درباره این اپیزود
In this episode of Decode AI, hosts Michael Plettner and Ralf Richter explore the latest advancements in artificial intelligence, including Claude 4's safety features, Google's VU3 for video production, the introduction of local AI models with AI Edge, and the reasoning capabilities of Gemini 2.5. They also discuss Project Astra's real-time assistance, the implications of AI in cybersecurity, and the legal challenges posed by AI-generated content. The conversation highlights the rapid evolution of AI technology and its impact on various industries.
Takeaways
Claude 4 introduces advanced guardrails for AI safety.
Google's VU3 democratizes video production, making it accessible to all.
AI Edge allows users to run models locally on devices.
Gemini 2.5 enhances reasoning capabilities for complex tasks.
Project Astra integrates real-time AI assistance into daily life.
AI agents are outperforming elite human hackers in cybersecurity.
Legal implications arise from AI-generated court cases.
Ethics in AI remains a complex and evolving challenge.
AI is transforming traditional search into an interactive experience.
The rapid evolution of AI technology is reshaping industries.
Reference Links
AI, Microsoft Build, OpenAI, language models, AI development tools, hardware advancements, Google Gemini, technology development
در این اپیزود
یادداشت ها را نشان دهید 🔗
رونوشت 🔗
00:00:17.570 --> 00:00:26.042
Hello and welcome to Decode AI, your podcast diving into fascinating, complex and ever-evolving world of artificial intelligence.
00:00:26.042 --> 00:00:32.649
Each episode, Michael, and I bring you insights from both the consulting and the technical frontiers.
00:00:32.649 --> 00:00:35.441
whether you are here to sharpen your business strategies.
00:00:35.441 --> 00:00:37.265
or deepen your technical understanding.
00:00:37.265 --> 00:00:38.481
We've got you covered.
00:00:38.481 --> 00:00:40.465
Grab your coffee and sit back.
00:00:40.465 --> 00:00:42.307
And let's decode AI.
00:00:42.307 --> 00:00:43.399
Hi, I'm Michael.
00:00:43.399 --> 00:00:47.561
I'm technically, a digital transformation.
00:00:47.561 --> 00:00:48.521
I don't know.
00:00:48.521 --> 00:00:50.593
You can put in bullshit bingo here.
00:00:50.593 --> 00:00:52.912
I'm more the technical, not the technical guy.
00:00:52.912 --> 00:00:53.673
That's the point.
00:00:53.673 --> 00:00:57.204
I'm more the business point, business part here.
00:00:57.204 --> 00:01:06.906
And I can help you to understand some ideas, figure out how to put it into your business and get AI work for you.
00:01:06.906 --> 00:01:10.590
bring it into your business and work smarter, not harder.
00:01:10.590 --> 00:01:11.643
And Ralph?
00:01:11.643 --> 00:01:18.001
Yeah, I'm Ralf, your resident tech wizard from coding to GitHub, Co-Pilot and Deep AI technology.
00:01:18.001 --> 00:01:20.596
I'll ensure we're grounded in reality.
00:01:20.596 --> 00:01:22.891
Well, the digital one anyway.
00:01:22.891 --> 00:01:25.483
So let's start with our first topic for today.
00:01:25.483 --> 00:01:33.165
And to give you a glimpse into our whole episode, we will talk about a lot of Google, especially the Google I.O.
00:01:33.165 --> 00:01:39.933
conference we had a couple of days, weeks, years ago, depending on whenever you listen to the podcast.
00:01:39.933 --> 00:01:44.727
But we would like to talk about the latest parts for us and our highlights.
00:01:44.727 --> 00:01:47.781
So we would like to start with Claude 4.
00:01:47.781 --> 00:01:57.314
And it feels like this is more about the AI safety instead of just we have a new model or something like that.
00:01:57.436 --> 00:02:03.751
Indeed, Claude 4 introduces advanced guardrails against AI misuse.
00:02:03.751 --> 00:02:12.697
So think of it as AI's own ethics, manually encoded directly in the system, essentially making AI safer by design.
00:02:12.697 --> 00:02:19.772
So that's the clue at what Claude 4, Anthrophic's, Leap in AI safety is about.
00:02:19.772 --> 00:02:27.586
That's interesting because we had some previous episodes talking about some security issues also with MCP service.
00:02:27.586 --> 00:02:39.507
So it's interesting to see that companies get a focus on security points and not only developing faster, cheaper, better models.
