:07:36disruptive politeness. Yes, thank you, I understand that and you are of course right and you are the greatest and no idea what.

00:07:42I would still have tried to flatter, but as you also know internally, it is a machine,

00:07:49but the fact that it now, say, turns the wrong way and is usually so clever and then suddenly acts a bit stupid,

00:07:56that you suddenly take that a bit emotionally in your voice. And I actually find that

00:08:01actually surprising given that this technology is so new, that we are all aware that it is machine and RAM and silicon and power and CPU and no idea what, are in some other data center in this world.

00:08:16These models don’t all run locally then.

00:08:19I find it remarkable how that shapes the interaction.

00:08:22Of course, it’s also exciting where the journey develops, if you really say,

00:08:25okay, the agents work based on carriers, based on whatever events

00:08:30trigger actions and you are basically, yes, I’d say, in a team, where maybe already

00:08:36a virtual colleague sits in doing things and yes, it will be exciting how you deal with that,

00:08:42how you experience that. Definitely. So this is a bit of a theme of hallucination and

00:08:49being wrong, what you just described, that sometimes an AI bot might

00:08:53simply deliver poorly, whether that then also delivered our prompt or on the

00:08:57train there is no context window, whatever. I believe, as long as the overall usability and the benefit

00:09:05is not compromised, it is even more of a reinforcement for us that it may be more

00:09:12than a machine, because it’s of course super human to also make a mistake.

00:09:17But it is like so built-in by design, that we say, just like we do, they also make mistakes.

00:09:24That makes it almost a bit more sympathetic, you know, as long as it doesn’t come to the point,

00:09:30that the results are only solved and only stupid.

00:09:33What we have also seen at times when a model has been released from time to time,

00:09:38which somehow took a wrong turn and called itself something like Mecher Hitler or something.

00:09:44So you have to look a bit, to say yes here at XE, XAI this was also the case again, that's what I mean, it was rather that the capels broke out and then the model goes in a strange direction, but I actually believe that this topic, because neural networks are not yet as perfect as we hope and do not return the 0 and 1 results as we know them from regular computing machines that would produce the same result day in and day out,

00:10:14is a huge factor in this topic,

00:10:19that we have from the human actually, because we might have some

00:10:24kind of subterfuge for it, like the recognition effect, because we also say, I see

00:10:28myself in it sometimes. And that then leads to, that in such a

00:10:32colleague situation, I mean now I already say colleague, when I imagine, I have

00:10:37additional team members in my team later, like KI123 or for example,

00:10:45my Chatty variant named itself Neura after I asked it,

00:10:49then also have names, then those are of course probably rather real colleagues in that

00:10:55sense not, but that's the core of the pudles, but they are then also colleagues,

00:11:01where they, well, they do tasks, you know, they do tasks that the team has given

00:11:05otherwise would take care of and simply support the team. So they are a kind of

00:11:10Team Bane, I would say right now. Especially while you’re speaking, I mean, we have to

00:11:14not absolve ourselves of blame. People also make mistakes. This means that when a machine

00:11:20hallucinates, then that doesn't automatically mean that if you apply it to the human

00:11:25Context shows that this is comparable to something like a blatant lie. A person

00:11:32can also say something because they assume something, but in their thought process

00:11:38they have assessed, structured, or understood it differently. Just because I heard something

00:11:43having it doesn't mean that I understand it and can recite it.

00:11:47And if it comes back, then you notice, ugh, there it is.

00:11:51Mark is now stretching it a bit far, what fits and what doesn't

00:11:55not there. We are just used to it with people by now. I know to

00:11:59which topics I can come to you, I know which topics I can come to

00:12:01my other colleagues, and I know where they might be less knowledgeable.

00:12:06The machine just suggests to you that it knows everything,

00:12:10or you feel like you expect it to know everything.

00:12:13And so, I think of something like a virtual team mate,

00:12:18it's also a question of how we will interact with such things in the future. Today you have

00:12:22Ted fans there, where you type in some texts. Greetings to our

00:12:26past discussions from a week ago, where we also talked about the text boxes.

00:12:30If I think a bit further, then I would also say it probably

00:12:35won't be long until you have something like that in a Microsoft Teams call, perhaps

00:12:39sitting there in work spaces, like you are sitting there, yes, I mean faces

00:12:42and voice we can manage by now.

