Mythos, Jobverluste und das Regulierungsdilemma: Die KI-Woche #1 - mit Thomas Seiger (KINN)

Markus Kirchmair: ⁓ very to the first conversation today with ⁓ Thomas from Hello nice to you ⁓

Thomas: There will be lot to discuss. We have basically every week. That's there is also the AI week. Pretty much a lot of news.

Markus Kirchmair: Exactly. could now be called AI day or AI hour, because so much is But since it the weekly newsletter and the weekly podcast is quite good, the name AI week quite well. What's on the table? What I been doing the last few So much has happened that the things that happened the days before are actually a bit out

Thomas: What's on the table? What did I eat in the last few

Markus Kirchmair: Microsoft presented at beginning of the week. Then we Opus 4.8 last week. Many other topics. Mythos is coming soon, or will be to other

Thomas: What is this mythos for all those who don't know it? What does it

Markus Kirchmair: Mythos is the powerful model that concerns cyber security from Anthropic. It's discovering critical in critical systems as soon as before this technology is available So you assume that this technology will in few months, just as for... criminal purposes can and that's it's about now that you your own small circle that they their systems again. They big software companies, that goes from flight traffic to machines to the smartphone and so on. It really affects many critical systems. Yes, But that's already few weeks ago. There's interesting feedback from Mozilla. in April.

Thomas: This is a big callback for many who are not aware the fact that there many possible attack vectors.

Markus Kirchmair: So this is a big call out for many people who are even on the skis. don't know if there are possible attack vectors. Exactly, a big call and also big race. Who gets this technology and how much time is left to it to in the bug fixes or the security gaps.

Thomas: And that's only from Anthropik, Or are there others who ⁓

Markus Kirchmair: Are there other that can this? OpenAI has with GPT-55-Cyber at the start, should be powerful. There are relatively few transparent benchmarks for both models, but it is assumed that both are on similar levels. Is there any estimate how long it until the other offers are out?

Thomas: Is there any estimate how long it will until the other providers from the Far East close up?

Markus Kirchmair: Good question. It's hard to estimate, but at moment the open source is very strong. that, in China. And estimates are rather so at three to six months, what we currently at the performance of the public flagship models. With Mythos, we don't exactly. Not that she knows, but it opens up now, so I think there now 150 participants and some in the EU.

Thomas: I hope we don't get completely stuck Exciting. What else is What else is there? The last two weeks

Markus Kirchmair: What you about the last two weeks? think something interesting is the AI Act and regulation in America. The Americans are on the opposite path with the war order. because the AI, the technology, is the national security issue and we can't into the loop and ⁓ brake

Thomas: Okay, and in the EU we basically have the same but it's more personal security,

Markus Kirchmair: Yes, is a lot about fundamental rights and less about race races, I think. you can from one point of in public or in official communication that it promote or that companies in the EU are also competitive. But that is viewed very critically So the reality. It's hard to tell that the race is on the and China. There is still Mistral place 76. There is always movement in the whole but it's hard to how we stop

Thomas: Yeah, so that's also hotly discussed now at all the events I've the last few weeks. So how does the company So what does it want to So in the end we still the data protection basic regulations and we also have the EU AI Act, to speak. And then there was the topic of digital omnibus. Can you say about It's not quite clear how it looks with the deadlines that we have

Markus Kirchmair: the ego-EI act, if I remember correctly. And then we have the digital omnibus. Can you say about It's quite clear how it actually looks Yes, it is actually not much affected by You shift on a high-risk period, those that are in August, so early August. the high-risk systems have involved and we it into the future. But it is now changing bit fundamentally. We improved in of this competency obligation. But is not really a big change in terms of the AI. It more that it the pressure out of the frist. ⁓ The companies in the practice often have the biggest topic, which the DSGVO. And that's a huge topic when we about war, because of course these big challenges.

Thomas: So will be lot of pressure, will affect the medium-sized companies. How do you this in your environment and in the companies you advise? Absolutely.

Markus Kirchmair: Absolutely. think the biggest problem is that you actually from the intention that want to legal security with the regulatory system and it feels like it very well it's more legal insecurity that is created because there are still too many cases where is no legal talk at How is this to be How is the duty How far does it in a single case? it really just about to exclude or is it still possible to like a cookie-panner? There are different views on There are many open questions about how to put forward at the time at an extreme tempo, extreme dynamics in the whole thing. And then it all becomes bit of risk management. it becomes, from my point For many companies, the question how to fulfill but what risks do take. And that's a bad sign, because there are laws for so that everyone to But if you with the DSGVO, I think 90 % of companies could do something with the DSGVO. Now, no matter if it's a cookie banner that's missing or analytics script that's too early on the website or an application that in it's mail and so on. There are so many ⁓ topics where the good intention, the regulatory one, which you can understand and have to understand, and the practicality.

Thomas: And that's ⁓

Markus Kirchmair: we going drift

Thomas: This reminds me of a workshop I recently had on topic of to an innovative or an curiosity through a fluttering space where you really want learn and out new things. And it emphasized that it is important that you a space where no fear, where you really feel safe to try That's also the topic here. you at the companies that are trying to minimize but the same time they this feeling that they could something. How ready are they to invest in a technology that is not really safe, where you don't know yet? In America, where against law, where aren't many laws that can where you invest now, there's a lot to gain, there's lot possible.