00:02:39.507 --> 00:02:42.098
So this is something which will have...
00:02:42.098 --> 00:02:45.924
a real huge impact into businesses itself.
00:02:45.924 --> 00:02:58.925
If you have to decide between, I wouldn't say unsecure, but if you have to rely on security, you may choose something which is more secure.
00:02:59.187 --> 00:03:04.931
And on the technical part, Well, it is not so much about technique.
00:03:04.931 --> 00:03:13.758
It's looking for ethics, so challenging itself, whether it is an ethical approach, which is commonly accepted or not.
00:03:13.758 --> 00:03:19.152
So it is going on to the ethical standards and it is challenging itself by reinforced learning.
00:03:19.152 --> 00:03:24.460
And that brings a lot into safety when it comes to ethical questions using AI.
00:03:24.460 --> 00:03:29.687
which is a huge leap forward to bring it on a safer space for everybody.
00:03:29.687 --> 00:03:34.467
And I mean, that's an important thing here, Yeah, well, don't get me wrong.
00:03:34.467 --> 00:03:41.784
I totally agree about every, from my point of view, every system has to follow a specific ethical standards.
00:03:41.784 --> 00:03:54.486
The well at the beginning of my sentence is actually we have a lot of different ethical standards if you look at different regions in the world, for example.
00:03:54.486 --> 00:03:57.930
So there are some differences in the culture, in the requirements.
00:03:57.930 --> 00:04:04.896
And that's the point why you have to decide actually, yeah, is it what you're looking for?
00:04:04.896 --> 00:04:15.689
And it means a broader business adoption, I would say, if you are in the exact same area and it will help to shape the market actually.
00:04:15.689 --> 00:04:30.259
And it's interesting that we, from my point of view, to see this focus on specific areas to improve the, maybe the way how models get treated and what's coming out.
00:04:30.259 --> 00:04:34.182
And it sounds like unethical for the people.
00:04:34.182 --> 00:04:36.894
Yeah, what sounds unethical.
00:04:36.894 --> 00:04:43.269
It's like that AI gets its own moral compass to navigate through all that ethical jungle.
00:04:43.269 --> 00:04:55.461
And I think it is more or less for the Western Hemisphere, less towards the Eastern Hemisphere, where maybe a different understanding on ethics is available, not on all points, but on some.
00:04:55.461 --> 00:05:08.194
Yes, can see the changes in Claude are going to help running in a Western Hemisphere, more European related, more Western related.
00:05:08.194 --> 00:05:09.565
So that's true for sure.
00:05:09.565 --> 00:05:15.471
And there are differences or differences in all models up from where they made it.
00:05:15.471 --> 00:05:18.574
And it is assured that they differ.
00:05:18.574 --> 00:05:20.687
from each other, country by country.
00:05:20.687 --> 00:05:27.391
And then we have another cool topic, and you've announced it by the introduction already, that we're talking a lot about Google.
00:05:27.391 --> 00:05:29.423
Google's stuff is really cool.
00:05:29.423 --> 00:05:39.149
And some of you may noticed in social media already that there are ton of millions of AI-generated videos available for now.
00:05:39.149 --> 00:05:43.045
And it's making a big wave that Google's VU3 is there.
00:05:43.045 --> 00:05:49.081
and is allowing users to generate a ton of AI videos in just a few days.
00:05:49.081 --> 00:05:55.237
And Google itself is understanding it as a democratization of AI-powered media production.
00:05:55.237 --> 00:06:00.514
So that's pretty tremendous what's happening out there with Vio3.
00:06:00.514 --> 00:06:19.504
Well, I would call it actually a kind of revolutionizing thing because this is not only democratizing and liberating or I don't know how we can put it in positive phrases.
00:06:19.504 --> 00:06:24.425
For the media production, it's also something which is incredible.
00:06:24.425 --> 00:06:34.362
If you see the first videos generated by this AI, not only was it plain video, but also with voice, with music.
00:06:34.362 --> 00:06:37.785
So you get the full video with audio and video experiences.
00:06:37.785 --> 00:06:50.144
And you get a video which is basically not different from any professional directed video from quality, from how people look like.
00:06:50.144 --> 00:07:00.783
It's not the typical issues like six or eight fingers or some arms and additional arms and legs or something like that doing motion.
00:07:00.783 --> 00:07:04.538
It's something which is really, really, really powerful.
00:07:04.538 --> 00:07:12.096
And if you take a look in the business part, we have companies who create actually those videos to earn money.