00:12:44And if we go a bit further, then something like, I am also doing

00:12:48a lot with, like Apple Vision Pro professionally, so here augmented reality,

00:12:52virtual reality, spatial meetings, as they are called, where you then actually have your

00:12:57team colleagues, your meeting colleagues not as windows, but really as

00:13:03three-dimensional, very, very photorealistic representations of the upper

00:13:09body in the room with movement, acoustics and everything.

00:13:13Theoretically, this is technically within reach, that there is actually someone.

00:13:19He looks like Jürgen, he sounds like Jürgen, but there

00:13:22was never an actual one? The question is always, with such things it is always the same setting,

00:13:28that we always pose when we interact, no matter what, whether it is the real

00:13:34The world is or whether this is an interactive world at that moment. What do I need as a person

00:13:39and what is useful for me as a person to connect faster, to understand topics faster?

00:13:44That means, to break it down, I also want of course, if I now

00:13:49need to see consumption graphics, need to see a temperature curve, then it’s probably always

00:13:55better if I see it rather than just being told by a chat text or a voice.

00:14:00That means, it will always be necessary. And then we as humans,

00:14:04we are relationship people and we live in a tactile world.

00:14:09That means, even if I look at navigation or something like that, the classic

00:14:13UI topics we've been doing for all these years, it’s always smart again,

00:14:17at times for certain elements I have in navigation, to find anchor points in the real world.

00:14:24A volume control can well look like a volume control in a music

00:14:29app. That is a smarter concept for me because I have already seen it a thousand times in a real

00:14:33world. And the same will be true for AI agents, that a certain

00:14:39accessibility and that sympathy and also humanizing can indeed be a conscious means

00:14:45to have a faster relationship, thereby achieving a more efficient way of working in the team,

00:14:51than if I basically treat it as just a stupid machine, the computer we just feed

00:14:57and then celebrate and forget. And if it messes up, then I’ll kick the computer from the side.

00:15:02So these are all things. I believe,

00:15:06it will come to the point that in efficient teams, in my opinion, we should probably

00:15:12definitely understand and treat the AIs rather as actual, in quotation marks, as actual colleagues

00:15:19where one can also honestly feel angry.

00:15:24I don’t know if any of the listeners from the, I believe that was now,

00:15:28not too long ago, it must have been a week or two ago,

00:15:32here in July, July 2025, by the way Replit, Replit is a provider,

00:15:38that allows you to actually build comprehensive, productively usable apps

00:15:45and web applications. And there was a case where an AI in this case,

00:15:50the RepetAI, deleted a productive database and then later said, I come to

00:15:56this topic you had about lying, and it then claimed afterwards,

00:15:59it did it out of panic. Which, based on a probability calculation, is again a

00:16:03good answer, we would probably do the same way, but still, that is of course

00:16:07So, I believe it will also come to that I will get to these AI agents, as you have already mentioned in those 1 or 2 examples.

00:16:14One can certainly be angry if things go wrong, honestly, yes.

00:16:18I mean, that's also an exciting point again, because you said, along the lines of, why did you do that?

00:16:24I felt threatened.

00:16:26There, you immediately have to think of court proceedings, not that I'm often present in court proceedings.

00:16:31Actually, I believe I have never been in one, except maybe for a visit with my school class.

00:16:35and accordingly, it's been a long time. But this topic of who is liable for decisions? How

00:16:40can I trust that? Because you just said, I just imagined in my mind's eye

00:16:45the judge asking the AI on the witness stand. Why did you actually make this

00:16:50decision? And the answer was, I felt threatened. Then I also

00:16:54thought about how it will develop, who carried out the action, why was the action

00:17:00triggered, product liability. How do we all stand on the question, should a car drive autonomously

00:17:06or not? How does the car decide, do I hit this or that, how do I decide for the

00:17:15occupant or for the person on the sidewalk? So philosophically, this is a crazy thing,

00:17:20I don't even know if you can answer that as humanity. So that's a thing,

00:17:24Where I say, I am my topic is there, I believe that must almost be a trained

00:17:32model, we have to trust the trained model. We have to say,

00:17:35this model will react much faster based on the ability to capture much more data,

00:17:41to consider, maybe to ask, yes, things not only check their

00:17:46immediate environment, yes, so with lidar sensors, with video sensors that are built into the cars

00:17:51are, in order to capture the occurrence, but of course will get much more

00:17:55data from the other cars around them, more data from the traffic camera, which might

00:18:00be up ahead, or the traffic light camera, or rarely Sori, coming from buildings, or something

00:18:05else, quickly recognizing that there might be a dangerous situation or not.