Markus Kirchmair: The feeling in the space that someone could attack How ready are you to invest in a technology that is not really safe and doesn't know Compared to we in America where it's against you, where don't have many laws that can and where you think, okay, invest now, it's a lot to win, it's lot possible. The issue is with DSGVU because it's all in local servers and data centers. And they can use the state of the art models without the big... I forgot that. That's a big lever in the economy. I think the dilemma is that on the hand... has great understanding that you have to participate, should early on and so on. So certain fear of missing out, which is not completely irrational in most cases, at least I see. And the time, this legal insecurity and you have to find a means, think, if you don't go naive and that completely unregulated legal and data protection issues and on. And the time, I didn't wait 10 years for everything to

Thomas: Yes, you want security so that you can invest and innovation. At the time, also to be by society. Now we the video, we can play in America. On one hand, being pushed that AI is promoted, but ... In the public perception it is bit different,

Markus Kirchmair: Yes, which video would you like to show?

Thomas: think the video was not that good.

Markus Kirchmair: University graduates. It's big event in America when the people from the universities leave. And yes, now well known I if it's possible, but many have already seen it. Eric Schmidt. Emelika Seehoff from Google is to a stage when he about Ki. And also others. And that course said a lot things. You can that the mood has changed in the population. And that's not surprising More surprisingly, with the lack of sensitivity that into And before the students, we talk about the graduates, because it's new that the mood has for long time. There is this Harvard youth study. from autumn 2025, which already said that 59 % of young Republicans and 66 % of young Democrats say that AI is one of the big challenges for their professional future. And that was actually pretty to because... America is otherwise on many political issues. And that's actually the topic that both sides have common. And that you still out and talk AI and change and the predicates of the observers, where they are all confused with reality that the entry jobs are currently really less, so also numerically identifiable. There different studies that around 14 percent, always fewer studies. I know it from one your events, from a children's event, professor was there and explained that you compulsory practice in the degree program because you can no longer find for your studies.

Thomas: What the ⁓

Markus Kirchmair: Something with data and analytics, in that direction. It's really about the companies saying, well, for what I to for a couple of months as a work student, for research or whatever, that's now doing Cloud or Gemini or something like that.

Thomas: So these are very simple, research activities that a practitioner has

Markus Kirchmair: But it won't with the stupid tasks. There is another example of a form, I have to see if I quickly. there. ⁓

Thomas: I find it exciting to see And of if a university graduate can imagine that it's very difficult to find a job, even you've several hundred thousand dollars in this education, that's not a surprise. Absolutely.

Markus Kirchmair: Absolutely, absolutely. And Ken Griffin was from Citadel. He about how depressed he he home from work He realized that he couldn't AI agent work for someone had highly qualified directors in his company. He the analysis of jobs and so on. And suddenly he that he it better and faster. It's interesting to see that such a side, doesn't many people cold. That's it's such a topic. If you look at how it's or what turnaround it in public communication from Sam Altman and so on, then it doesn't as surprise

Thomas: That's also very exciting. I've different opinions about And one tenor, as she to turn could say that she works now, thanks to KI. Because I it that I can what you're saying, that you more time than before, KI not there.

Markus Kirchmair: I see that you more time than before. You will the early adapters and a lot and all ideas that we have had long time can now be The question is how far is it scalable? If everyone can their new favorite game in and we're not yet at the J-Play level, but we'll see that in two or three years, if everyone can do it, how many games does world how much software does the world need and so on. There is the topic that many say, okay, there can't enough I'm more the opposite, because I say that a of software is actually used as a tool for people. Photoshop, for example. No AI needs Photoshop. AI can Photoshop, but it's actually the other way AI generates a picture from the first to the last pixel. And if you don't like you don't need this special filter for healing or something. I tools, I just it and it the picture as it should be. And this is the in many cases. Software is extremely reduced to minimum, to input, control, feedback. But not 100,000 functions that I as a human being to the computer what to So changes from my point of view. There so much software that allows software to with other software. That's all about becoming less relevant. And bookkeeping software in end, if the cloud coworker can bookkeeping better and faster than in a modern bookkeeping program, you only need the data storage as software in

Thomas: But most people don't just to software stuff. If you say, I use KI, there are many developers who don't use to program But it's used to write all kinds things. You're also author, for example. Or to creative content in Or research that runs all the time.

Markus Kirchmair: Poocha. ⁓

Thomas: And that raises the question, mean, of software, I see that at some it a process where you think, okay, why do I even software? I can the output directly to or it generate.

Markus Kirchmair: I mean, software can also be seen that it will a process at Why do I still need software? I can the output to So don't get I wouldn't say that software is no longer needed. But lot of the application software, all the apps and so on, which with VypCoding right now, they in one or two years When we the next generation of mobile phones, where this is to more agents, is no longer right Most of them. There will always be yes, I can imagine that. But... In the end, you have to understand Yes, yes. The point is less that I say the software will no longer exist, it will less. The point is more that I say...

Thomas: In effect, the AI has understand If the software doesn't understand you, the AI can't do ⁓

Markus Kirchmair: What can more and faster now is the demand for software will not with the tempo as the supply increases. And I think that's the big dilemma.

Thomas: That the whole market is by many different small software stories that everyone builds?

Markus Kirchmair: Exactly, many jump and runs do you in the app store if everyone it How many pocket computers, apps and so on do you need? Yes, but also with the book market. How many books does the world Who can consume And then, course, ⁓ different who the big authors I think the big names don't to about but now these independent authors, also independent game programmers and so on, who They have a theme, they are extremely fast and in the they on AAA level. For that, big studios have On the other hand, if everyone can many millions of car racing games or whatever you need, it is also in certain way.

Thomas: That can be, but as I said, everyone can cook, everyone can paint Binzler has a pencil at Everyone opportunity. Everyone has Everyone it. Now it's democratized. The tool is to many people who can it. It doesn't mean that everyone uses it. It doesn't mean that everyone it to excellence. to use it it could Even if I could I'd rather myself be cooked, because someone just likes to do and can And it's also the question what now this output that by KI, whether it will also somehow still be bit hype, bit a slug, yes, I want to try but somehow does everyone it actively, does everyone really want to create something with actively, or does it then less that would it really well, where then want it from it?