00:07:12.096 --> 00:07:14.079
if you have a...
00:07:14.079 --> 00:07:15.579
How can I say this?
00:07:15.579 --> 00:07:18.632
A pretty cheap help in this area.
00:07:18.632 --> 00:07:19.524
I know...
00:07:19.524 --> 00:07:30.386
If you take a look at the license prices, about, I think it's in a larger, I'm not sure, 250 or 299 dollars per user per month.
00:07:30.386 --> 00:07:42.889
It's not cheap, but compared to a professional who creates videos like this, with artists, with lights, with audio, with all the stuff you need to create a brilliant video.
00:07:42.889 --> 00:07:48.213
Whee! That will change and disrupt the media circus definitely.
00:07:48.213 --> 00:08:04.175
Yeah, it feels like that Vue 3 is the enabler to bring a disruption into all this stuff in the, into the media world, because it is like going to be that AI generated media is more and more becoming mainstream.
00:08:04.175 --> 00:08:17.636
And technically, technically spoken, is a diffusion model, which is very optimized for the video data production and is going to allow you to rapidly generate coherent and contextually relevant videos.
00:08:17.636 --> 00:08:21.627
It's like having an entire film studio at your fingertips.
00:08:21.627 --> 00:08:31.898
So that's pretty cool because that means it is pretty easy to use and brings up technically good results with a high quality at the end of the day.
00:08:31.898 --> 00:08:33.576
And that's, that's the thing here.
00:08:33.576 --> 00:08:34.297
Absolutely.
00:08:34.297 --> 00:08:45.517
yeah, as we always say, there's a huge opportunity in this, especially for companies who work with the new technology.
00:08:45.517 --> 00:08:54.844
So they can add specific scenes, some specific parts into production, real world production films.
00:08:54.844 --> 00:09:03.649
But it's definitely something which is disruptive ground-shaking, I would say, for video creation.
00:09:03.649 --> 00:09:06.712
think we still have some limitations, right?
00:09:06.712 --> 00:09:16.197
We still have a better movie or a better scenery if a human can interact and maybe improvise into space.
00:09:16.197 --> 00:09:18.070
in some specific scenarios.
00:09:18.070 --> 00:09:20.692
So it's definitely something different.
00:09:20.692 --> 00:09:28.966
But, and you know you can ignore everything before the but if there is a but in the sentence, you get some impressive results from AI.
00:09:28.966 --> 00:09:31.989
And as you said on the fingertip, it's just there.
00:09:31.989 --> 00:09:33.120
It takes some minutes.
00:09:33.120 --> 00:09:38.200
In the near future, it will definitely change.
00:09:38.200 --> 00:09:40.341
the way how we think about videos.
00:09:40.341 --> 00:09:45.565
Don't think about social media, that's scary actually to think in this scenario.
00:09:45.565 --> 00:09:52.272
Anyways, the businesses have to adapt and use it for the good and to improve a lot in the business.
00:09:52.272 --> 00:09:53.700
Yeah, true story.
00:09:53.700 --> 00:10:14.956
Well, and then there's something which is maybe, I don't know, it's not a secret anymore, but it was released pretty quietly from Google, the local AI model, actually, which is, we had a discussion about small language models already, which feels definitely like Google.
00:10:14.956 --> 00:10:22.474
did something which is a good competitor to AI, the Apple Intelligence feature.
00:10:24.215 --> 00:10:25.471
You missed that, right?
00:10:25.471 --> 00:10:31.649
that Apple is so much stepping on the same place every time, so they're not coming forward.
00:10:31.649 --> 00:10:33.350
really far behind.
00:10:33.368 --> 00:10:50.480
I love Apple for the whole ecosystem and integration, but on the AI perspective, it's absolutely, and it's a shame they call it Apple Intelligence to associate AI with real AI.
00:10:50.480 --> 00:10:51.312
my gosh.
00:10:51.312 --> 00:10:52.984
I think that hurts a little bit.
00:10:52.984 --> 00:10:53.955
To you.
00:10:56.399 --> 00:11:00.485
Coming some years up, you need something, a break or so.
00:11:00.485 --> 00:11:02.227
Shall we pause the video?
00:11:05.253 --> 00:11:06.067
I'm sorry.
00:11:06.067 --> 00:11:13.296
this is the, the app is actually the AI Edge Gallery app.
00:11:13.296 --> 00:11:20.414
And it enables users to download and run AI models directly on their Android device.
