Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
#12 - Sven Spöde - The AI Transformation: Redefining Business, Empowering People, and Reshaping the Future
76 Plays6 months ago
Ever wondered how AI is reshaping industries right before our eyes? In this episode, I sit down with Sven Spöde, a PR expert and seasoned AI enthusiast, to discuss the transformative power of AI in the business world.
Sven breaks down the five levels of AI development—from chatbots to fully autonomous systems—and we explore how businesses are already leveraging AI for growth. We dive deep into real-world applications, from improving workflows to empowering individuals and small businesses with AI tools like GPT models.
We also discuss the future: What will Level 3 autonomy look like? How are traditional industries being disrupted? And what are the big questions we still need to answer as AI continues to evolve?
Tune in for an insightful conversation about the impact of AI on business, the opportunities it’s creating, and the challenges that lie ahead.
Enjoy!
Recommended
Transcript
Introduction to AI in Business
00:00:00
Speaker
Hey, what's up, everyone? This is Ajmar Savary and welcome to another podcast episode. Our guest today is Sven Sperter. In this conversation, we talk about AI in business. In particular, how it is being used and what is about to come that will affect us all. Enjoy.
00:00:31
Speaker
Hey, everyone, and welcome to another podcast episode. If you're new here, my name's Ajmal. I'm a neuroscientist and entrepreneur. On this podcast, we explore the links between science, technology, business, and the impact they have on all of us.
00:00:46
Speaker
Today, we talk to Sven Spöhrer. Sven is a digital communication strategist with over a decade of experience. Currently, he's a senior consultant for digital communication at Olivaschodt Communication, focusing on AI-driven solutions. Sven's expertise extends to the metaverse as the chief metaverse shaker at Metaverse Monday, and he's previously helped global brands thrive with digital strategies, especially in China.
00:01:11
Speaker
His background in IT, PR and digital marketing makes him a versatile leader in the field. The list goes on and on. All right, enough background. Let's get into it, shall we?
00:01:24
Speaker
All right. Well, welcome to another podcast episode. um You already heard in the intro, our guest today is Sven Spöder. And well, it's been such a long time since we know each other. I mean, I still remember being in the machine learning classes with you when nobody really knew anything about machine learning or AI, however you want to call it.
00:01:44
Speaker
And back then the classes were full of math. I remember they made us to you know manually by hand do the first few steps of back propagation um for the partial derivatives. And that that was painful, but definitely learned something. And well, I can say this has certainly changed to how it's done today. um And to make the switch to nowadays, really, I mean, you've been working in the field of PR for a while now.
00:02:14
Speaker
And the reason I wanted to talk to you is to get a real world understanding of what is actually happening with AI in business.
Understanding AI Hype and Reality
00:02:21
Speaker
um You know, I honestly don't like all the hyperbole that's being used from various camps. And most of all, a lot of people are still left in the dark of what is actually happening right under their noses. a They live their normal lives and it's all fine but you have to hype on the one side and you have to doomsday on the other side. They often say the truth is somewhere in between but that also is not always the case. So that's why I really wanted to talk to you.
00:02:50
Speaker
So I mean, let's start with the basis that we can work through. um The service or app that has made AI be catapulted into the massive mainstream was JetGPT. It has been developed by OpenAI, US-based company. And OpenAI has talked about their five stages of AI development. So can you walk us through those five stages? And afterwards, we will go into detail into each.
00:03:18
Speaker
give the mic to you. Hey, first of all, thanks for the invite. And yeah, it's almost been 20 years I think since we had those machine learning classes. um Actually, a few weeks ago, I saw a website called AI by hand.
00:03:34
Speaker
where you can do all those nice things, including all the steps of back propagation. I had to think of you. I don't know why I didn't send you that link, but maybe look that up. You can go through all the basics of also current models and techniques and ah yeah see how they work to all the nice math, which I sucked so much at.
Impact of ChatGPT on AI Accessibility
00:03:57
Speaker
Fun, fun.
00:03:59
Speaker
Yeah, and why why I love ah those ah supposedly five levels of AI, which are talked or classified internally at OpenAI, because it actually dumps the whole thing a bit down. So level one is basically chat GPT. The cool thing about chat GPT and what made it such a big hype was actually, wow, people can talk to AI using natural language.
00:04:28
Speaker
So it just everyday language. And this was a big thing. and I mean, it's not even two years, but it was a game changer for many people who were not able to set up a AI system, a large language model or others on their computers before and use them. But this easy interface together with use it with natural language without too much syntax Um, really changed something from the, a lot of people were suddenly able to use the system without, uh, understanding that math or understanding how a large language model is trained or something similar. So this is why I think it's somehow deserves the hype. Even though for us, uh, as people who knew know how to code and how to speak and and syntax to machines, uh, this was not so relevant maybe at first.
00:05:22
Speaker
right Yeah, i mean you're certainly right. i mean it was
00:05:27
Speaker
if The systems were so complicated and were safe well I shouldn't say safeguarded, but behind a very strict high high skill level wall, um So it couldn't really be accessible. I mean, my parents never really knew what I was doing when I was saying this. Yeah. And you needed to apply to get access to a lot of models, right? You needed to prove that you're an AI researcher to get, I think, access to GPT-3. And I applied for access. I'm not working in AI research. I'm not working at an AI company. I'm working in a PR ah agency.
00:06:03
Speaker
um I wasn't able to get access. So even I who maybe could have used it from a skill level, I wasn't able to get access to a better of that or to to the model. Right. Yeah, you're right. I mean, that definitely changed. And I mean, we saw the usage numbers as soon as what was it, GPT 3.5 was released to the public.
00:06:26
Speaker
we saw the number of users went through the roof. I think it was but within one month or something, had 1 billion users or something like this. I think it was the same. The app which fastest had 1 million users, I think the number was. But but maybe let's look that up again. But yeah but but it was from the adoption rate. It was the fastest ever. So I mean, that shows that I think it somehow deserves the hype.
OpenAI's Five Stages of AI Development
00:06:54
Speaker
But of course, the the media is full of ah topics, artificial super intelligence, sci-fi stuff. Yeah, well, this is maybe, I mean, hey, I can talk to a computer using natural language to AI system that's not yet there. So this is why I think
00:07:12
Speaker
Also, these five levels are actually helpful. And just to briefly mention them, maybe what those five levels are. It's level one is the chatbots, so chatgbt. Level two is reasoners, so AIs who have human level problem solving. Level three are agents, so systems that can actually take actions on their own, given a mission. Level four are innovators or innovative systems. um What does this mean?
00:07:40
Speaker
To be honest, I don't know yet to be exactly where the border to agents is. So they would probably do something and come up with their own mission or ah I don't know exactly. So that's the next step and it would be a different or level three and level five would be basically organizations. So AI is one or many AI systems which take over maybe a whole industry or a company organization of tasks or missions.
00:08:10
Speaker
um Yeah, maybe maybe we talk in detail about level four and five later because that's really far out there to grasp. But I think this makes it more concrete. And I think there are many cases which are discussed in the media, but also in business.
00:08:25
Speaker
where then people always come up with creative ideas for AI. um You're talking about chat GPT and how you can use it and they come up with level three or four or five stuff and I'm like, yeah, sorry, but no, that's not possible because it's level three or four or five. um It helps me in arguments and also to sort use cases people are discussing.
00:08:47
Speaker
um just to see where are we? What kind of system are we talking about? I think there are these classification is helpful. If it's still helpful in a few years, I'm not so sure. But for now, it works for me. Right. No, I think the classification and and I think it's really good that they talk about levels, because you can't have, let's say level five without having a solid level one. Right. That's it's it. It doesn't work that way. um And it's a build up that happens over time. The complexities and the levels of abstractions are just just like a pyramid. It's building on top of the base layer of the other. so and after you Now you mentioned all the five levels. um Level one, you said chat bots and it's basically chat GPT.
00:09:36
Speaker
where Where do you see, I mean, I use it quite a lot just because it's quite, it it saves me a lot of time um for sure. I mean, it certainly can write code faster than I can. ah The good thing is I can still check if the code is correct and if it isn't, I can just tell it to, you know, change a few lines here or there or variables, or if it made errors in connecting modules or things. So I'm not necessarily using it to come up with solutions.