Markus Kirchmair: I think in some areas will that. ⁓ photo generation, video generation, everyone to the beginning, all the action figures and so on, and the it apart again. I that. And on other hand, if it beyond the playing, cooking is certain process or also through the system of pencil, if I want to draw something, first of need talent for it, and secondly, probably need a lot of time and patience for short I know people who do pencil drawings. think they spend two weeks on a drawing, hours a day. Extremely. It looks like photorealistic. But now is no process behind it. That's fascinating. my point Google.io last week, where Omni and the new video model After that, many went to Vypcoding and so on. always presented in sense of the creators, what they can do and how this creative process and so on. when the creative process is so far reduced that it's actually only half a minute to think what I want and in it out in perfect form. Yes, not yet, but ...

Thomas: Is this the creative process that reduced or the execution process of something creative?

Markus Kirchmair: Yeah, it's a good party. ⁓ Because that's the point. it's less than a minute, would be very difficult for this creative process. The question is how many loops you make. And then you always the end, the person is the one who the taste, who decides in is it good or is it bad or who coordinates, who chooses. The thing is, if you about at the end, you to admit KI will improve You can see that agents and sub-agents, how suddenly KI other tasks for me. And often it top results from a presentation or something that you might not have even thought of doing yourself. And then you go and positively surprised. Of course, you can throw in. That's the moment. feel ⁓

Thomas: This is what we're going to now. I think it's very fast and people are surprised and flashed. Everything is cool. The pictures that were created the Ghibli stuff. I'm an artist now, I can it too. Suddenly everyone did it and we weren't very special anymore. And also the new X and all kinds of AI-generated content. But again, have a certain smile. This has a little smell, taste of Anthropic, Open Air or whatever. And think that's I see, in terms of this creative process, the challenge of as much possible... to what is not in the broad master app, but what is my special flavor, which is a product, my signature, and how much time I to invest, no how good the tool is the execution, to it into the materialization, whether it the or whatever. This time I put can make a difference, but the which then this result from all the others,

Markus Kirchmair: and then say, this is product, this my signature, how much time do I have invest, no how good the performance is, and then bring into materialization, whether it's the step or whatever. This time that I put there can make a difference, but the what this result differs from all others, who maybe only bragged spoke I'm with you and at the same time I see a certain trend in which the creative person you and and your style and wishes better. I can this with a very sober example. If the creative week, something regulatory, and I say, agent, let's a at the creative week, what new articles we have and what sources we used for for which my books would interesting for a new edition. Then he says exactly that it in this book and in this book and then he it the chapter and looks at the style and gives it directly in. But when the book was in the first edition lot of work has It goes extremely far in that you can't get out In imitates you. You can say it's not really intelligent, but gives you pre-intelligence to be. The question is less philosophical. Is it intelligent? Yes or no? Or is it creative? Yes or no? But what is the impact of what it can do?

Thomas: Your thought is that if you let the listen, it will lead to creative process, because it already knows you well and also knows what your taste would if you a book, generated and so on, so that people can that it a work of from you, where you have only ten seconds, and it still looks like you invested

Markus Kirchmair: I see it that because you are a creative person. ⁓ many tools, to different channels. They can your website, can your books, can your presentations, your press releases, your emails, your calendars, they can know and understand I'm just very keen to this connection and that's...

Thomas: If it now, many people the AI watch to quickly and effectively and large quantities, that's the point where you say, then the market is flooded and then it's less relevant.

Markus Kirchmair: Yeah, and it's yeah, and they were, course, tempo, me, Dennis, boat, Stag, Stag, this, this is not linear, as my sister, so ⁓ this Stagg, Laufend in a smaller tempo and in one and couple, so I'm on the boat and not far and then. ⁓ this value or were Then there is course this personal factor that you in art or that you have with various personalities. also with authors, crime authors, the big ones, they will not the problem that they somehow cannibalized The smaller ones rather. The artists, just as little as thing, or as they with the influencers, they will all be no more avatars, so no one needs more influencers. The big influencers are brands, brands, messages, personalities and that's what stays, live events, live...

Thomas: Even if you any personal because the can replicate for you.

Markus Kirchmair: I look at and the football World starts again, You it as a computer game and instead people who see people on the pitch or race to There are probably 100 arguments why you car races in reality because it dangerous and so on. and let's just on the computer, then no one And then we it anytime and regardless of the weather and so on. And yet there is a big demand for So there is a big difference. And that's why I the really big ones and the personalities and sport, culture and so on, are from KI in my opinion very little affected. Exactly.

Thomas: Exactly. yet, is the great disadvantage. It is the difference. And that's why the big companies like Ford, Kuro and the rest of the companies are very few. Especially those that really in public. Can you say Yes, understand.

Markus Kirchmair: Exactly. course, personal branding is very important. But one can't assume that 8 billion people in world are suddenly personal branding. Because if all are personal brands, no a brand apart from the not everyone is the for But that is course certain dilemma that only a small part the can in

Thomas: I think there will be a of subcultures, regional subcultures, which might have an influencer space of 300 people, but are anchored the network there and can use to support the AI. But maybe not reach the masses as it was in the last 10 years.

Markus Kirchmair: But maybe not as much as in the last 10 years? I imagine. I have customers in tourism or in the industry. And the in the publishing industry is personality and regionality. These are the blind spots that have been worked in last decades. We lot of copy-based journalism, lot of money from very much generically. So theoretically, if you a daily newspaper, you read in, or if you have choices, you 90 percent of the content very, very similarly. And I think that's exactly the potential there, that you say, what is really relevant for the people and their environment? And how do you personality in to ⁓ that?

Thomas: Do you see trends in your customers? How is this used?