00:11:20.914 --> 00:11:24.740
For instance, yes, without needing an internet connection.
00:11:24.740 --> 00:11:25.288
Mm-hmm.
00:11:25.288 --> 00:11:26.094
impressive.
00:11:26.094 --> 00:11:26.595
Yeah.
00:11:26.595 --> 00:11:30.648
So it was a bit like a sneak peek into something.
00:11:30.648 --> 00:11:34.043
So Google didn't announce it broad and was loud with it.
00:11:34.043 --> 00:11:38.547
But it is like Michael said, you can now run an AI model in your pocket.
00:11:38.547 --> 00:11:40.149
And that's tremendous.
00:11:40.149 --> 00:11:44.511
So Google released a so-called AI Edge Gallery.
00:11:44.511 --> 00:11:56.062
And therefore you can download a model which is a light RT enabled or is capable to run in the light RT runtime and your device is supporting that.
00:11:56.062 --> 00:11:57.623
So there are some limitations.
00:11:57.623 --> 00:12:02.784
We don't want to say that you can run everything on your phone now.
00:12:02.784 --> 00:12:11.551
When that's done and that's set, you can literally try out the most models which are capable to run on light RT runtime in your pocket.
00:12:11.551 --> 00:12:13.243
And that's pretty cool.
00:12:13.243 --> 00:12:24.475
Technically, is really that you run an AI model efficiently on your user's device and you can import whatever you want there.
00:12:24.475 --> 00:12:35.477
it has, as I said, some limitations, but it is also providing you with a high flexibility regarding models and stuff you can do on your custom applications and stuff.
00:12:35.477 --> 00:12:38.740
So it's really, I'm looking forward to get my hands on it.
00:12:38.740 --> 00:12:42.890
Didn't have the time yet to try it out, but sounds promising to me.
00:12:42.890 --> 00:12:49.932
And the fantastic, fantastic story behind this, and that will make Michael very happy.
00:12:49.932 --> 00:12:56.583
There is the promise that the iOS app version is expected to be released soon.
00:12:56.583 --> 00:13:00.416
So you finally get real AI on your app iPhone.
00:13:00.416 --> 00:13:07.263
Yeah, and Apple will not allow to replace the existing AI technology they implement.
00:13:07.263 --> 00:13:08.904
Anyway, different story.
00:13:08.904 --> 00:13:17.130
I like the idea to have AI for specific needs, especially in some areas.
00:13:17.130 --> 00:13:20.953
We live actually in Germany where we don't have...
00:13:20.953 --> 00:13:23.764
good Wi-Fi or internet connection all the time.
00:13:23.764 --> 00:13:31.461
So if you have a limited connection or you just want to remove the footprint and use online connectivity.
00:13:31.461 --> 00:13:38.725
And that's something which sounds pretty interesting for me just to have it right into my pocket.
00:13:38.725 --> 00:13:42.538
I don't have to care about where am I right now.
00:13:42.538 --> 00:13:44.360
I'm in an elevator in...
00:13:44.360 --> 00:13:47.648
the underground or just traveling through Germany.
00:13:47.648 --> 00:13:51.702
So there are still different scenarios and it will still work.
00:13:51.702 --> 00:13:52.985
That's fantastic.
00:13:52.985 --> 00:14:00.585
Yeah, also is touching other areas like emphasizing privacy and not only the offline capability.
00:14:00.585 --> 00:14:07.695
So it is also regarding privacy a lot and it will definitely change the way applications are developed anyways.
00:14:07.695 --> 00:14:13.801
I hope this will come to my iPhone pretty soon.
00:14:14.528 --> 00:14:17.222
Otherwise we can recommend you an Android device anyways.
00:14:17.222 --> 00:14:20.440
So I'm running both and I like Android still more.
00:14:20.440 --> 00:14:26.874
Yeah, but you know, I live in the whole ecosystem and that's so, so much better.
00:14:26.874 --> 00:14:29.355
You live on a Windows device, right?
00:14:29.695 --> 00:14:31.427
I don't, for good reasons.
00:14:31.427 --> 00:14:33.106
Some parts are good.
00:14:35.948 --> 00:14:38.481
The quality of the OS.
00:14:38.481 --> 00:14:39.001
Okay.
00:14:39.001 --> 00:14:40.072
I'm great with that.
00:14:40.072 --> 00:14:43.183
So Google then there was more on Google IO 2025.