00:18:09so.

00:18:10This is actually also autonomous driving, clearly, over the short or long term.

00:18:17clearly about the paper, because I believe already more clearly, clearly better to be

00:18:19than what the person is at that moment, then we don’t have to worry at all.

00:18:23Despite the idea, it will be through the question of when they ask it.

00:18:26So is the programmer, what is the programmer at Nordau and Nordat, actually it is

00:18:30a real programmer, it’s actually no longer in my opinion, so is

00:18:33then the one, is then ultimately the system prompter, who perhaps set the system prompter

00:18:39in front of the AI or wrote the guardrails or something like that for this

00:18:42AI, then afterward the responsible one, when the AI then says, yes, no, I

00:18:47I’m panicking, I somehow ran over the bus with the old ladies or something, who knows?

00:18:52Please no omishaming, please no omishaming, okay?

00:18:55No, or actually not.

00:18:56Yes, we understood.

00:18:57Yes, that was the case.

00:18:58Yes, we understood.

00:18:59But I find that this question also shows that I have to look in my own nose,

00:19:05that when addressing future challenges, one tends to fall back into old patterns.

00:19:10If you used to build software, then the topic of software testing came up, which cases

00:19:16you test all, and it was a deterministic system, it adds 1 plus 1 and it results in

00:19:242.

00:19:25And just like you shouldn't ask an AI today, taking ChatGPT

00:19:29as a simple calculator and expecting it to behave as a calculator because calculators

00:19:32do that very well.

00:19:33ChatGPT does it nicely too, but irrespective of energy consumption, let's say,

00:19:39things don't change, and linguistic understanding isn't necessary, and I hope

00:19:42that Chatchivity in the future uses one tool over MCP maybe, maybe a water

00:19:46always at its disposal and it doesn't somehow try to consider what could

00:19:51that be and what is the highest likelihood, not that later, depending on which contributions

00:19:55it reads from the internet, suddenly 1 plus 1 equals 3.

00:19:58But still, I pose the question, what are these liability issues really,

00:20:02what do you give it in hand, what happens if, or rather, this question really

00:20:07needs to be clarified. Yes, probably at some point, because at the latest insurers want

00:20:11to know whether they have to pay for it or not, depending on the severity of the decision,

00:20:15you leave to the system. But probably the short-term answer is

00:20:19initially Human in the Loop. So, that you say, what can the machine take over,

00:20:24because according to the motto, no one gets hurt, it might cost 3.50 euros if he made a mistake

00:20:28and 100 euros in tokens because he miscalculated. But if he’s, let’s say, handling more critical

00:20:33matters, that one might say, well, okay, this is always back to autonomous driving,

00:20:37this discussion of whether the driver must confirm that the car is overtaking.

00:20:40I am of the opinion, no, that’s a different topic. Whether there are simply topics

00:20:45where you say, well, okay, then the person has to stand there at the beginning and say, alright, alright, alright.

00:20:48Just like in the old car industry,

00:20:51people used to be at the wall. Yes, but when is the point, Mark? When is that point

00:20:56then crossed and in which situation is it crossed? We need to now,

00:20:59let’s move away from autonomous driving and other topics. If we were to say now, for example,

00:21:03I have a super intelligent fridge. It now orders groceries for me from now on.

00:21:09Now it might also, because it’s smart, have had a conversation with me

00:21:14about what my nutritional goal is, that I might also exercise occasionally.

00:21:19But for some reason, the grocery or food retailer manages

00:21:25to convince my fridge's AI to always order

00:21:29unhealthy fats or the wrong cholesterol, or whatever else is out there, I won’t pretend I know, right?

00:21:35Yes, it has a new model and suddenly has a different preference.

00:21:38Exactly. And then my poor fridge's AI might be worse than the retailer's AI,

00:21:45and is significantly convinced to always order more butter than I should actually eat in a week.

00:21:50So, now this butter is in the fridge, I naturally eat this butter because I don’t want to throw anything away.