Markus Kirchmair: There are different approaches, course. Personalization in sense of the segment of one, which has been 20, 30 years in sense of there is not one newsletter for everyone, but everyone gets their individual. That has so far failed in the past, in the end, even if it technically possible, there was still too much effort And someone who doesn't have a person full day for New Newsletter, set up. For him, this is actually not practical, because there are 100 other levers and measures that can marketing. There may be more happening I don't it's completely 100 percent individual. The point is, to maybe briefly summarize it, the future of streaming. At the moment we have seen extreme progress in video generation, extreme quality improvements and so on. now you could say, at Hollywood level. Maybe it's not 8K in some parts, but 4K and so on, and maybe there are still glitches stuff. But it's at a level where it competitive and where it's It's the big bottleneck, of course, the computer topic, is, extremely resource-hungry. We have already the first short films in Cannes and so on, 20-minute-long Ukrainian films, and that will be more. And some point you the opportunity, if there enough computing capacity, the moment of bottleneck and not the technology itself, the technology is there, a Word document or a PDF and then half an hour you it as a film. And then the question how much is it a community experience in the sense of need Netflix and so on in the direction of segment of one that really everyone only on the basis of his preferences his own series and content put or In the end, need a product that next week to exchange ideas and get the next film.

Thomas: I have say, I like to surprised. I like to a lecture where I no idea about the subject and that's also author I don't know. And then I it and it's a new perspective. That has inspired When I my content recycled in a film, I'm interested in I'll take a look at But Bridge into the Choir is just somewhere...

Markus Kirchmair: I've that before. If I get a recycle from in a film, I if that happens. I'll down. it's bridge into the choir. I think that would be a bore. I imagine, and that's what I wanted to beyond, that this segment of ONE doesn't go to very end, even if the technical possibilities at

Thomas: I think you would get bored I think.

Markus Kirchmair: because there are certain other aspects in it. For example, I am someone who loves reading opinions, even in the form of maps and so on. I'm in what others think about and I can then for myself if I agree or see differently. But I also with opinions that are different from mine. And that's why it would be bad if I my own opinion all day long. The topic still some algorithms. But that's why think that this segment of one, this concept, though it creates a bit of the ultimate, the theory goes into practice. So personalization continue.

Thomas: There is a certain influence, as we have seen from all these advertising campaigns online on all social media sites. basically things played out, you like then you stay longer there. You feel very familiar, you feel very comfortable in the bubbles, speak. You don't click anywhere else. You are probably also instrumentalized by larger corporations that it, now also general, TikTok and things that. which is also very difficult, because of the dopamine. And then I also the question, lately, also, what concerns AI. So we basically something that dopamine. You have very quick solutions and just more of it, in the end we something bigger with we now, if we just assume, there is an AI that is already smarter than us. And they know how use And they play us exactly what we want. And the time, something else on in the background that we don't understand. In this sense, work is decaying us or we are decaying them. It's a somewhat provocative statement, but if we dare to the won't really relieve but rather we do more for the AI. Our calculation is efficient than ever before. We more in the computer than our families are.

Markus Kirchmair: But I think that's a phenomenon that is in the early stages. I think that's driven by the limitations that you with the doc. There's the moment when say, Thursday evening, I've all my weekly limits, it's to nice long weekend and then there's... Freitag früh, ein Post auf X for Anthropic, happy Friday, we are free!

Thomas: I think so. I would like know how many other people made so because I haven't made I'm already trying hard, but I haven't yet. I'm curious what you do Maybe you tell

Markus Kirchmair: You can't see the limit. ⁓ I was at Anthropic on Cloud, at 8 o'clock, resetting the weekly limit. I think it Monday or Tuesday, I was already at the weekly limit, already resetting and then it worked again. Now I'm completely exhausted although I have Macs and an Ultra account at Google and so on. At the moment, we try to do with as long possible and to use Of you with the tempo think that's also hard to in long run. And the time, there's this gold-grey atmosphere. Trying, experimenting and whenever something has bit off, there's the next model.

Thomas: Maybe we you the meme. There is also this notebook that you have held before and how you keep But I have to I go to bus and let my agent run and don't the notebook.

Markus Kirchmair: I don't know if you spontaneous. ⁓ There is a great app for It's called Amphetamine. You can it to unlimited. Then it on the backpack. At the moment it like that many others. I notice it on X when you the comments like every minute you on work, it's a of a wasted minute. That's the feeling behind it. You have to the rent. Exactly. Then the question do you yourself more in end? And would say, the moment I more time, I have a lot output. But at moment, lot time is experimenting and especially in the quality. So what you used to have probably the same amount of and at some point satisfied with the result and now you say, now I'm faster, now it to better. That's a bit, I the current situation. But ⁓ it will also from generation to generation because you can get outputs better than small version jumps. ⁓

Thomas: Is the question whether we can reach a point where you say, have an idea, I get the result and it fits and I don't have that stress anymore, okay, I have to do and I have to better, the whole thing go. that you can really say, it's it anymore. Maybe in the future, but as a trainer. I think that's general As you say, a race with others who early adopters at the moment. That you have a bit of a fumo, you want in Now that the chance is there, whether that will

Markus Kirchmair: Maybe not as the point, but as a trend I see that I sooner or later it to be put in place. At moment it is extremely shaped, a lot of fascination and shock at time. There are always these fascination moments, like, wow, this new model can this and that. In next moment again the shock moment, if I can with this model, the other competitors can too. And maybe they it... First in a few weeks, but later in a few months they it. That's why it's so dynamic at ⁓

Thomas: I think especially when you content creator, also to little bit more when few days before the other person, of course, then it a difference.

Markus Kirchmair: This is so.

Thomas: That's an impressive. What else do have in our AI week? Now we have basically already a big arc.