00:10:06
Speaker
But I mean, it it just can type faster than I can. So for that, it's quite helpful. um Where do you see you know levels of chat GPT and let's maybe say chat GPT level four um being used in business nowadays already?
00:10:28
Speaker
ah You mean level four of these five levels where that is... No, no, no. I meant chat GPT model four. Model four. Okay. The model four. Okay. So there we are still at level one. That makes it easy. Yes. Yes. At level one. So actually what I like to... um So if if we talk about AI in business and adoption of gen AI and level one chat GPT kind of models, um I think it's also important to look at the work a lot of people do. So, I mean, basically shifting one content to the other, reformatting, rewriting content, doing adaptation of content. You had one base content which somebody came up with and then it's rewritten and rewritten in different forms as a PowerPoint, as a report.
00:11:16
Speaker
As a social media posts for different channels as a newsletter, then you update a website text with it. There you have a lot of tasks actually, which are done in corporations, which are level one jobs, basically it's content rewriting.
00:11:32
Speaker
so And all this is now heavily being um augmented or replaced by GPT, GPT-440. I think everybody is starting to adopt it. I think, ah I mean, you talked about the hype earlier. There is a hype and there is definitely, let's say, an American tech bow hype. I mean, we saw now how much money OpenAI got in funding and how much the valuation is. It's just crazy. I mean, we're talking about hundreds of billions of dollars there, right? So it's insane. But on the other hand, um I guess every big corporation or medium sized enterprise, they are at the moment implementing an internal GPT solution, either with the API, OpenAI or maybe open source models, which they self-host.
00:12:24
Speaker
Even all other models hosted on a hyperscaler or are they are just using the chat TVT enterprise. So, and this is a huge market and the adoption rate is crazy. So using
AI Augmenting Productivity in Business
00:12:37
Speaker
the transformation to a PC.
00:12:39
Speaker
Auto Internet wasn't this fast. I mean we are talking less than two years since chat GPT and um I mean GPT-4 was released April last year. So it's one and a half years and the adoption rate of this is crazy. So again there I think it deserves the hype up to some point.
00:13:00
Speaker
There are a lot of expectations for the future where I'm not 100% share the hype. But ah when I look at my clients, if I see to other industry where I talk to companies, everybody's looking at adopting and implementing training their staff in it.
00:13:18
Speaker
I mean, we trained everybody who had time and was interested. So because it's just relevant for so many people. I remember when I had a meeting with a client who wanted halal food and ah the our back office her staff, ah she basically put menus into eligibility and let her check it if it's halal.
00:13:40
Speaker
And I was so proud at the moment that we also trained the back of his stuff. So people were like, why do they need it? They are not creating content, but ah they can also use it in their work for things like this. I was like, okay, yeah. It makes sense in a lot of places to put these this and additional intelligence to work.
00:14:01
Speaker
For sure. And I totally agree with you. i mean it just says It helps me writing faster code, um not necessarily anything new. It's these little mundane tasks that just eat up so much of your time. um and and And for most people, it's most of their time, especially when it comes to content. i mean it's you're you're You're much more in that field than I am. Repurposing your content reformatting it to you know the specific platforms where it's supposed to be shared. It's an insane amount of work. and Creating the base from the start
00:14:42
Speaker
I would say it's even the least amount almost. As soon as you start to try to publish it around, it takes so much time. And if you have these little tools, I mean, I really see it as a tool. um That's, I think that's what many people should actually look at it. It's it's a tool to.
00:15:01
Speaker
to augment your productivity if you let it. But of course, you need to understand how it works. yeah But I mean, it's ah not now you and me, we are not content writers. so Yes, that's true. But we are now able with these tools to write or create content, which is on a quite good, quite okay, maybe even pretty good level. And this is insane on a certain level. i mean there ah hundred thousands of people in Germany alone, which made used to make their money and their living from being able to do this efficiently. And I mean, being able, we had this whole field of the internet, social media, content marketing, it's a whole industry of agencies of people and companies and marketing departments and communication departments.
00:15:52
Speaker
who whose job it is to create content and rewrite content. And now, ah given these tools, even people who are not very good at, let's talk about myself, at creative writing, are suddenly doing stuff which ah challenges them. And I'm not sure you saw that yet last night. or I published chatgbt Canvas, a new function, which basically means before you had to you've got a text,
00:16:21
Speaker
And then you had to see, hey, please write it, make it, put this point in more strongly, rewrite it in this and this tonality. And now you can basically click on a text part and just edit this text part. So you have basically in painting, which you had an images where you could say, hey, my journey only changed the head of the guy. And here's a form for that, make it a,
00:16:46
Speaker
draging ahead ah You can do the same for text now. You can click off on a piece of text on a few sentences and say, hey, please put in this fact here. So you can don't have to get the whole text updated as before, but you can only change some lines of text easily. with It's basically an editor.
00:17:05
Speaker
And this is for writing and you have the same for code. So yeah when you before you got code out of chat GPT, you said, Hey, please make sure this and that and he'd rewrote the whole code. Now you can click on the lines of code you want to change and say, Hey, please put in this and this, this there.
00:17:21
Speaker
I don't forget forget to do this API call or whatever. right So you can just change lines of code, which is again a game changer. I think overnight a few thousand freelancers lost a substantial amount of their ah yeah of the jobs.
00:17:39
Speaker
Damn. Okay. No, I haven't seen that. So thanks for letting me know. I will, now it's going to make my writing code writing even faster. that's ah hello Let's see how long I will write any code for the time coming.
00:17:55
Speaker
And I mean, I mean, we were talking about this hype, right? I mean, they can update a UI, they have an UI update overnight, which really affects people's living out there. So this is why I think this deserves to be called a hype, which has some substantial basis. But when we then look a bit further into the future, I'm not so sure anymore. Yeah. So, so this is already very good
AI's Role in Efficiency and Job Impact
00:18:21
Speaker
to know. I mean, a lot of these tasks that, that many people that many people are doing on on a day-to-day basis, they're being, well, for some things are being sped up and for others, well, it gives them a tool where they can challenge actual professionals without needing to hire an actual professional. So that's ah so that's already quite something and we're we're still at only level one.
00:18:51
Speaker
what ah What is happening at level two, the human level problem solving? Yeah, I mean, here also again, um I'm not so sure if you can maybe call some models from Entropic, from Claude already level two, but for sure this chat GPT-01 preview, um which they published a few weeks ago, is this. So it could do math,
00:19:19
Speaker
problems which we also had in our math class if I remember correctly. um Something with ah what age are the princesses if they are half the age and double the age and so on and ah it could come up with the solution much faster than I was able to read through the task.
00:19:39
Speaker
so um And this is insane. I mean, this is again doing something which I'm not able to and in a lot of domains, including math or physics, but it can solve problems there which I couldn't solve and it's on human PhD level. And this is crazy. And to maybe give you an example and also to, um yeah, you spoke earlier about those levels augmenting each other.
00:20:06
Speaker
So it's they're not replacing each other so for example if I'm preparing a presentation for a keynote I'm giving next week at a conference so basically I wrote down my thoughts I put it into in that case I actually put it into Claude and that come up with a presentation structure.
00:20:25
Speaker
Then I copy that over to chatgbt and let 01 preview challenge this presentation and argue and do something more out of it and ah boil it down to a hypothesis. And then I take gbt4.0 and let it rewrite the presentation because 4.0 is still better at creative writing and using language.
00:20:50
Speaker
So I'm using basically the strengths of level 2 and 1 in combination. You can change the model in chatgbt, so you also have to think which model is good for which task. So it's not that this new level 2 model is better at everything. If you do content creation and whether it's new novel idea or reasoning inside, I would still prefer a level 1 large language model than to than this 01. And yeah, this this is pretty cool already. So to be honest, I felt dumb, because I had access to the model. And then I was like, Okay, let's try a few things. And I was like, Okay, how many of the tasks I actually have to do every day are reasoning, it's not that many. ah
00:21:34
Speaker
Well, that's true. Yeah, yeah, yeah. ah and and And I think what's also really cool is, I mean, the O model, right? The GPT before it's O1 preview, the GPT for O standing for Omni. I thought that was already crazy because as you mentioned, you know, when GPT, chat GPT came out, they had this interface, you know, you could ah communicate with this.