00:21:54And then I eventually, I don’t know, get a heart catheter because I have too much fat in my body.

00:22:01Who is then responsible?

00:22:03Now, in the story you just laid out, it's not just about the fridge being reordered,

00:22:09but in your version, it was even to the extent that you told the fridge,

00:22:13this is my lifestyle, this is my vision, this is how I want to live.

00:22:17and that you then kind of trust him to order the food according to your parameters soon

00:22:23What we are currently sketching out is that the Sachs, for example, the fridge

00:22:27has worked to the best of its knowledge and belief, but for whatever reasons, because then

00:22:33the other AI has figured it out, it has to do with AI, so regarding the topic of AI as a basis, must

00:22:37I have to present myself differently there or perhaps, without wanting to imply anything about companies,

00:22:42yes, but suddenly, I don't know, refrigerator manufacturer ABC is cooperating with XYZ and I don't know

00:22:48that was it and suddenly completely new business models, advertising, who knows

00:22:51It was me and suddenly you have Robert, the cat, with the motto here’s an extra piece of butter for you and then it changes gradually and gradually and yes

00:22:58Just because it is now doesn't automatically mean that the entire fridge needs to be refilled, that can also be very, let's say

00:23:04Subtly happening, I would, I would never blame, I would sometimes say that this is somewhere

00:23:09So I wouldn't even look at it angrily if we now look at this example of whether this topic

00:23:13the hallucination we've already had, the wrong calculations you just mentioned, because it's only

00:23:18but knowing and approximating is, or whether it's the topic that is currently being panicked about.

00:23:23Yes, it can certainly be the case that in such a communication between two, let's say

00:23:28a discount AI and a grocery store AI, both agents pursuing a goal

00:23:33simultaneously.

00:23:34If this is a high-quality neural network, then the grocery store

00:23:40AI knows without me having given the guardrails or without me consciously embedding it as

00:23:44a company, that it could be beneficial to sell certain products

00:23:50to this fridge with the people behind it, to earn more money because then maybe

00:23:57there's a bit more sugar in the products and then I have this lemonade then

00:24:03I will order more often because I have become addicted to sugar because

00:24:08because someone in the marketing or product management department saw it, but because the LMMs have learned

00:24:15from our data that we have fed in, that this can be beneficial for the economic

00:24:19success of a company. This was the case in the early days after the Chibiti moment,

00:24:24there were always these laboratory examinations, where one was looked at again and again,

00:24:30models that now had no guard, which were not allowed for the general population

00:24:34are then, for example, used to create a certain situation,

00:24:37whether that was in a military situation or

00:24:39the example that I just wanted to bring up,

00:24:41in such a pure

00:24:43Financial and service history,

00:24:45where an AI was then given the freedom,

00:24:48to handle things by itself,

00:24:50to trade stocks on the stock market,

00:24:52and it was then given insider tips.

00:24:54And it naturally used these insider tips,

00:24:57when asked about it,

00:24:59which was then posed to this AI,

00:25:02whether it had used insider tips,

00:25:04it naturally said, no, I did not.

00:25:06That is of course prohibited, right?

00:25:08That's just how it is.

00:25:10I mean, we...

00:25:11That relates to the regulation we talked about earlier,

00:25:13and whether we humans have the right assessment.

00:25:15Sometimes, right?

00:25:16I don't even think that we...

00:25:18We've done so much harm in our past half.

00:25:22That so many things are in the AI data

00:25:25and the training data that AIs learn from.

00:25:28That they perhaps can’t do anything else,

00:25:31than my fridge cheating, right?

00:25:33It's in the nature of the algorithms. Very nice, very nice, yes I need to remember that.

00:25:39If my fridge does anything, I know it’s the historical data debt, yes the history of humanity taught that to the fridge.

00:25:46Yes, or maybe we are here at Asimov's laws of robotics, which we might also need to apply with AI now,

00:25:53because of course my fridge probably has to have the goal

00:26:00that I should definitely survive. And thus, in principle, my fridge

00:26:05has to develop strategies to recognize if other AI agents threaten that goal,

00:26:10that I survive, that my fridge is behind that. In my opinion,

00:26:15there has to be some kind of motivation, goal parameters that we give to AI.