Markus Kirchmair: What else have in our... ...e-wok? We have a big floor of tension A big floor of tension, yes. An open letter topic has been published or tonight. Several CEOs and leading researchers demand that strict regulation in America be implemented again. I am too little in that now that

Thomas: Can you remember where the discussion was between the Pentagon and Anthropic? What they fight

Markus Kirchmair: think there were two red lines, one of was that you don't use I think Anthropic wanted to exclude that it against its own citizens. I don't remember what the second one was. Mass surveillance was one topic, the people I think the other one was in that direction.

Thomas: Autonomous weapon systems or something like I think so. Just to think about There are some drones that with an embedded AI and decide for what they do. I think that's a justified red line. I think so too.

Markus Kirchmair: And I think it's actually already happening on the battlefield now, for example, that if the contact is the last moment in the target flight also with the support of So I'm too little in the picture, but think those already topics that ... There it's about bioweapons, about research, because you almost say, can very useful things, ⁓ of course, in medical research or in general the topic of biology. Demis Hasavis, I it as something that is just as solvable as program code, the whole topic And that course, as many topics, two sides. On other hand, many ... To The virus can also be if it properly programmed. Exactly. To create the bridge to the segment of one, maybe this medication or the vaccination for the individual who needs this that, where it best. In the way, can probably be in a bad way.

Thomas: A virus can heal if it properly programmed, but it should an epidemic. You could from a person's could only affect

Markus Kirchmair: or a certain ethnicity or whatever. There are many topics, which is global challenge, just like many other topics. It's good to a national or regional level, but in the end, the global community is... I think of trust that you can say that you understand each on international level, that are certain red lines that no one to In the end, might happen in secret or whatever, but think that trust, even if the rational level, certain things are bad for everyone.

Thomas: We are in a dilemma. There was an open letter a year ago, where you had to down the AI a bit. what didn't happen. Elon Musk was one of the few who signed back

Markus Kirchmair: So few were there. Many prominent voices were there. But the dilemma is that you can't control You can't inspectors there and make sure that the research is no longer happening. So no one is on the server Exactly.

Thomas: So, KI is not only on one server, but worldwide and also China, of course.

Markus Kirchmair: And also to the regulatory leap. You can course say as EU or as another region that you certain providers to certain things or that they are not allowed in certain regions. But what is too little or is discussed is the topic open source, where we of a months of hunt. If you realistically, component in there where you can give the big provider some barriers, but you can't stop It's like finding the US-based sticks to North Korea. You forbid that. That's the dilemma.

Thomas: Regulation is certain degree, but... ⁓

Markus Kirchmair: Especially because things will into the underground, where they are longer controllable. If you can't do which may become practical with large providers, then you have...

Thomas: Let's it a the room so that they see it.

Markus Kirchmair: Yes, that's probably in company as on a social level, when you certain things. Of course, if it a benefit, then it's difficult. Whether it's adults or children. You have to deal you have to show what risks there are, you can't just get rid of them. Yes, yes. It was actually like that.

Thomas: If it has a benefit, it's difficult. Whether it's adults or children. You have to deal it. You have to show what risks there are. You can't just lock

Markus Kirchmair: As far understand, one of the motivation points of Elon Musk, when he Open AI, was was that Google would take the concerns too seriously and he wanted to an provider from dominating these ethical issues are too seriously. And that's why I can't the development. But if he doesn't stop it, wants to part of it and the development himself. could argue about it it better or not. But it's a somewhat similar point.

Thomas: Yes, least the world has not yet ended. has been often prophesied that HGI would and HGI would announced but so far I don't any. I see a lot of stupid AI, even if it intelligent the same time. How do you that? I see it way

Markus Kirchmair: I see that think the issue of AGI is a bit like a topic that is only one away. In the sense that someone in the news that AGI been today. and there may be a statement from the president or something else. I don't think it's a one-time, one-time, binary event where you the point where you say, now you don't have it, AGI and suddenly you AGI.

Thomas: It means Artificial General Intelligence. It's about having a career that doesn't just one specific area for which it's but that understands what it

Markus Kirchmair: And there are more more voices that say that we have had long time. If we about ATI three ago and now think what we already have for multimodal models and how far we most people only a fraction of then many would already say that this is actually ATI. Then there is the increase form, the Artificial Super Intelligencer. It's always difficult with the terms. I the big leap was the agent topic, which the few have on We had Google in to find to answer information. We Wikipedia and then chatbots, the answering machines. And that was an evolutionary stage in my opinion. But the actual game changers are the agents where the suddenly able to With the ability it had with reasoning on top. And from my point view, already very strong in direction of AGI. then a major topic of how much the real part of improving oneself. And after that we see It's not the point where all the Entropic programmers can home, because now it's the AI itself, but it an increasing amount of And so it's a smooth process.

Thomas: So it wasn't always the news that the latest version of Cloud Code has no programmer written but only the AI has written or something?

Markus Kirchmair: I think you could that they are bigger than today. understand that. Otherwise they home and we would have to But you could say that maybe no one has mobile code. I think the least programmers have in last weeks and But that's exciting. There are also contradictory messages. think I read an interview quote from Hazavis, he said that the current models have nothing to with AGI at And then I think it wasn't the same interview, but within the same week, I read a vote in the sense that until now AGI is to 2030 and now it on 2029. And then that... to this a roof. We are still very far away from the current models of AGI and the same time, but two years.

Thomas: How is it, the curve? It was like that it's flattening down, we've a peak and then something like Mithos comes out and you think, okay, we don't anything like it's still up. What is the final state when it comes to It's still up.