00:14:43.183 --> 00:14:55.615
So we had two of them now and we should have a look on to other five top AI announcements and the future impact of Google IO 2025.
00:14:55.615 --> 00:15:00.686
The landmark event when it comes to Google technology.
00:15:00.686 --> 00:15:03.438
So I think we already talked about this.
00:15:03.438 --> 00:15:05.080
Let me stick to the order.
00:15:05.080 --> 00:15:06.072
I do it.
00:15:06.072 --> 00:15:07.332
Let me stick to the order.
00:15:07.332 --> 00:15:12.456
But I would like to have the fifth topic, The fifth.
00:15:13.347 --> 00:15:15.570
No, but you can stick to the order.
00:15:15.570 --> 00:15:18.393
So the first part is reasoning.
00:15:18.393 --> 00:15:20.775
Reasoning is coming to Gemini.
00:15:20.775 --> 00:15:23.696
with Gemini 5....
00:15:23.696 --> 00:15:28.457
Sorry, Gemini 2.5 and not 5.2.
00:15:28.457 --> 00:15:36.799
With Gemini 2.5 Pro, deep thing, you can get reasoning into Gemini.
00:15:36.799 --> 00:15:43.461
And I've heard so much good parts about Gemini over the last months, unfortunately.
00:15:43.461 --> 00:15:44.109
And...
00:15:44.109 --> 00:15:55.019
I'm afraid to see Gemini coming to docs and Gmail, connecting some dots like Microsoft did with Copilot.
00:15:56.881 --> 00:15:58.063
which they didn't do good.
00:15:58.063 --> 00:16:00.149
Oh, I'm sorry.
00:16:00.149 --> 00:16:01.059
Yeah, yeah, yeah.
00:16:01.059 --> 00:16:09.230
um it's not so bad as Apple is doing with Apple Intelligence.
00:16:09.230 --> 00:16:13.413
There are still someone on the market doing it more bad than, than them.
00:16:13.413 --> 00:16:13.903
Okay.
00:16:13.903 --> 00:16:14.234
Great.
00:16:14.234 --> 00:16:24.498
So Jimini 2.5 pro introduces a deep thing and is enabling AI to perform complex multi-step reasoning tasks, which is a thing in chat.
00:16:24.498 --> 00:16:26.327
GPT is already available and get up.
00:16:26.327 --> 00:16:27.568
Copilot is already available.
00:16:27.568 --> 00:16:31.831
And now Google's Jimini is also capable to do so, but you need to pay for it.
00:16:31.831 --> 00:16:33.383
on the free version, you won't get it.
00:16:33.383 --> 00:16:35.863
So you have to pay like, what is it?
00:16:35.863 --> 00:16:38.908
24 bucks a month for Gemini Pro.
00:16:38.908 --> 00:16:40.289
Sorry, didn't get it yet.
00:16:40.289 --> 00:16:43.533
Sorry, it is 22 bucks in Europe to get the feature.
00:16:43.533 --> 00:16:53.094
But it is a tool made for researchers and developers in the mainstream and will give them the opportunity to go on.
00:16:53.094 --> 00:16:59.442
reasoning tasks, which are over multi-steps executed then to assist them by researching.
00:16:59.442 --> 00:17:10.946
So, and as Michael said, across Google's platform, it'll connect the dots to make users capable running over their docs, emails, appointments whatsoever, and work over that.
00:17:10.946 --> 00:17:12.328
That's pretty cool.
00:17:12.328 --> 00:17:25.010
Yeah, and another cool story is that the AI is coming to Google search and that's going to be personalized now and will be very interactive.
00:17:25.010 --> 00:17:37.211
Many of you may experience it already that Gemini is sitting in Google search and that's going to be more and more that way.
00:17:37.211 --> 00:17:52.640
And that means that the search itself will transform into a more interactive platform providing personalized responses based on your users of the context of the user, your location, whatever you did before, your behavior.
00:17:52.640 --> 00:17:55.892
I don't know yet if I will like it or not.
00:17:55.892 --> 00:18:09.006
mean, you know, was sometimes a struggle when you had a search history and you need to clean it up to get valid results back because someone misused your Google account or was searching on the same IP address you are using.
00:18:09.006 --> 00:18:14.821
I don't know yet what that will mean to us, but it will go and will be multimodal.
00:18:14.821 --> 00:18:21.621
That means you can have tags, you can have voice, you can have images, and we'll make the searches more intuitive.
00:18:21.621 --> 00:18:24.983
Well, I'm not sure if it really makes search intuitive.