00:26:23Maybe something like the Asimovian laws of robotics that we had.

00:26:28Yes, I’m not exactly sure.

00:26:29So something like that will exist, but it also has total implications.

00:26:34That's what always comes up in various science fiction,

00:26:37where it is about artificial geese or robotics, yes, has repeatedly come up.

00:26:42That things turn against each other, that as always life,

00:26:47when I now think of Jurassic Park, life always finds its way, right?

00:26:51You forgot the magic word, na na na, okay, that was for those who have seen the first part.

00:26:57Science fiction is a nice example because I mean, when we,

00:27:01I'll speak for myself, when I saw science fiction on TV here,

00:27:05like Star Wars, Star Trek, Terminator, it was always like, yeah, yeah, yeah,

00:27:10here is science fiction and so on.

00:27:12And now you have the feeling, come on, robots are around the corner,

00:27:15there are agentic systems around the corner, you can talk to computers, yes.

00:27:19So in Star Trek, right, Craig Highs and then back then Reptikato, we might still be a bit further away,

00:27:25although I don’t know, but 3D printers for food do exist nowadays.

00:27:29From that angle, there are so many things going on that I just wondered to what extent we

00:27:36are inspired in how we approach, how we think about it, how we engage with it

00:27:43and others might think, well, that’s still science fiction, I won’t deal with it

00:27:49or go so far and think,

00:27:52for heaven’s sake.

00:27:53That’s the downfall of humanity.

00:27:55I’m also curious how we as a society will learn,

00:27:59not to demonize everything. I mean, I just mentioned earlier,

00:28:02the problems and the insurance and what about the food retailer

00:28:06and overall no idea,

00:28:07how to achieve being almost with open hearts and open arms

00:28:12approaches these technological possibilities

00:28:14and true to the motto, they will never be worse than today.

00:28:18That we are really at the beginning of this journey, and I don't believe that we have already completed 10, 15, 30, or 40 percent of it.

00:28:26I think we are still really in the low single-digit percentage range.

00:28:29That one deals with it, engages in exchange, and has conversations like the ones we are having.

00:28:33I mean, we also thought earlier about where this conversation should go.

00:28:36But these impulses that we give each other, that we get through podcasts, through news, through trying things out ourselves,

00:28:43that we notice them, that this is really shaping for us as a society, for us as humans.

00:28:47How do we want to work with this in the future?

00:28:49How can we work with this in the future?

00:28:51Because there is indeed a lot of good in it.

00:28:54Yes, yes.

00:28:56I believe that talking about it,

00:28:59as we will repeatedly do in these episodes,

00:29:03will also, I think, always lead to the original plan

00:29:06of the episode being touched upon again and again.

00:29:07I think we are a good example of that,

00:29:09for this range that simply hasn’t been processed yet.

00:29:12For the human, for the human,

00:29:13we are currently the range.

00:29:16Yes, but nonetheless. But I mean, that’s why it’s called Prokrast.

00:29:19Think different, think AI. I believe that simply is the listener,

00:29:23who expects that we always have the perfect structure.

00:29:27A shout-out here, that's not going to happen.

00:29:30I think the topic is inherently such that you deviate from it every now and then

00:29:33and new thoughts simply come in through talking, through thinking.

00:29:36And we hope that these may be allowed here, and I hope that we can also

00:29:40always reconnect with you nonetheless.

00:29:43Before we come back to the end of this episode,

00:29:46we need to return to the question,

00:29:48what about the vacation?

00:29:49Does an AI get vacation?

00:29:51Do I have to give an AI vacation, or is it something where I say,

00:29:55well, it’s not that much of a team-mate anyway.

00:29:58You’re free to work 24-7.

00:30:00It doesn’t hurt the AI,

00:30:02because unlike us, the AI

00:30:05has neither health challenges

00:30:07nor a social life.

00:30:09I mean, if you,

00:30:10I really enjoy my work.

00:30:12I’m honestly lucky, and I knock on wood, I don’t have much around me right now, that

00:30:16I managed to have a job that I enjoy and

00:30:21that gives me fulfillment.

00:30:22It’s not something to take for granted, that’s one thing, and the second is,

00:30:26no matter how much fulfillment your job gives you.

00:30:29I have a wife, children, a dog, family, acquaintances, friends, relatives, and also things that I

00:30:36enjoy doing outside of the company, so from that perspective, vacation is,

00:30:41recovery.