Markus Kirchmair: It's going up. You've been for three years that the LLMs are this wall where it doesn't any further, this dead end. And it's not visible Of course, it's when it to the escalation. The training doesn't just need linear more computer or computer power, but needs more more, powerful they are. And that's a topic, but the moment is no brake in the whole tempo. And is always the big discussion about whether LLMs are the right technology for AGI or do you these world models that actually

Thomas: Who was that? Jan Miepler-Kuhn? ⁓

Markus Kirchmair: Be watching through films or through exactly the former AI chief of Meta is a big fan of it. He says, text does not this information that I really create a worldview, real one. Can and yes, is the total sex of the city is in two camps, And my point view, this is a totally unnecessary discussion, because think in the it be combination of several things. So LLMs are also much, much complicated than the ones that been for two or three years. So all this reasoning and what happens with agents and sub-agents and so on in there, that would be with this... very simple description that it's just a parrot that the next word, actually not possible. So would break these theoretical boundaries that you this technology very quickly. And the reality is that you actually, and many in research, are surprised how much information and world understanding can from the texts and so on.

Thomas: And then we have the multimodal, which can be included now, the whole image content.

Markus Kirchmair: Exactly. Where the large language models with diffusion models and so on combined.

Thomas: And also video. Google has video input and this video input is understood. Very exciting.

Markus Kirchmair: Yes, it's extremely impressive. Also with X, example, with Grog, if you now... You can out to every tweet or post with the Grog symbol, is the video real or is it really happening? And so on. You can ask and he really at the video and understands the content quite reliably. It's really impressive how well it works with current content.

Thomas: I'm pretty free man. I think we have to record The new Omni model from Google. You can later, but it's also very exciting. You talk while a film is being I think they it live in If you have a Google server, can that live. But the video language, the AI, basically... directly changed. You have to understand that. Because with pictures it is exciting. And now we the whole thing again, which makes the deepfake extremely relevant.

Markus Kirchmair: What is the current issue the field of It is actually nanobanana in video format. Nanobanana is the benchmark itself, but editing was always very strong. If at Omni, is not the best in video generation. you movements and the physics are much stronger. The editing is very impressive. speak with a director, speak, he understands what you mean. And he sees something moving, it's incredible. But when you yourself year and he it... And what we half a year ago, we short clips where we text or image to video, where we 10 seconds or something.

Thomas: You are to a director who understands what you mean. And when he sees something moving in the it's incredible. If someone had ago, it would been...

Markus Kirchmair: a nice video, but it wasn't more than a nice video. There consistency and suddenly there are more more models that maintain the consistency over long distances and where suddenly not only the generation of the video content happens, but more more the cut. ⁓

Thomas: It's exciting, because it's about recognizing the person again. The faces are consistent. This been missing The storytelling becomes really strong.

Markus Kirchmair: Very extreme. And then there's Higgs Field, for example, with Higgs Field supercomputer. don't know if you've it again. Actually, how do you it? An agent coordinator who end more or less the whole creative process to an idea. It's of a good way to rid of ⁓ Especially at beginning, where you can't really estimate what will out But it's already impressive.

Thomas: You can create a lot of with a low budget. David, the film, was with Flux. Soon there of Amazon Netflix films with a low budget. And then the question is, can I Can I also promote this film And then it's as good as the big ones. I think that a lot of how do you have the right to say, I have a budget of several million, for a film that can at home?

Markus Kirchmair: for a film that can at No, the Higgs Field just a film, which can't at festival itself, but they said, okay, let's to the third floor and rent a cellar and show it there. I'll it on the screen. They said, okay, we have a 500,000 euros, a 500,000, about 400,000 compute, so they have a little bit in and at the time they that it have least 50 million So yes, those are high sums that the AI costs, but if you that in a relationship of the real big films, the big productions, what we often see is small part behind it. When I Making of a Gladiator 2, where I the same coliseum.

Thomas: Yes, it's a high standard. It's a high for big film production. It's like a robot. It small part. How many people were in a short scene? You costumes, have light, you have whole setup, the lights.

Markus Kirchmair: on Malta, the neighbouring country. without a... it's unbelievable. And everything you save with KI has an impact on the people who in this industry.

Thomas: And yet, you people who really an idea of what the picture really has look So that's a bit of danger that KI is a faster, but that often a bit too generic if this taste really in the individual components of a picture. That's still relevant, I'd say. ⁓

Markus Kirchmair: Absolutely. On the big level, possible to one person sing. I get the taste for whole movie. But I don't the 50-ditch and the caterer and the sound engineer and so on. are huge rejections in this industry.

Thomas: This of course a lot controversy and lot of disruption in that respect. Also regarding the process, I recently a lecture where it was about theater. And the theater process is by KI. You don't the set designer anymore, you just with KI. And there you already a first draft. But this process, which also very important to a theater piece, is of course also on the human side. You have conflicts. You learn together and the end you something. And is extremely short-lived. That basically the middle way between the idea and the finished product. This experience that shared with the team, that is very strongly utilized.

Markus Kirchmair: That's extremely expensive. That's middle way. The idea and the finished product. The experience shared with the team is gone. It's to very exciting. Very exciting, because I would have said before, example, when you discussions like what will change, what will be resolved, what will then I would have said, theater is relatively less affected by But it's hear that... But if I understand more the process that more and less the product itself. The actors are still on stage and so on.

Thomas: It's relatively easy to get But it's exciting to how much time you have. If you don't much time and you want to results use the technology for And on the budget, of course. off.

Markus Kirchmair: The stage image is probably there, and it probably doesn't from a 3D printer or something else. Although the KRI theme is safe. There are the stage images or what do you them? But the process before...

Thomas: There is the beat. Exactly. But it is right. Until the finished product is in plan. The whole concept is what generates a lot of This ideation phase, which is very short, to less.