00:22:03
Speaker
model via written or let's say type text. And now when they came up with 4.0, you could talk to it. So it's it's not only written, now it's also audio processing.
00:22:19
Speaker
And you could also provide it with um image input and it would you know combine the context of e which whichever channel you wanted, it communication channel. And to me, i I didn't really know what to say. I thought that was insane because you know what? We were doing all of those things.
00:22:40
Speaker
each of those channels by itself it hard was very, very hard and different. there There were different approaches and all of this. So, I mean, the transformer-based model is just... i Yeah. um How would you say it in German?
00:23:04
Speaker
That was quite something, I must say. Yeah, it's really genius, right? So it's really cool. And I mean, this I don't know if you had a chance to test advanced voice mode yet. So no we I was able to use it if one day using VPN and then they blocked that.
00:23:23
Speaker
So, I mean, it's really detecting the human emotion tonality in your voice, right? And it's depth to depth. So that's a further, not only like using it to do speech to text, but it's also annotating it. If you say it in a certain tonality, if you have a certain base ah emotion when you speak,
00:23:46
Speaker
This is also an information channel for GPT-40, which is using in the advanced voice mode. And this is also insane. I mean, it can also do this. And this is already now not being rolled out in the EU because ah it's emotion recognition, which we ah prohibit in AI models or need certifications for.
00:24:07
Speaker
And I mean, we're still speaking about level one here, right? ah Talking a natural language with an AI. Already this is on a level which where we get into trouble with the EU AI act and legal. yeah it's it's It might not be legal in the EU to do this. Oh, wow. So we'll we'll get, certainly we'll dive a bit more deeper into the differences there. But do you already see the attempts of level two being incorporated into business?
00:24:37
Speaker
No, I haven't seen it because I mean, some people are testing things, but um I mean,
Potential of Level Two AI in Business
00:24:46
Speaker
as I mentioned, so as in communications and marketing, most of the stuff you do is level one.
00:24:53
Speaker
So and maybe where level two now comes in handy is to come up with new creative ideas challenge them like i did with the presentation but i think this could be become really interesting on a product level on the other hand it will kill a lot of ah companies.
00:25:09
Speaker
I mean, you know, there are a lot of, ah if you if you go to a big, let's say, D-Mexico digital marketing expo, which was last a few weeks ago in Cologne. If you went there a few years ago, half of those companies now don't exist anymore because they solve problems which now either don't exist anymore or solving them doesn't make enough money. So the margin is just not big enough. to Yeah.
00:25:35
Speaker
Yeah, that you really need a specialized company or business offering for this. And I mean, imagine what you could kill with level two. I don't know, but I assume you could put a lot of research paper on visual attention and UI design into it. And you probably could kill eye tracking companies. ah i That's possible. Yeah. I mean, if it understands the research and then you input an image or a UI mock-up, it should be able to give you a pretty good recommendation of what not to do and where to improve. That's true. That's true. um You know, I'm wondering if you know what you said with level one, it's a lot of the
00:26:17
Speaker
um you know marketing and and writing stuff and and rewriting things. I'm wondering if level two, those are the consultants that will go.
00:26:30
Speaker
So I'm not sure that so many consultants really do level two stuff. and let's Let's give them the benefit after that. I mean, that's that's at least what they say they do. Yeah, but but if you look at, I mean, this this also raises a lot of interesting questions about the way we work and structure our work in corporations. I think most corporations are very complex structures systems.
00:26:56
Speaker
But they want to break down the roads in those structures. If possible, they want to reduce every employee to a level one agent. They don't want them to be too important in their role. There only needs to be a certain number of actually people who come up with, who have reasoning or who are actually agents who come up with new stuff in an organization. If everybody would do that, it would be pure chaos and anarchy. That's true.
00:27:25
Speaker
So, yeah I mean, this is how corporations are structured with different roles and functions. And again, I see a big, I mean, I use the same terms and ah as in those five AI levels, right? So it's the, I see a big similarity between how corporations work and you have different roles and expectations to people. And I don't think, I mean, probably consultants have some form of reasoning, hopefully. And I mean, I mean, the ones like, like McKinsey, you know, I mean, those you you hire, theyre you basically hire Brains, that's that's why you that's why you go to McKinsey. you know That's what they claim they are. that They're the high level problem solvers. Yeah, but again, right if you how how much of their time is then level two? So I mean, they have probably data collection research. So yeah this could be called level one, talking to people. So you need to collect information. Then you need to like reason.
00:28:23
Speaker
And then you have ideas how you could transform a company. Let's say you come up with new ideas. This is what you actually pay for. But then and probably 90% of the time is putting this into PowerPoint and holding the presentation and ah doing various round of feedback and adapting them and so on, which I don't think is level two anymore. So probably 90% of their work is again, level one.
00:28:47
Speaker
So coming up with this reasoning part. So this is what I mean with I have a few tasks, which are level two, definitely. And, uh, but then I need to do all kinds of stuff with the stuff I came up with, which is then rather again, level one level. So, I mean, even the consultants don't do level two stuff all the time. So it's probably 10%, 20%. I don't think so. So it's yeah not more than that.
00:29:15
Speaker
That's true. um And you charge by hour and the hour rates, you don't distinguish between level one and two work. That's true. And I would say most of the time they just write emails. Yeah, that's level one again, right? Very expensive emails. And I mean, if you reduce the hours by 80%, I'm not sure ah consultancy would be still a lucrative business model. Yeah, that's that's that's fair.
00:29:43
Speaker
And I think, okay, this this brings us then, I would say, also with your example, already to level three, um agents, systems that can take action. but I would just maybe go a little...
00:29:56
Speaker
little before, because what you explained in the example um that you gave is you use one system to to just drop your thoughts in to make you an outline of a presentation. That output you threw in another system to challenge it. um And that output you took again to another system to make them the presentation for you.
Microsoft's AutoGen and AI Collaboration
00:30:20
Speaker
It's almost as if you have different already different roles ah assigned for different systems out of which you know this system is good at this, that system is good at this. They have their clear roles and tasks assigned.
00:30:35
Speaker
And I would think this almost sounds like agents to me. I mean, there was this this framework published ah about one year ago from Microsoft. It was called AutoGen, which is a programmatic framework for AI agents where you can link multiple large language models together. And each large language model is assigned a specific role.
00:31:02
Speaker
and Just like in a company, you know okay, you're my creative writer, you're my product manager, you're the marketing head, you're the CEO and a short description of what that role entails and what is expected of them. And then you give them a goal, you link them all up together and then you see them kind of funnily talk to each other, you know the the creative writers really producing something that's shifted over to, you know I don't know, the marketing head who checks, ah does this make sense? Does this sound good? ah Maybe give some feedback back to the
00:31:41
Speaker
a creative writer to adjust the thing. Then it goes back to the marketing person, or that I can't say person, the agent. And then that goes to the product manager because they have to make sure that this is actually in the roadmap. Does our product do this? Is this okay to say? And then as a final check goes to the CEO to say like, okay, fine, this is this is good. And when I saw this, I already thought,
00:32:07
Speaker
That's insane. I mean, it makes sense to have, as again, another level of abstraction to build on top. um But you could see just by setting up this type of system, just like you did manually, the result, the the the final result that was produced was massively, massively better than what just a single system would have produced, even if you as a person would have interacted with it. And that just really blew my mind. And nobody really talked about this AutoGen stuff from Microsoft, which was, you know, published, open source, everyone can use it, and you can plug in whichever LLM you want.
00:32:50
Speaker
Yeah, but I thought it was quite crazy. It's super cool. But again, the usability and the access is very restricted. ah You were able to run it, but you are one in a million there. so Thank you. But i I think even tech teams, you know, I think tech teams, they could have used that, but I'm now wondering.
00:33:14
Speaker
In all honesty, you know i've been I really don't like siloed professions you know that ah that that try to um they try to go against progress just in order to to to to safeguard their job, even though, i'm I mean, of course, we're all I would say sooner or later are being affected by this. Um, and there needs to be a conversation about this, but what I've never really understood is people just saying no categorically, um, to safeguard their job. Uh, and I'm wondering if auto gen has been, you know, flying under the radar because now developers ah were thinking, holy shit, this is now coming for me.