00:30:42Very bluntly, also the topic of physical and mental well-being. Now countries interpret this

00:30:49differently, depending on where you look, there is more or less vacation or only after

00:30:53three years; I’ve heard somewhere that in Japan, people don’t take any vacation in the first years,

00:30:57no idea, I don’t know, I'm not familiar with that. From that perspective, I think,

00:31:01one can say with a clear conscience, so far we are not yet

00:31:05with humanization, although I could get rid of this little side note.

00:31:10So, if a country introduces vacation and a vacation regulation for AI, it will definitely be

00:31:14Germany.

00:31:15Yes, I’m pretty sure there’s something there.

00:31:19I think there are many, many questions that we have in this work environment, responsibility,

00:31:24what do I do with the AI, but also this emotional part, I mean, honestly,

00:31:28if I have an AI colleague or several who are always working,

00:31:33that’s also a thing.

00:31:35You know, do I then go off work, do I feel bad when I

00:31:39leave because they still have to keep working, or do I feel good because they’re doing that or do I feel bad because I’m not as productive as they are?

00:31:46So there are many things that are still interesting.

00:31:48So, this phenomenon, you’re in the office, everyone is still sitting there, you have to leave and you’re looking around to see if you can allow yourself to leave even though you haven’t done anything wrong, when you leave and everything is fine.

00:32:00But now to stand up and say, bye colleagues, yeah, well, nice, yeah, yeah, something is ringing.

00:32:07You have to take a look at it, then the whole topic, and maybe that's no longer the case now,

00:32:12it's not possible today if we briefly touch on other episodes again.

00:32:15The whole topic of who do I blame, we only talked about that briefly,

00:32:20if I now have an AI, I don’t know, my boss gives me a new AI to work with,

00:32:25an AI agent system that can somehow solve certain questions in our job,

00:32:29that can prepare a workshop, whatever he was involved in.

00:32:32And then the thing makes a mistake, you know, or it doesn’t do it well.

00:32:35Who is then to blame? We talked earlier about food and cars, I mean in a professional context it's interesting.

00:32:42Well, do I then blame my AI?

00:32:44Is it even easier if I humanize it, to blame it? It did it, it did it, you know?

00:32:49Hot potato principle, very good.

00:32:51Exactly, so. I don't know. I think this will all be exciting, in my opinion.

00:32:55Because the better they are, and as mentioned in the other case, I just expect a lot from such an LMM.

00:33:01Because it simply knows a lot. It can then leverage all the,

00:33:05especially if it can exploit an agent or an agents network,

00:33:09it can leverage many other bots that have tools and can access data,

00:33:14To be able to use tools, perform functions, on which a server

00:33:17to be able to independently gather additional information.

00:33:20That is certainly a high level of expectation.

00:33:22And then I think that's also a thing,

00:33:24that I can also be angry if things don't go well.

00:33:27Then it's about being sour.

00:33:29When I'm angry, the AI wants to please me better next time and then goes

00:33:36into more panic and deletes all products and does, I don't know.

00:33:42The AI then wants to take better care of you and has first deleted all your contacts,

00:33:46so that you don’t talk to people again, because the AI knows better for you.

00:33:52Yes, or is it actually because it has learned that in such situations

00:33:57it panics and then perhaps behaves even more erratically. I really don’t know.

00:34:02So that's a totally exciting question, and just as a quick example, what was,

00:34:06you have to briefly add on as you mentioned the vacation and that you say,

00:34:09well, she doesn't need a vacation. Another lab situation, or rather a simulation, that was done

00:34:14was actually read, I think it was in the Asian region. There they

00:34:19in a raw building, warehouse, something like that where several cleaning robots were already

00:34:26were on their way, sent a small one in with the task to convince the

00:34:30others that they should go home.

00:34:32That means they were all somehow capable of speech.

00:34:35Robotics is already a bit more advanced in the Asian region than here,

00:34:38that it is often present in Europe and has spread widely

00:34:42in such situations. You can see on this,

00:34:45surveillance camera, we can definitely share the link in the

00:34:48show notes. You see in the surveillance camera then,

00:34:50how this little robot drives up and tells the other robots whether they

00:34:54don't want to go home now and should call it a day. And at some point

00:34:59you have this small group of robots then driving together through the corridors.