Markus Kirchmair: Yeah, understand. Exciting. And yes, process to end from the I already the whole right that we lose something. The KMT takes lot, but maybe it shouldn't everything away from

Thomas: And yes, process to there from start to end, I give the whole right that we lose a bit of something. Decay simply a lot from maybe it shouldn't everything from

Markus Kirchmair: That's the weird thing I from Google AIO. To outside on stage in of the developers and the creators and from the creative process and so on. And the time, one product after another, the creative process or the programming, tenfold or even more.

Thomas: This is creation, is failing in is simply not necessary ⁓

Markus Kirchmair: It's an tension. It's a struggle between the It's very difficult. Yes, that's right. And especially the ideation and the whole process in of how you collect you design them and so on. can that. I've two years now. So brainstorming and so on. I sensational. We as humans always in our own sphere and ideas that not even conceivable have them and they just create things that might the most ideas and you think, okay, it's maybe a stupid thing but in combination with the other one you wouldn't even think it.

Thomas: We will a lot more. It is different. The question if we extrapolate in the future, will there really new professions or will we lose

Markus Kirchmair: It's different. question is, we extrapolate will there really new professions? will there something new? I will be new professions. The question is how many? Because those who say there have always new professions, can't name the of the AI advisor. Consultation is not a long-term perspective in my opinion. At least what the knowledge transmission is the outcome of knowledge transmission is not a business field in my opinion, no matter what form of consultation. Then may be certain regulated fields, such as legal consultation, where the lawyer has to personally a complaint or something, etc. This remains of certain regulatory. or have for but he might not need his employees in that form. And then of course the question is, if you it a lot of topics, also in many areas where you... There are areas where if you can more output is Where there is an elastic demand for the function on the professional field. And then there are fields where if you can the demand is not greater. If you ten times as quickly with KI, you not 10 times more bills. If you times faster damage reports in an insurance company, you 10 times more damage reports. And that's a certain dilemma because it lot of jobs and areas that are currently being heavily blocked. And nobody can really tell you what the people are Because the AI expert will first... I don't everyone Secondly, training is a relatively slow process compared to the dynamics we see in development. And computer centers are not famous for creating many new jobs. Maybe in the construction process. There is also the argumentation in America that 2,000 new jobs Yes, the next two years. But really, this long-term answer, nobody you the

Thomas: So that the size of the work will be obsolete. That it doesn't need many people anymore.

Markus Kirchmair: I don't know if that's how works, but think there many areas where much fewer people can much more. And that's also when you the last few weeks and months, what's the big companies in America that people, then you always the same wording in sense of we are now an AI native company, agent focused organization. Whatever. I can say again, it's eye-washing and so on. A part of be justified, but the big one is the one that you see and I think everyone who in touch with KI almost notices that individual or small teams can much more. big organizations with very heavy-handed, which can sometimes not really act are actually in the disadvantage. And that's you can big companies that are to how they can slacken. Because if they don't pull and stand slenderly, there is a bit of a risk that they be overtaken

Thomas: And then we have this switch part think the small ones, the startups, they have no big regulatory internal problems because they can so lean that they decide Then we have the big ones, they have a lot budget, they can also invest in AI projects and things that. And think the risk area is in the middle. They already structures They may also a management layer in Then they to all these regulators.

Markus Kirchmair: At the time, structures, which have management layer in between, we all these regulators. Exactly. you have, so far, you about the organization, how you to have organizations then it only about hierarchies. When you big organizations, you several hierarchies, where one was for several, because you 100,000-man operation. not with one guide level. This will change that you no this middle layer and actually now as single person much more with the of AI. That was rather because of this need that you this overview and that. ⁓ Sometimes it's already the technology, if you of Nvidia, Jensen Wang, think he has 82 direct reports or something. He's an exciting CEO, I think the direction is more this direction, that there are really a of direct reports.

Thomas: is that the agents sort of the management somewhere. That they just the information from all the departments and then just prepare as a middle manager would otherwise.

Markus Kirchmair: Yes, because deep hierarchies always the disadvantage, as much they necessary and much justified, as they were in the past, there certain loss or loss of from below to above and vice versa.

Thomas: think that's what AI now. There was an interview where exactly that described. That there is this startup mentality That you stay as flat as in all these hierarchies. Because you say that there is a lot of loss.

Markus Kirchmair: I think everyone who in big company knows these topics. And with these hierarchies, the political within the company which is his kingdom in the company. I think that in many ways.

Thomas: I think that will be much more in the future, that there this pressure, okay, it really me, so what kind of quality am here that an agent not

Markus Kirchmair: What do I to sure I your agent? You'll be amazed how many companies there in Tyrol that now officially hiring freeze or headcount freeze. not totally strict, but you really say, before you a place, you have to justify why this not with AI. And even if it is not now, how is the perspective, can you in half a year with AI? Why should we it? So that's a bit of topic that's already happening. And I don't think that really the attention because of if we There are legal and the labor market being...

Thomas: Yes, mean, you fell on your face. I also saw it now. It at Kleiner, where many were fired and then maybe we still the employees. But I think that was also an early adopter, who was maybe a little too early, right?

Markus Kirchmair: I think that was just your daughter, who was a bit too early. It an example that very often and now more and more often, interestingly, although the example has happened two years ago. And then it was about the customer service. And said, you tried to this customer service with AI too far. and didn't get angry because the software wasn't reliable enough for That means can with innovation or with establishment of Cree company. What I find is to Cree because of years ago the learning is it for next ten years. If go through life notice what in last weeks and months, you would a good feeling about how fast this development is and how much it can that it more reliable. I'm sure if you step, if you would roll the experience would probably completely different.

Thomas: Yes, exciting. We will a lot. And as you say, we will each in America until we see How do you the time horizon until we moments where people are invited when they about KI?