00:18:24.983 --> 00:18:36.268
You know, two years ago when we started to explain AI to regular people, the first difference to explain was it's not a search engine.
00:18:36.268 --> 00:18:41.474
Please don't drop it like a supermarket opening hours next to me.
00:18:41.474 --> 00:18:43.676
something like this in a search query.
00:18:43.676 --> 00:18:50.684
um nowadays, two years ago, it was a little bit hard, right?
00:18:50.684 --> 00:18:51.157
that's true.
00:18:51.157 --> 00:18:55.259
But let me explain what I want to try to say here.
00:18:55.638 --> 00:18:59.172
We have used search engines for more than 20 years.
00:18:59.172 --> 00:19:02.864
now, maybe more.
00:19:02.864 --> 00:19:06.606
And now Google is moving this into the next era.
00:19:06.606 --> 00:19:11.911
And we will get something which is quite more interactive.
00:19:11.911 --> 00:19:12.981
And that means...
00:19:12.981 --> 00:19:19.567
it's maybe not so well known and needs some time until it will be intuitive.
00:19:19.567 --> 00:19:20.317
It sounds good.
00:19:20.317 --> 00:19:27.540
I really like to get the summary of the first steps I need to do for some certain parts I was looking for.
00:19:27.540 --> 00:19:34.068
So I get a short description about the next steps I need to do, for example, well, I don't know, create...
00:19:34.068 --> 00:19:37.093
create a Docker container or something like that.
00:19:37.093 --> 00:19:39.826
So I don't have to open a website anymore.
00:19:39.826 --> 00:19:45.252
I still get the references, I see what's going on, where the ideas are coming from.
00:19:45.252 --> 00:19:46.064
So I get...
00:19:46.064 --> 00:19:49.457
There is still some real content behind that.
00:19:49.457 --> 00:19:55.528
But I don't know how it works for the person who wants to publish blog articles, news or whatever.
00:19:55.528 --> 00:20:03.972
And Google is crawling everything together, bringing it into the search engine and giving you quick answers instead of the results.
00:20:03.972 --> 00:20:06.105
And you can verify, I know this person.
00:20:06.105 --> 00:20:08.518
I've read another article of this person.
00:20:08.518 --> 00:20:11.520
I trust them or I trust this person.
00:20:11.520 --> 00:20:17.373
I've seen a source I like because I know they write the style I prefer or something like that.
00:20:17.373 --> 00:20:18.364
And now we get...
00:20:18.364 --> 00:20:25.755
Something which is preselected already and not something which I can choose from.
00:20:25.755 --> 00:20:28.414
doesn't feel like...
00:20:28.414 --> 00:20:43.217
From my personal point of view, I may be at the age where I get some old man yelling at the window and, don't change anything! I don't know if it's this or I...
00:20:43.217 --> 00:20:45.237
I don't know.
00:20:45.237 --> 00:20:46.696
It sounds good.
00:20:46.767 --> 00:20:54.140
to get some personal driven, AI driven personal information that sounds good in the first place.
00:20:54.140 --> 00:20:57.972
And then they get shivers about they know everything about me.
00:20:58.773 --> 00:21:07.962
I don't, yes, but now someone can give me some more context, right?
00:21:07.962 --> 00:21:08.253
So.
00:21:08.253 --> 00:21:10.214
It's scary, but they did it already.
00:21:10.214 --> 00:21:15.357
So there was already that analysis, analysation of your account profile.
00:21:15.357 --> 00:21:17.318
you don't need to be scared about that.
00:21:17.318 --> 00:21:34.358
So, but coming back from your introduction of your talk, you were like saying that it is not intuitive and I'm going from the perspective of ease of use and I would say, yes, it is going to be more intuitive because Before you had to learn on how search terms are going to be used.
00:21:34.358 --> 00:21:41.023
Like how do I include a whole phrase instead of just a word for Google?
00:21:41.023 --> 00:21:44.537
What keywords do I have to use that I will find the right thing?
00:21:44.537 --> 00:21:56.174
And now it's going to be an assistant you can walk over with your natural language and you don't have to think about any highlightings or stuff around it'll understand your search term a little bit more.
00:21:56.174 --> 00:22:00.596
I will optimize it and then find maybe the better result in an easier way.
00:22:00.596 --> 00:22:05.340
problem I have there is the distrust that the results are okay.