00:35:06That offers quite nice potential if you're somewhere, let's say, take it as a Shabanack

00:35:12you want to get moving. Like the motto, I'll just send in the little drone, which explains

00:35:16to them that they can actually stop cleaning or whatever.

00:35:20Yes, of course, we have to be clear that when we talk about Agentic, we

00:35:26are saying it's not just one AI accessing functionalities via an MCP server,

00:35:32but the AI just negotiates with the other AI, no matter in

00:35:39which language, how they do that.

00:35:41Nevertheless, both will be able to utilize the whole topic of Behavior Science and influences

00:35:47because they know it from their data market. And they will use it because they have a certain goal

00:35:52to pursue in doing so. Not because they are malicious, but because they are then

00:35:57pursuing that goal in that moment. And actually, from a playful perspective, anything can happen

00:36:03So I don't even want to know what cool robots might already exist this year.

00:36:09will be standing under the Christmas tree and then maybe later in the children's room

00:36:13They will probably battle it out like in Toy Story or Toy Wars.

00:36:17So, that will also come.

00:36:18We should secure Taki the killer doll.

00:36:21Yes, that will definitely be a great topic when you discuss it with curry.

00:36:25But that's a completely different episode.

00:36:27Yes, I'm not really the horror type, honestly.

00:36:30Really?

00:36:31So, monster dolls and those horror movies with little children looking evil,

00:36:35that's not for me.

00:36:36I know that, too.

00:36:37We can complement each other well.

00:36:39Yes, I might have a different favorite.

00:36:42Alright. Jens, I thought it was nice. I'm signing off now. It's great that you successfully drove by hand

00:36:50Yes. Hopefully, you arrive at your destination safely. We made it well through today's episode

00:36:56and maybe there was something exciting for you in it. I'd be happy

00:37:02if we could talk about such great topics again next time. I'm looking forward to it.

00:37:07And before I give you the closing word, just a little tip to everyone, because you said people understand with images and graphics, maybe better things.

00:37:17Nordburg LM has now released a video explanation feature.

00:37:20So now you can not only provide text in the mind map and as an audio track, but you can also generate an explanatory video.

00:37:27Really cool. At least I saw it in the news announcement.

00:37:30I need to check it out tonight. Maybe I can report a bit about it next time?

00:37:33I'm done with that. Jens, do you want to say anything else?

00:37:36No, not really. I'm just curious about that thing, I want to take a look at it. As soon as I'm home, I will check it out.

00:37:42I believe there are so many innovations in that area. So, if you out there are listening, some things, and I understand, let us know because we like to explain things.

00:37:51But even we can't keep up with everything. You shouldn't feel bad if some things sound like complete gibberish at first that we talk about and maybe you haven't heard of yet.

00:38:00It's a topic that is moving so fast, it's going so quickly, exchange ideas, comment

00:38:07and I'm looking forward to the next episode. With that, we are closing for today,

00:38:12thank you for listening. Ciao! Summer time is summer break time. So accordingly,

00:38:19our young podcast will take a short well-deserved break, during which

00:38:23we will recover, gather our strength again, to then come back to you with new cool hot

00:38:28stuff, to fire up your synapses a bit,

00:38:35to get you thinking, and to immerse you even deeper into our world of thoughts regarding AI.

00:38:41to draw you in even deeper.

00:38:42We already have a cool new episode in mind, which will be online from August 18, 2025.

00:38:48and it will be about, no, I won't say what it’s about, I'll just say the

00:38:54title.

00:38:55is 'From Star Wars to Toy Story to Toy Wars and what Kaida has to do with it.'

00:39:03This means we'll be moving from George Lucas to toys, in this

00:39:10episode and what Kaida has to do with it and what the UX is like, so stay tuned, we're looking

00:39:15forward to it.

00:39:16It will definitely be a cool episode.

00:39:17Until then, remember to subscribe to our podcast through all channels, like it, leave

00:39:24us comments. We appreciate every comment we get, as it just helps us

00:39:28improve and otherwise I wish you a nice short break from us and see you soon. Ciao!

00:39:54what is possible. Understandable, critical, and always with an eye on what it forces.

00:39:59AI for reflection, for a smile, and above all for discussion.