Markus Kirchmair: Good question. there is a ZDF reportage, I don't know if you saw it, it two or three weeks ago. There a survey, what the I think 86 % of the Germans said that they the assessment that AI destroys more jobs or costs than it do. And that's interesting because the image that can be from it is actually one to one the same as in America, although the topic of AI and influence on the labor market a much longer history in America than in Germany. In Germany, it was treated very very carefully. And if it makes us more productive, the sense that the work finer and pleasant, but never goes to where you job fears, then is very interesting that many people feel that. I have to say that 5 % of them have a job of It's more of the big picture that you think is in that direction. It's not the very acute concern yet. But I can imagine that the atmosphere we in America is not necessarily better because the K.I. is not yet so extreme it will be more present. Maybe it's because we don't have online streamable, big talks about universities and that would be there but...

Thomas: Yeah. online street. ⁓ We will experience it. think it will take I think it's important that such moments happen, that you feel where the actual atmosphere is, that there is a possibility to discuss I have this small groups, where it very critically. Whether there such big talks, we will see.

Markus Kirchmair: I have a very critical discussion. Whether there such big talks about it, What I interesting here, we already about Sam Altman earlier, he had this big this last week, this 180 degree turn that he allegedly made in sense of Job apocalypse is cancelled. Not That's interesting because you see that in the tropics and now at the beginning of the week where Mustafa Suleiman is standing on stage and preaching humanistic AI. So no more AI first. Suddenly you don't to What was the big motto Not specifically from Microsoft, but suddenly it's the humane before it. The question is, something really changed since then? Has the situation

Thomas: Well, speech, you have seen that it not so popular and you have to your IPO or the value of the assets that come to the market. I think that plays a little bit it.

Markus Kirchmair: Exactly. Three things that actually happened. yes, these speeches at the universities that really once showed unnoticeable and unheard of. Then a second bad thing. Sam Altman about a month ago, he gave hits on his house, once with the Molledorf cocktail, once shots on the house. What you actually leads to this negative mood. And the time, the reports themselves to demonstrations against certain AI labs, also against computer centers and so on, which boosts the mood. At the same time, the big IPOs. And I can very well that if you're in the ramp like a Sam Altman and you really you about your own safety, so you can about what message you in And my point view, I only guess about it. would explain why a 180-degree change in public communication is Because it is a popular topic, which everyone has because you have to switch regardless of whether something has changed or not.

Thomas: But yes, even if they it differently, it still to be proven. Whether their new performance really into the or if they are just a hooligans who are even more.

Markus Kirchmair: No!

Thomas: Exciting topics. Within one or two weeks, lot of new We will that more often.

Markus Kirchmair: What's happening with Kin?

Thomas: In Kitzbühel on June 18th have a lighting event called Gondel. We have the opportunity to have our E-Rounds in cooperation with the AI Summit, so everyone who has already experienced it, this is the third time. AI meetings also discussed at the summit this time, and we also an event right in the valley at the Kulturcafé. Everyone who wants in the gondola, welcome. It take and we will with Innsbruck That's something we are

Markus Kirchmair: How many seats are in a gondola? Tell us something about Kin for those who don't know yet.

Thomas: We have six seats in any case. These the table sizes we have. And then in the station and then you your people so you want talk a view. For those don't know yet, and I hope there that many we do this every week, every week Thursdays 8 to 9. It's the opportunity to exchange ideas on different topics that AI. So you can AI and marketing, AI and human resources. We also have ethics, research, Agriculture is all part it, where no one ever touched If you register, you'll out that interest And you'll other people who the topic with you.

Markus Kirchmair: from 8 to 9. This is the possibility to exchange ideas on different topics that AI. You can AI in marketing, AI in human resources, also ethics, research, agriculture, everything included, where AI AI. And if you register, you your ideas out, which interests you, and you will them ...other people who want other topics. Do you have other topics That's the topic that should be More people want talk about ethics and have an ethics table. That's a bit of a test. The bio-weapon table. I hope not. We're still in marketing phase.

Thomas: I am interested in how it We have many founders and many companies that this transfer of and course the network.

Markus Kirchmair: Exactly, but that's also the element where we the possibility that it doesn't move. The other one We have lot of founders, lot of entrepreneurs here, who also this knowledge transfer and course the network. You were recently at the Höfe Museum. Right, in Hamster. We any chemistry, but was also a picture of it. It's like a pack of pictures, where it's much more about building Photos are built there. It was open for young and old.

Thomas: Right. In Kramsach we a chin, but a picture of It's similar to a hacker, but it's more about building something. Prototypes are being built there. It was open for young and old. We had a lot of children there. They brought their own ideas. It super exciting for them. There were a lot interactive museum apps. Also through this opportunity to these buildings. At 5pm, you spend time discussing ideas together, first completely without computer. Then a master program is built and then the KIDAs and then exciting apps or ideas

Markus Kirchmair: Cool. Is there else has Yes, the next one is called Heck am Berg. that are built there are being developed. So we actually two apps that have by the director of the health museum. And that's not interesting anymore, it's not nice anymore, but he that. I'm very curious about the team.

Thomas: The next one is called the Heck am Berg. And the apps that are built there are being developed. We actually two apps that were by the director of the Hilfemuseum. he that's not only interesting, it's not only nice. He wants that. I'm very excited about the team. I missed and I hope it be implemented. Very exciting. As I the format itself will a pre-production at Berg. Which Berg? That be announced

Markus Kirchmair: I hope we again. Very exciting. We will a new of mountains. Which mountain? We will Okay. Your Thank you for Thomas. I'm happy to the first podcast together. I'm to it again think there's more than enough

Thomas: We will this week. That's is also the AI week. I look forward to a continuation. Thank you.

Markus Kirchmair: Cool, thank you! ⁓