00:22:05.340 --> 00:22:11.816
My experience is that sometimes I'm reading it and then I'm going to say, I'm I'm not 10 % convinced.
00:22:11.816 --> 00:22:13.036
I will have a look at it.
00:22:13.036 --> 00:22:14.717
And, that's more my issue.
00:22:14.717 --> 00:22:20.992
And as well, the other thing that I don't want to have my search history involved into that.
00:22:20.992 --> 00:22:35.299
I want to have the control about on how the search engine is behaving instead of it is more likely the chance there that it's going to advertise me something which it knows that it is on my agenda.
00:22:35.299 --> 00:22:38.948
oh you said something.
00:22:38.948 --> 00:22:49.039
I've heard about a story where someone was wondering why they get some advertising about topics they never searched for, never did before.
00:22:49.039 --> 00:22:57.309
And they figured out it was because someone else was using the same Wi-Fi and they searched for those topics.
00:22:57.309 --> 00:23:03.953
And then just imagine someone is using for a birthday gift for you.
00:23:05.612 --> 00:23:09.864
someone, you are looking up for something, maybe something related.
00:23:09.864 --> 00:23:17.152
And then they come up with, hey, remember yesterday evening at 2 a.m.
00:23:17.152 --> 00:23:20.246
you have been looking into something this.
00:23:20.246 --> 00:23:23.659
Will you continue there?
00:23:23.659 --> 00:23:26.891
Let's see what the future will bring to us.
00:23:26.891 --> 00:23:28.632
It sounds a little bit scary.
00:23:28.632 --> 00:23:30.272
I like the idea behind that.
00:23:30.272 --> 00:23:42.615
You don't have to be this person using just the specific words you are looking for and try to figure out do I need a plus, do I need a minus, to specify the query actually.
00:23:42.615 --> 00:23:45.097
It was not a question, it was a query.
00:23:45.097 --> 00:23:53.253
to get the search results and see how good it will be with the intuitive way to work with search engines.
00:23:53.253 --> 00:23:56.950
What I really like is the development of Google search.
00:23:56.950 --> 00:24:02.712
It felt like there was just a minor change over the last five, six, seven, eight years.
00:24:02.712 --> 00:24:04.915
And now we get real...
00:24:04.915 --> 00:24:06.689
AI in front of everyone.
00:24:06.689 --> 00:24:13.430
that's from the, let's see how the society will react to this perspective.
00:24:13.430 --> 00:24:14.442
Really interesting.
00:24:14.442 --> 00:24:19.757
I mean, I haven't brought up the topic yet in this episode, but we need to talk about AI browser.
00:24:19.757 --> 00:24:22.881
The browsing of the internet will change a lot.
00:24:22.881 --> 00:24:32.939
I've read a couple of things the past week and I'm really going to try it out and look into it, into deeper details, but you won't like it that much then.
00:24:33.715 --> 00:24:36.423
Yeah, sounds scary already.
00:24:38.134 --> 00:24:44.083
And I'm good to try something out, but please give me the option, the alternative to turn it off.
00:24:44.083 --> 00:24:47.933
Yeah, they claim it's not a human browser anymore.
00:24:47.933 --> 00:24:49.644
It's an AI agent browser.
00:24:49.644 --> 00:24:50.955
stay tuned.
00:24:50.955 --> 00:24:57.736
So the third project or the third topic we claimed or figured out at Google I.O.
00:24:57.736 --> 00:25:01.527
2025 was a project Astra and Gemini Live.
00:25:01.527 --> 00:25:06.518
So that there is happening something to this real time multimodal AI assistance.
00:25:06.518 --> 00:25:24.119
So Project Astra is going to evolve to be Gemini Live and will offer a real-time AI assistance by integrating the camera, voice and web data, which you maybe have seen already on merchandise on YouTube, TV, wherever.
00:25:24.119 --> 00:25:31.215
So when the guy is messing up the sauce and it's going to be a cookie at least with the assistance of Gemini.
00:25:31.215 --> 00:25:32.757
That's coming to be live now.
00:25:32.757 --> 00:25:40.329
So the multimodal approach, which enables the AI to understand and respond to complex real world scenarios.
00:25:40.329 --> 00:25:44.913
So it's just the guy is talking to Gemini, Hey, I messed up the source.
00:25:44.913 --> 00:25:46.105
What can I do with it?
00:25:46.105 --> 00:25:48.227
And Gemini answers, do you have this?
00:25:48.227 --> 00:25:50.538
And then add some flour and mix it up.