00:34:05
Speaker
Yeah, I mean, I think actually, if you look at big corporations, hardly people know what the person in the same department and at the next desk is doing a whole day. Actually, a lot of conversations and enterprises are no idea what that guy is doing, but he is always nice. Nice to have a coffee with him. So I think there are a lot of, let's say,
00:34:30
Speaker
Yeah, roads and tasks and companies where actually people do stuff and it's kind of like something comes in and something comes out, their organizational requirements to them. But I'm not sure about how much value, let's say, is actually created by these roads. I mean, the system, the complex system of an organization company has to create value to survive, to make money. But um I'm pretty sure there are a lot of functions and roles in companies which by itself don't produce this value, but ah which are still necessary. I don't want to call them unnecessary. So they are there for a reason, usually. There are sometimes there for historic reasons, which don't exist anymore. And again, maybe the consultants come in to identify that.
00:35:16
Speaker
But in other cases, there are things which still you need to have for certain situations, for risk mitigation, for compliance, whatever. And to be honest, whenever and I work with a tech team who is a doing things for communications and marketing where I work,
00:35:37
Speaker
um I'm always like, okay, that is what you think I'm doing. No, sorry. There's such a huge gap. And ah I mean, half half of the projects I'm actually currently doing is companies who introduce an internal GPT system. And ah If you just let engineer write a prompt for a PR department, it's one or two lines because data you don't think very highly of a communications work and they think it's very easy. So I saw prompts which are like translate from English to French.
00:36:18
Speaker
And if you work with a professional translation agency, you would write them a brief how they should translate it. You would give them a glossary, so our company uses this word, we use American English or British English. Here is the list of terms. For me, it's our brand names. This is how we translate them for different markets. These are the industry terms. Please make sure to translate them like this and not like that.
00:36:43
Speaker
um So there's a lot more than saying translate from one language to another in the prompt. So as all these details, they they they are not very, let's say, tech teams don't really know these details or the depth of all these roles and works.
00:37:00
Speaker
So I can assume that nobody wants to like hide this, but the people who would have needed it maybe or could have been helped by this, they were not able to access it because you would have to run it locally, I think. right just said so and yeah Nobody in PR and marketing has a ah dedicated GPU, for example.
00:37:21
Speaker
the standard laptop has ah only integrated graphics, you cannot run most models. So that starts, you only have ah graphic designers and coders who have actually a discrete GPU. Then there are already these small differences where actually most people who would need something like this cannot run it currently. So this is an access problem for me, why it's probably wasn't used. Yeah, that's crazy. Yeah, you're right. I mean, there's some thought that it's already in the role description. I just Yeah, I think maybe for me, it's just hard to understand how you can just work all day next to somebody else and not know what they're doing. I mean, not maybe maybe not even be interested in what they're doing, right? I mean, it's like, sure, I have preconceived notions about communications and marketing, but if somebody is hired for a full-time role, maybe.
00:38:14
Speaker
Give them the benefit of the doubt and ask what they're doing. um Yeah, but it could be KPI based, right? If you have a media manager and there's nice stuff going on on Instagram and LinkedIn and you're reaching the KPIs you said, hey, who cares what he is doing half of the time.
00:38:32
Speaker
Yeah, that's true. Yeah, that's true. I'm fine with that. But ah yeah, you mentioned that that system autogen.
Autonomous AI Agents in Action
00:38:38
Speaker
And I'm still wondering if this really would be level three agents. Yeah, that's not my question. Yeah. yeah So I'm i'm not 100% sure yet if the definition is clear to me or if it's clear at all. Because for me, ah agent systems should run. So I like to call them autonomous agents.
00:38:58
Speaker
So I think there are a lot of systems which are agent systems also in level one and level two, but this autonomy question is actually for me the biggest differentiator in level three because they can run for a longer period of time. I don't want to say indefinitely.
00:39:15
Speaker
But from the idea, they should be able to do tasks when I'm asleep. I give them a mission, and they work on that. I mean, the AutoGen, I think it can run a few hours. But at some point, it has a solution and will probably terminate. It won't constantly browse the web for better input and update the solution. So I'm not sure if this would maybe be a requirement for me, or if it was, in this case, let's say, different level one systems.
00:39:44
Speaker
where you threw in a ball and they played with it until they stopped playing, kind of. um But again, right if you have this together with ah maybe a combination of level one and level two, and then you have one agent ah who is running autonomously and starting level one and level two systems, once there is, ah let's say,
00:40:06
Speaker
You have an agent which is listening for your podcast RSS feed. And whenever there is a new episode, ah it gets awake. Then it puts that into a reasoner or first into a level one and says, here's natural language. Please do a transcription and give me the gist of it. And level two is then doing reasoning. And ah then the agent is briefing level one agents again, who are doing copywriting for your social media assets.
00:40:33
Speaker
and which are automatically posted every few weeks on X or LinkedIn. So this would be for me an agent. So you give them the task to do something for new podcast episodes every two days and then repost something every few weeks. This would be the mission and the agent would just do that until you stop it or until there are no more episodes in the feed.
00:40:57
Speaker
So if I get you right, you see an agent more as like, um,
00:41:05
Speaker
how can I say like an almost always ready to go employee who is waiting for things for you to give them if things change, but they just run in the background, um, taking care of things. So I just like it in a team where.
00:41:27
Speaker
I don't know, i'm i'm I'm publishing a podcast and then I have a content distribution team, but now I don't have a team. I have a content distribution agent who'll just take care of it and orchestrate the whole thing um to make sure it's getting everywhere. The content is written by level one, maybe, um and just lets me know, Hey boss, all is good. That's it.
00:41:52
Speaker
Yeah, it it has to be a good content creation team because basicly I mean, you wouldn't throw the ball. You would just upload your episode and then it would know. Yeah. Okay. The job starts. You wouldn't need to notify it. So it would get triggered somehow. Maybe it's a notification by a human.
00:42:09
Speaker
but ah then should know what to do. So basically, somebody press somehow the button is pressed, but it's online all the time and could react to things. So I assume it could be used also for things. I mean, um there is a news article about something and then a crisis mitigation for the company. um Let's say there is a PR scandal, maybe some German carmaker has problem with the diesel engine somewhere.
00:42:38
Speaker
as a few years ago, right? So, yes and then you could have an automated system, which is directly creating content and raking up the right people and ah analyzing the whole situation before it gets out of hand PR wise. Well, yeah, that would be, that would be insane. I mean, I, you know me, I like to automate things away as much as I can, but this is a whole, the whole another level to do that.
00:43:05
Speaker
Yeah, you could finally take a vacation or two. Yeah, they i could I could. I don't think I would, but I could.
Challenges of AI Energy Consumption
00:43:15
Speaker
I guess ah on vacation, I'll be just talking to the agents all the time.
00:43:20
Speaker
But is there is that what you explained with that's how you see agents? Is that also how OpenAI describes their level three? This autonomy and 24-7 always on is, I think, a big part of the definition. And I mean, there were rumors that the next AI levels or the next chat GPT costs not longer $20 a month, but $2,000 a month.
00:43:48
Speaker
And I mean, if you have a level 3, system 1, 24, 7, I think that's quite expensive compute wise, right? And oh yeah imagine, I mean, ah all those statistics now, ah one query for GPT-4 is a few glasses of water and so on, so much energy.
00:44:11
Speaker
I mean, that's already ah insane, right? And this level three is not only, I mean, 20 seconds answering me or level two GP here, this 01 preview, it's thinking for up to a minute. So yeah that's nothing compared to being on 24 seven constantly browsing and scraping the web. I don't know what it's doing. I mean, that's an yeah times thousand ah probably from a compute perspective right and I'm not sure we are ready for that and it will be probably very expensive to run this um yeah I can imagine that this is creating a lot of new issues
00:44:54
Speaker
Yeah, that is definitely going to create a lot of new issues. I mean, what I thought was interesting is Google published, I think it was so one or two weeks ago, a paper um about how they can improve AI performance, their model ah performance, without increasing the size of their parameters.
00:45:13
Speaker
um ah by just making it run much more efficiently. you know It's almost like ah having different levels in there and you're transferring almost like a tree search algorithm, um which made the results better without actually using more energy. And it's it's still under debate. And of course, we will never know of OpenAI.
00:45:42
Speaker
um 01 preview is running this way, but I I guess it is I mean they're all They're all using the same tricks here and there but the advantage is I'm i'm trying to say is so At least there is something being done, not in order to just blow up the models and use more energy, but now to think about, okay, how can we make this also run more efficient? But nonetheless, if everybody wants to run this, it's it's going to take a lot of energy.
00:46:13
Speaker
Yeah. I mean, what what the O1 is probably doing is it's a Monte Carlo search tree and running through all the different solution and then comparing them. That's why it's taking a while. And I mean, probably they have a just the parameter of chess. Don't do this for longer than a minute until you come up with a solution which has this in the score and ah you could run it longer. and Probably they did that.
00:46:40
Speaker
And that could be interesting what those systems then come up with. If you let them run a day or two days on certain problems, they can come up with probably new stuff. I mean, who knows what they could come up with. And there we are kind of in level four already, this innovators. I mean, for me, that would be a combination of this level two coming reasoning, maybe let the search tree go really wide and a big, and then to combine that with an agent.
AI Innovators and Future Missions
00:47:11
Speaker
So maybe it's not longer your, let's say it's your creative copy team who is automatically using new episode content to create social media content for us LinkedIn and Twitter. And then you wake up one day and your level four AI has started a TikTok channel, found some old footage from you and then made some nice things that
00:47:35
Speaker
but know I hope not. I hope not. But yeah, I'm just, yeah, you're right. I mean, that definitely could be, I mean, especially for the innovators.
00:47:48
Speaker
I don't know where this could go. I mean, I think you would have the agents, I mean, talking about building it on top of each other. You have the individual agents, as you said, maybe some you let it go a bit wild ah with some ideas. And one agent is just then pushing it to the, to the what would you call it? The simulate simulator agent. So the simulator can check, hey, is this could this be something? And then um it's almost like you have a complete research team.
00:48:22
Speaker
um running a complete engineering department. The crazy thing is it wouldn't need to be a simulation necessarily. I mean, there would, there would come maybe also already a bit to level five then, especially when we talk about the combination, like the whole research team and doing a whole organization. Why shouldn't these maybe innovators come up with something where then a level five AI organization, that's basically what you described.
00:48:51
Speaker
and says, let's not do a simulation, but why don't we found a dummy corporation limited somewhere and test this out in a country and build up the Boston business model and see if it's feasible and build a new business. um So one example I came up for level five was, um I was at a conference, I hold a keynote at a company in the business travel industry.
00:49:17
Speaker
And there are a lot of companies who are doing the payment, the booking, the expense reporting. And so there's a lot of different companies with different solutions. A lot of layers. A lot of layers. But if you think it even bigger in cross companies, so a lot of companies have travel business, travel spend.
00:49:38
Speaker
Then on the other hand, you have hotels, you have conference venues, you have speakers, you have caterers, you have taxis, you have airlines. So this is a huge industry, the travel agency industry, we're talking about billions. So and why not have an AI organization between these talk about things like, oh,
00:50:01
Speaker
Our companies still have a lot of open travel budget this year. And ah other AI systems are, oh, our hotel hotels in, I don't know, Las Vegas are underbooked in November. And others are, hey, look at the mad tech industry in Germany.
00:50:17
Speaker
take They could be in for some innovation and then the AI systems organize a business conference and invite all the important players there. um They optimize the business band because the airlines and the hotels had good pricing.
00:50:34
Speaker
They ah create value for all companies and players involved win-win-win on all sides. So this is how I imagine level 5 to be like orchestrating something which maybe humans in the individual companies couldn't see because they're like forks in a rail or something. yeah So they cannot see outside of their entity.
00:50:56
Speaker
And ah these levels could orchestrate something which then um solves things or problems nobody knew were there or create opportunities which weren't there before. And this is, yeah, I mean, this is not now artificial super intelligent sci fi. But if this would be done, it could create so much value for that industry of, and and everybody involved for the, even the cab drivers, probably if they would be auto numbers by then, but um yeah, you see where this could be going.
00:51:35
Speaker
Yeah. And we're not even talking about, um We're still only in the, I would say, in the cyberspace realm. We're not even talking about things being being affected by it, by an effector in the real world, like a robot or anything yet, which of course is all happening in parallel to all of this stuff at the same time. And I think what's most important here is You don't, I mean, to get to level five, let's say, we don't need to have artificial general intelligence. We don't even need to get there. It's just a different, it's just a different level of abstraction of all these different, what would you say?
00:52:19
Speaker
layers. I mean, if you have level two and three and combine them, you would probably already have level four. And if you have level four, you could combine again, the layer one to four and have level five. So this is how I think it's not substituting. So once you have level five, you still need level one to four. And there will be probably still be advances in level one AIs.
00:52:43
Speaker
for the foreseeable future. um And especially, like you mentioned, I think running them more efficiently. I think a model distillation is a big topic. So where you can basically reduce them to a much smaller parameter space, and then you can run them on devices, on smartphones, and so on. Maybe even on smartwatches, maybe in smart glasses, maybe then in your brain at some point in time. Let's see.
00:53:09
Speaker
and Man, that's insane. I mean, it really is crazy.
00:53:17
Speaker
I would say, yeah, do you know, I mean, when it comes to level one to five, what timeline did open AI talk about? They they never really published this. So it's supposedly an internal discussion. um I have no idea, to be honest, if you look at the timelines,
00:53:37
Speaker
um It's accelerating, isn't it? It's accelerating and usually I'm conservative when it comes to these timings because as you know, those are very hard problems. Of course. Yeah. But then I didn't expect us to have have a model like 01 preview this year in 2024. I didn't expect that so soon.
00:54:00
Speaker
um And same with most, right? Like you mentioned, GPT-40, at some point you always think, okay, now it's slowed down. We have chat GPTs in the world. Now we can start to integrate it. And so what do we do? How do we do it? How do we train people? And then suddenly, oh yeah, okay. Now it works with vision, voice, nevermind. Okay. Next. I mean,
00:54:23
Speaker
Sometimes you get bored and say, hey, nothing has happened for a few months. And then suddenly one model is dropping after the other. I mean, we talked about text or language most of the time, but you have also in diffusion models, you have this huge development with Flux suddenly, and you have a German image model company, which I didn't assume would be happening still. It looked like it's California-based business model and suddenly you have players in Europe as well, which nobody ever heard of before outside of the industry.
AI's Influence on Company Practices
00:55:01
Speaker
Yeah, that's true. So, I mean, irrespective of how fast the transition happened, I mean, that the development from level one to five happens. I mean, how fast do you see the transition happening within companies? And you know, you said it's going to be
00:55:18
Speaker
Level one, level two came out, people were trained. They were told, look, you can use these tools for this and that. Of course, that takes time. And I'm just wondering, I mean i guess, where the humans are now, the bottleneck.
00:55:33
Speaker
Yeah, kind of. like and And I mean, even when it comes to you finished with the training, you know, everybody is now, everybody in your staff is up to speed. You spend, of course, you've also spent quite some money and time on this to then down the line save it because they're now much more efficient and effective.
00:55:55
Speaker
But now, two months later, and you actually have to restart doing this because something new dropped.
00:56:04
Speaker
so let's Let's see it from this point. I think there are different transformations happening at the same time, also with these levels. So you have the level one transformation, which basically is a new tool. And a lot of people can use this for their work and be more efficient, ah reduce certain parts of their job, ah the the hours, and hopefully you can do more clever things in the meantime, be more productive, maybe come up with new things, and so on. um So this is, let's say, a good thing. So if not, too many people lose their job at least.
00:56:41
Speaker
um But at the same time, you have a lot of business models which are then under ah pressure or you don't have the margins anymore. I mean, if you if you had a business model creating content for Google, let's say, um Google is less and less relevant. The more people spend time also in LLMs and social media. So the value of this decreases now.
00:57:10
Speaker
and the work you do can be more easily done using chatgbt and you don't know need a specialized agency for this anymore so it could be rapidly filtering this huge industry i mean i don't know there's a i feel there's an sao agency in every germ town at least as so it's a huge it's a lot of people and People working there with chat GPT are much more efficient. But if they are so efficient, is this the enough budget to have a margin? I'm not sure. So this is what I mean with there's a transformation, then even if they are more efficient, it probably is not a sustainable business model suddenly anymore.
00:57:52
Speaker
And on the other hand, I mean, as you mentioned before, you can now do a lot of things you weren't able to do before. So freelancers and micro companies could actually take a bigger ah role in our economy because a lot of business models could be now possible, which were not feasible before on the other hand. so Right. what ah um And I think this is a really great thing. I mean, it enables you to do so much more. I mean, do you have some some side projects where you already make use of this to augment basically your your output? I mean, i'm I'm thinking about these points I just made for quite some time. And of course, I now decided to try it out myself. So I always wanted to write a book. So a really clever book, but this is not what I did now, but I started with something low content.
00:58:47
Speaker
So I feel that actually, if you if you also look at Instagram or maybe also TikTok, a lot of content there now is, let's say, coaching, self-improvement, everybody wants to become better. I mean, you said earlier, how can we humans keep up with this evolution of AI with the rapid progress it's making? And I feel a lot of people are trying to ah get better, get there, sort out their life and so on, self-organize better.
00:59:16
Speaker
And this is a huge need. So I decided why not do something like a habit tracker planner. I'm working on a meditation journal. And then I augment that with a custom GPT. So basically people who are using this these books, they can fill out, they can track their that's say weekly habits, how they change, how their sleep quality improves. That's what I've been doing the last year now.
00:59:40
Speaker
So and basically with my girlfriend, we are going every Sunday to a cafe and then opening our trackers and ah yeah, like having some check in where we are with our habits, with our progress we are making and the different KPIs we set ourselves also discuss what is up the next week.
00:59:59
Speaker
And ah this together with a custom GPT integration where you basically can take a photo. As you mentioned earlier, you can take a photo of something and upload it. And then it's giving you recommendations how to improve your sleep, for example. Or you're trying to learn Spanish and you write in the progress you have made. And then it's recommending you, hey, why not watch a movie in Spanish next or something.
01:00:20
Speaker
So yeah, and this is something ah which is now possible for me to do on my own because I'm using all the nice gen AI tools. If I would have done this a year ago or even half a year ago, I would have probably needed to hire somebody. I couldn't have done it on my own. So I so find this idea of doing something which ah took a substantial amount of labor before and just do it um I don't think the margins are high enough for me to live on that. It's just a side project at the moment. And maybe also an experiment where I try out new AI things and see how I can actually use them for a very small business that's publishing and just see where it goes and also to learn something from there. I mean, with all those tools, you also need to assess what is the value they create just because you now can generate
01:01:20
Speaker
Yeah. Thousands of pages of text in an hour, make them into a podcast with a notebook LM, create a lot of images for that. You create a lot of content, but what is the value in this content until somebody buys something because of it or pays for it? So this is, I think, the big question. And this is why I'm doing the side project to try that out. and And so it's a bit a research project rather than a business.
01:01:49
Speaker
But that's pretty cool. I mean, you know, when I was doing my PhD, we had this device, it was called a robotic manipulandum, where you could do um reaching movements and you could measure forces and things. And the advantage of this device was you could set up experiments insanely fast. So if I dreamt something up today,
01:02:15
Speaker
I have it coded up tomorrow and have ah my first participant the day after tomorrow. And in the research world, that is insanely fast. And yeah the danger was, well, now you're just going to do stuff. i You're just goingnna just going to make new experiments, experiments until something something works and then you publish a paper and off you go and you rinse and repeat. And I never really thought about this much until I was at a conference and then there was a professor who actually said like, look, you have this tool and it's amazing, but now your job is to think really hard of what
01:03:01
Speaker
to actually pursue. you have you Your job is now most of the time to think about what to do, not how to do it because the how to do it has been made so much easier now for you.
01:03:15
Speaker
He said basically what I think what he tried to say to me in a very nice way was, don't produce just more noise, please.
01:03:26
Speaker
And I think i think it's it's it's it's the same with what you just described. Sure, I can provide content. Sure, it's written text. What is the value? of of all of that. So maybe you should think about that first and not just produce more stuff for the sake of producing more stuff. Yeah. I mean, I can still remember when we had this discussion when I think it was the end of 90s in Germany, we wanted to transition to a service economy that was like, we cannot survive by all cutting each other's hair.
01:04:03
Speaker
so that you need to have some productive part of the economy where you actually produce new things, um then people earn money and cut each other's hair with that ah and pay for a haircut. But we couldn't just all do service for each other because somewhere we need to have, I mean, we would be basically exchanging money, but there would be not an increase in money in the economy.
01:04:29
Speaker
so And I think the same thing is now with all the content we can generate, right? It's so easy to go generate a lot of content. But is there value in all this content? How many AI-generated books can Amazon take? And ah how many pages of never-read stories are there out there now? How many images on mid-journey?
01:04:52
Speaker
ah probably not even the person who created it ah took a second look at. So as you mentioned, that's a lot of noise. And I mean, this is kind of like also content pollution a bit, right? And um oh yeah if you then train you AI on this, that's gonna not be a good good thing. So how do you how is your How's your industry dealing with this? Because I can imagine that at the beginning it was, oh, great, our work is now made easier. Well, and sooner or later, everyone is going to be a content marketer or content marketing company just run by AI stuff. how was your How's your industry trying to deal with this kind of a dilemma?
01:05:40
Speaker
Yeah, I was on a panel at New Mexico a few weeks ago and I feel like many are still in denial. So, all so I mean, everybody's kind of seeing it coming and everybody
AI Reshaping the Content Industry
01:05:52
Speaker
is, again, I mean, you mentioned that already, everybody is now training their employees in it, also in agencies and content production. um Everybody should be training in it.
01:06:04
Speaker
And I mean, this is level one, right? As I mentioned, and then you have a lot of these companies which won't have sustainable business models anymore. Because if you charge by hour, which most content producing agencies do, you will write one you will you will charge less hours after time. So you need to increase your hourly rate or you need to look at different revenue models. It should be a content piece, pay by content piece or pay by to a subscription based model. So basically everybody is probably looking at business model changes because this paying by hour wouldn't work out in the end anymore. I mean, also if you look at the, let's say, I'm not talking only about my industry, but if you talk about, if you look at a lot of industries in Germany, the work satisfaction rate is not very high. So let's say like this, there are a lot of unpaid over hours
01:07:01
Speaker
um I mean, ad agencies are even worse than PR agencies for this. So there's a lot of ah young people to spending an insane amount of time producing things which companies do not pay enough. This is the reason for unpaid over hours because there's not enough value in it that there's a sustainable business model already.
01:07:25
Speaker
And the big question is then what are companies actually paying agencies for? What are they paying freelancers for? What are they doing in-house? Is it outsourcing workload and content creation? Is it genius ideas? Probably if you have ah genius creative ideas for marketing, you should be still able to charge a lot of money. You probably won't do that by hour.
01:07:48
Speaker
um But if you have an idea which can change a multi-million or billion dollar company, this should be worth more than 100 or 200 euros an hour, to be honest. um And I think this will change the way we work, not only in my industry, but I think across the line, across all industries.
01:08:09
Speaker
on Yeah. So basically at the core, the model, I mean, the hourly rate model still comes from the industrial revolution, you know, where we worked in factories and stuff. And that doesn't hold already for a while anymore. yeah But I think now.
01:08:30
Speaker
It's really in your face. There is no denial anymore. It was like a leaking water bucket and now they took the bucket away. Yes, exactly. exactly
01:08:42
Speaker
hi and um Well, what can companies do, right? So let's say you are a company and you want to transition. You want to incorporate these things. um You want to incorporate these tools. What can they do? Except for, of course, train their staff.
01:09:02
Speaker
Yeah, to be honest, I think there needs to be a more clear discussion about maybe these five levels are also good there. So about the kind of work which is being done in companies. So often when I speak to companies and look at, so explain to me what do you do? Okay, why do you do it this way? Why don't you do it another way? Okay.
01:09:24
Speaker
And then you try to understand why and how people work. There are a lot of inefficiencies there. And I mean, earlier we said level one is cool because you can talk to large language models using natural language. But then people came up with this notion of prompt engineering. So it sounds like, okay, you need a new skill to be able to do this.
01:09:45
Speaker
But actually, if you look at this prompt engineering, it's just good briefing. So it's basically be clear in giving instructions, be clear about your own work, be clear in what information, what context do I need to give somebody or something in the sense of a large language model that it can actually do this work. And I think if you look at a lot of companies and roles in companies, people are not that clear about what they're actually doing and what information you would need to do this job. And they're not good in briefing. I mean, I'm working in an agency. I get client briefs all the time. They're all super good. Almost.
01:10:30
Speaker
I bet, I bet. so I mean, the joke is you always should know better what your client wants than the client, right? so it's That's, I guess there's there's not only a grain of salt in there.
01:10:44
Speaker
Yeah, so I mean, again, this prompt engineering is for me only be clear about the role you are having, what ah role you are assigning information and a task store and so on. And um I mean, companies need to get clearer about this. And of course, you then need to from go from there. So who are the employees who need these new tools and train them? So my recommendation is usually train everybody.
01:11:12
Speaker
What also do the people who are doing this or that you're working in the back office or um I wouldn't go as far maybe as the cleaning lady because they're often not internal employees. But I think there are a lot of also that's an organizational jobs where people rarely use a computer.
01:11:32
Speaker
um Which would profit from this i mean if you have chat gpt vision and you need to repair something, um maybe it's not hundred percent there yet but it could be i mean i don't know my vacuum cleaner broke down i took a photo and it actually came up with something so it's.
01:11:51
Speaker
I mean, I think I used in that case perplexity, not chat GPT. Um, so this worked quite good. Then I got the manual and knew how to fix this. Um, so, I mean, I think this kind of system could be useful for everybody. So my recommendation is really train everybody in level one AI, how to use it and then see, do they need to use it on their laptop, desktop, or even with a smartphone, if they are not doing having a desktop.
01:12:17
Speaker
and I think this would be good for everybody, but then I think the let's say transformation people, the higher management roles, they would also need to understand the level 225 and also now willll start looking at the impact for the business model.
01:12:34
Speaker
and ah do something about that. um As I mentioned before, it's not just enough to train people in level one because it's simultaneous transformations happening ah huge where huge value chains in industries are faltering and business models are not sustainable anymore.
01:12:54
Speaker
yeah Yeah, and I think what you mentioned about good briefing, no matter which level, that's a prerequisite for whatever you want to do. let's Even when you work with other people, the more concrete you are and the more information you can give them for what you want to achieve and what the output should look like.
01:13:15
Speaker
That's super important. It shouldn't just be, as you said, with translate, translate this text into French. I was like, sure, then you get something, but that's, you get something. You can't really be mad at the output really, if it's not to your liking. I was asked about an obvious advice, what to do with the whole AI thing. And I said, start meditating.
01:13:40
Speaker
So because I think the huge point is you have to be more attentive, mindful, reflective of what you do, what you want to achieve. And so you again, also there we have to look at how can I become better at at thinking, at whatever I'm doing to be able to interact with these systems.
AI's Impact on Mental Workload
01:14:05
Speaker
I mean, I wouldn't just recommend it for this. I think it's in general a good idea. ah It's also a good thing from a mental health perspective wise, but I think having this mental clarity is ah will become more and more important.
01:14:20
Speaker
So, I already assumed for quite some time that, I mean, if we look five or maybe if even a few years more into the future, maybe it's even earlier, people will also talk a lot also about mental enhancement, probably taking some even more microdosing and other substances to be able to keep up with those systems because it's also mentally challenging.
01:14:45
Speaker
and's um I have nowadays where I'm using so many AI systems on such a high level, reasoners, where it's too much for me. I'm burned out after a few hours of ah working on that level. When I said only a few percentage of probably consultant work is level two reasoning. I also mean that its and you you can't do that any longer without having a burnout. So it's mentally challenging. It's a lot of mental work.
01:15:14
Speaker
And then jumping from topic to topic with high-level AI, wow, probably people will come up with all sorts of things to be able to do that. That's just how humans work, unfortunately, right? Right, right. I mean, we're we have to sleep, we have to eat, the AI does not. Yes. But, and I mean, you already mentioned a few things, but which other pressing questions do you think still need to be figured out when it comes to all of this?
01:15:44
Speaker
I think the big question will be also so for politics on society. So what is happening? How fast these things are happening? Can we adapt? So there will be potentially be thousands of people who work in companies who don't have a substantial business model anymore.
01:16:03
Speaker
So I have friends who run restaurants and they have now a lot of people who apply there who used to work in agencies doing web design before. So and it's people with graduated degrees, they have a masters.
01:16:18
Speaker
They were designing websites and ah now there's not such a huge need for that anymore and you need less people to build a website. but So a lot of people apply for waiter positions.
01:16:34
Speaker
ah with a master's degree now so that's crazy and this is just the beginning and yes these are the questions we would need to ask ourselves and of course there are people who would say no this is not a good thing people should still work their eight to nine hours a day and I don't know if that's a good idea to work that long every day. um yeah And as I mentioned, there's such a big job unhappiness. Maybe it's a good thing we talk about the way we work and also the impacts on society this has. And we need to do that rather sooner than later. Otherwise, yeah it will just happen have happened to us.
01:17:14
Speaker
Yeah. And I think this brings us back to maybe the EU AI act. as ah
EU AI Act and Innovation
01:17:23
Speaker
What do you think? Do you think the EU jumped the gun with wanting to be the first with the regulation and maybe, well, went a bit too far or didn't think too too concretely about these things?
01:17:40
Speaker
Let's say like this, I'm a big fan of the JDPR. So everybody hated that when it came out because it affected every company. Everybody had to do something. Everybody, I mean, i'm I still get annoyed if I have to click all those cookie banners and so on. But in general, it's a good thing. A lot of ah privacy regulations worldwide are now modeled after that. So I think it was a good idea to be early here.
01:18:09
Speaker
And um I think what is not being discussed enough is at the moment, it looks like it's really stopping innovation from reaching Europe. So all the open source meta AI models are not really.
01:18:25
Speaker
ah Yeah, I'm not really here, ready for download and so on. um If you look at WhatsApp, some maybe most boring app on every phone here in Europe, um outside of the EU, it has meta AI built in, in the EU it doesn't.
01:18:44
Speaker
Apple doesn't bring out their Apple AI features in Europe just at a later stage. This advanced voice mode and chat ability, I was able to use it with VPN. This is now blocked. So they were also didn't release it yet in the EU. So it's stopping innovation or let's say, especially American tech company products reaching us.
01:19:07
Speaker
So is this a good thing? I'm not sure. Is this a bad thing? I'm also not so sure. I think it's too early to tell. I think it probably will need to be adapted. I hope they are able to, um I mean, it was kind of a really long discussion and they only put in generative AI at the very last minute because they didn't see it coming. So yeah. They already planned the EU AI before and it was actually not made for generative AI. They only put that in and the and I think the last draft before it being ah approved. So I think they need to revisit that and maybe change a few things.
01:19:46
Speaker
um If we look at how things are developing, I think it would be in the big interest of Europe to do everything possible for a big open source AI scene. So they should rather support meta, which they probably hate because of her everything which happened with social media and scandals there.
01:20:06
Speaker
um But this would be still better than all leave it in the hand of a few tech players as actually now look at your at the systems we two are running. It's all by big tech as big tech. And I mean, it should be in our all interest that open source models are there or actually, I think open AI should publish more research about their models. I think it's dangerous.
01:20:32
Speaker
Same as I always had the opinion that closed source software is not a good idea. um Yeah, but ah I think the EU, it's good that they were fast for once, but I hope that they can also adapt that, fail short of what they wanted to achieve. So, but let's see. You know, I think what was good.
01:20:55
Speaker
and Not to, I mean, I'm always quite critical of the GDPR and the UAE Act. In fairness, um I think the GDPR had an advantage, which was it had to regulate business practices or data practices that were already taking place for quite a while. So they already had a good idea of, okay, no, this is not okay. You can't do this and this should be like this.
01:21:25
Speaker
But now the regulation came at the EU AI Act, the regulation came in before practices you know established themselves. um I'm not saying that's bad. I'm not saying that's good, but it makes it a bit more difficult because you you're walking a bit more in the dark. As you said with Gen AI, you can't see what's coming, especially if technical experts are not in those committees, especially then um because they need to be. And I think.
01:22:01
Speaker
yeah Sure. Now you see all these posts on all over social media. as i I saw this funny meme. um What was it? You saw like a group of people at a party and there was everybody on the world having this party and the EU um being in the corner, having their drink and solitude and just the speech bubble above it saying, yeah, we got the best regulation. It's kind of funny. um And it kind of hits the nerve when it comes to what is going on at the moment.
01:22:32
Speaker
It doesn't mean it has to be this way long term. I just really hope that they will not take too long with adapting these things because the tech it's always difficult for regulation to keep up with technological changes. But we as we talked about it, this now We're now on a different racetrack. This speed is different. It's very different now. And we are in a global economy, especially when it comes to the cyberspace. You use VPNs, you use any of those other things, then there's not much you can do to always block people from using these things.
01:23:15
Speaker
And I think if they have these regulations and I have nothing against German companies or European companies being you know being protected and there's a push for innovation within Europe.
01:23:27
Speaker
There, I just don't see the same level of financial investment then taking place to rival the US tech companies. It's always easy to say we protect our market. we We do all of these things. No, you can't do that. We do it in-house. And it's it's also happening with GDPR and the Mettech sector. No, it's German companies, German servers. That is all fine. But sooner or later, that argument is not going to fly anymore.
01:23:54
Speaker
If the solutions that are being made are just subpar to what the competition does on a global scale. So if they do these regulations, I think it's good. But at the same time, when it comes to innovation, you can't just hope it's going to come out of thin air with no financial backing. insist that I don't see it happening. you know Yeah, and I'm here with you it's it's on the same page. um I mean, if you look at it, I think there need to be ah needs to be a clearer definition what we do not want, but also what we want. Yes. And it's, of course, financial investment is lacking, but also I don't see a vision or some some alternative to that, say,
01:24:39
Speaker
Big tech us hidden sauce models where we do actually not know how they work because they stop publishing research papers on it yeah and um i think this should be a much more bigger factor in the regulation.
01:24:55
Speaker
I mean, we all, I mean, I was a big fan of social media. 20 years ago, 15 years ago, I saw that could make the world a better place and so on. We saw what happened there. Basically, social media turned into big ad companies at monopolies.
01:25:12
Speaker
And now it's a bit funny that Meta is proposing this open source AI. And yeah I mean, they are investing billions of dollars into models and then publishing them for free. There could be a huge economy using these models and ah doing stuff with it in Europe. I mean, the the head of Meta AI, Jan LeCun, he is French. I think so they should listen more to him.
01:25:38
Speaker
He's a very special guy. I don't know if you follow him on X. now And he's always, ah he's always ranting against Elon Musk. And that's very funny. He's that's really, I really love to listen to podcasts with him. This is maybe a good recommendation. He has good things to say about open source AI. For sure. About what the EU should do differently. I mean, I think he is French born. I don't know if he has dual citizenship now. um Yeah.
01:26:08
Speaker
Yeah, I should check him out and I was and I have to agree with you when When I heard about the Llama models, which were the ones from Facebook, I first heard about just, yeah, there are these open source models, Llama, and I played around with them and was actually pretty impressed. Oh, this is pretty good. ah Who made this? And I saw Facebook, where it's like, what? They gave this out. Meta of all companies gave this out for, okay.
01:26:40
Speaker
Interesting. like That's all I could think of was interesting. But nonetheless, it shows the power of making this accessible to as many people as possible. Because I heard so many companies, you know, that once I was also advising, they they always said like, yeah, you know, we should do this with AI, we should do that with AI. No, we should make our own. It's like, okay, how much budget do you give us to do it?
01:27:07
Speaker
And it's like, well, and then they came up with, yeah, like, I don't know, a hundred K and they thought they were already generous. Like, Hey, what do you want for this hundred K? Yeah. Like, you know, chat GPT type performance. And it just stopped. I just couldn't stop laughing. It's like, what what are you like? Microsoft just invested 10 billion. You know, do you think that's comparable to anything that you want? That's yeah. So when you have these open source models, it can.
01:27:36
Speaker
It allows the small players just as, you know, when it came to WordPress, making websites, you know, it's just and and and enabled so many small people, small, small players, I should say, not small people, small players to do big things because these tools are like an accelerant if used in the right way. So I hope that you will, well.
01:27:59
Speaker
Get their shit together. Adjust it along the way and get their shit together. Yes, yeah yes. like except They need to do much more. so i mean It's good that they do regulation. Let's not be against yeah that, but they should know they should should adopt it and they should allow things um yeah which enables companies and people here. um Yeah, for sure.
01:28:22
Speaker
Man, are there ah there any books or other types of resources that you can recommend for the listeners that want to learn more? Or maybe, I mean, I'm sure they go they should follow you anyways, because I always like talking to you. you got You got the latest inside track on these things, but are there some resources you think that would be quiet quite good for them to know about?
01:28:44
Speaker
You had to be honest. I mean, i when I prepared for the conversation, I thought about that. And I think it's very difficult at the moment to recommend just resources, right? I mean, everything is outdated so soon.
01:29:00
Speaker
I mean, it's, ah I could now say, I find X, so Twitter, no still a good source. I mean, a lot of people working for OpenAI, Jan Lekun, in his rants, and ah people like that are on there.
01:29:18
Speaker
during the conversation, driving the conversation. So if you want to know about what is going on in the US, which is most advanced, I think this is the best spot to look. Then there are a few newsletters. So I can also recommend this Decade community.
01:29:35
Speaker
They have pretty cool newsletters to keep updated and other than that connect with the right people, visit events and I think most of all start using those tools and don't just use them to be able to use them but come up with side projects, be creative, use them to do something you couldn't do before.
01:29:56
Speaker
um Yeah, you will be surprised where you can actually find some value for yourself and your work and also your life besides work and using them.
Conclusion and Future Conversations
01:30:07
Speaker
Yeah, for sure. m Is there something maybe that's really pressing that you wanted to to mention that that I didn't ask you yet?
01:30:18
Speaker
No, and not really. ah thanks Thanks a lot for this super interesting discussion, I would say. um Hope to continue it someday next time we see each other. And yeah, there are other listeners out there. I hope you enjoyed the conversation and let me know if you have anything to add or ah refuse what I said. Very very happy to discuss.
01:30:44
Speaker
No, also thank you so
Guest's Social Media and Projects
01:30:45
Speaker
much for your time and this great conversation. And this will be in the show notes anyways, but where can people find you if they want to reach out on the various platforms? Yeah, so I think the platform I'm using most personally for this kind of stuff is LinkedIn.
01:31:02
Speaker
So I'm not posting as much anymore because there's just too much AI-generated stuff on LinkedIn, I feel. So there's not that much more value in the platform as it used to be a few years ago. But I think this is still a good channel and I think it's the best ah platform to connect with me.
01:31:22
Speaker
And then um the project I mentioned earlier, so I'm now running it under the handle inactive content on Instagram and TikTok soon. um Yeah, there will publish these books I spoke about and where I work. Hopefully I work through the whole five levels of AI with them. And ah first it will be AI integration into print books. And
Future of AI in Meditation
01:31:49
Speaker
maybe who knows, maybe in one or two years I will have AI agents who teach you meditation. That's going to be fun. I mean, all of that has been the show notes anyways, all your handles as well. And well, Sven, thanks again. Thanks a lot. We will see each other soon, finally, again in the flesh. And well, to everyone listening, have a great day.
01:32:16
Speaker
Hey everyone, just one more thing before you go. I hope you enjoyed the show and to stay up to date with future episodes and extra content, you can sign up to the blog and you'll get an email every Friday that provides some fun before you head off for the weekend. Don't worry, it'll be a short email where I share cool things that I have found or what I've been up to. If you want to receive that, just go to Ajmal dot.com. A-D-J-M-A-L dot.com and you can sign up right there. I hope you enjoy. it