00:25:50.538 --> 00:25:53.690
And then you're going to have some nice cookies at the end of the day.
00:25:53.690 --> 00:25:54.141
it's.
00:25:54.141 --> 00:26:27.375
something like that or a girl is going to is that the washing saloon or so we're washing saloon for for clothing and sees a man pulling out pullover or something like that and that's shrink to a tiny size and and she's holding up her skirt and going to ask Gemini, hey Gemini holding up the camera to the washing label and says, how may I wash this and how may I dry this?
00:26:27.375 --> 00:26:31.405
And Gemini is answering on how to do that and stuff and so on and so forth.
00:26:31.405 --> 00:26:36.830
That's meant to be a life, Gemini life with real time AI assistance.
00:26:36.830 --> 00:26:41.913
that's, mean, yeah, we will have a 0.5 there.
00:26:41.913 --> 00:26:46.155
So it's going to be really, uh okay.
00:26:46.467 --> 00:26:47.617
I cannot hold it anymore.
00:26:47.617 --> 00:26:52.019
I want to talk about point five because it's hardly connected here.
00:26:53.038 --> 00:26:54.288
Sorry about that.
00:26:54.288 --> 00:27:01.780
Do you remember, I think we talked in one of our earliest episodes, we have been talking about Google IO last year.
00:27:01.780 --> 00:27:08.644
And there was a scenario where someone had Google glasses or another product.
00:27:08.644 --> 00:27:10.777
I'm not sure if it really was Google glasses.
00:27:10.777 --> 00:27:11.559
Mm-hmm.
00:27:11.559 --> 00:27:17.244
there was a device, glasses, with a camera and a voice interactive assistant.
00:27:17.244 --> 00:27:21.518
And this person was walking through an office or something like that.
00:27:21.518 --> 00:27:25.244
And then at the end, the person was asking, I forgot my keys.
00:27:25.244 --> 00:27:27.965
Where have I lost my keys?
00:27:27.965 --> 00:27:31.563
Or where have I seen my keys at the last point in time?
00:27:31.563 --> 00:27:36.421
Wasn't that the Ray-Ban Smart Glasses?
00:27:36.421 --> 00:27:37.092
Yeah, maybe.
00:27:37.092 --> 00:27:44.045
I'm not sure about the glass itself, but it was definitely in the presentation of Google as well.
00:27:44.045 --> 00:27:46.667
And then this person got the answer.
00:27:46.667 --> 00:27:51.762
Yes, you have seen it on the table next to the plant or whatever it was.
00:27:51.762 --> 00:27:52.913
That was mind blowing.
00:27:52.913 --> 00:27:55.336
And I think now it's available.
00:27:55.336 --> 00:27:57.911
This is coming to us.
00:27:57.911 --> 00:28:06.068
We have it in our daily life and it's not only connected via glasses and that's maybe the development.
00:28:06.068 --> 00:28:11.997
You can use Gemini Live for mobile devices, for my understanding.
00:28:11.997 --> 00:28:17.282
And it's also working with integrated into your regular life.
00:28:17.282 --> 00:28:18.494
You just use it.
00:28:18.494 --> 00:28:21.766
as you usually would like to use the product.
00:28:21.766 --> 00:28:24.836
And then you can interact with that.
00:28:24.836 --> 00:28:28.548
Your examples and I think also the keys.
00:28:28.548 --> 00:28:31.470
But yeah, that's crazy.
00:28:31.470 --> 00:28:36.781
Yeah, so Android XR, smart glasses, or a game changer.
00:28:36.781 --> 00:28:40.311
Now you spoiled the name of topic 5.
00:28:41.165 --> 00:28:43.182
Yes, you're true.
00:28:43.182 --> 00:28:44.603
that was, that wasn't me.
00:28:44.603 --> 00:28:45.413
That was you.
00:28:45.413 --> 00:28:48.463
And I'm now going to give them the name that they can look after it.
00:28:48.463 --> 00:28:57.619
So it is Android XR smart classes and it's a development with, with Samsung as well as Warby Parker.
00:28:57.619 --> 00:29:04.800
And they want to aim or they are aiming to seamlessly integrate AI into the user's field of vision.
00:29:04.800 --> 00:29:07.622
and they want to provide real time information.
00:29:07.622 --> 00:29:25.361
So that idea is not that new, but combined with LLMs and overall AI, it's getting now to be a game changer as the models got that good in quality and can handle much more complex situations like before.
00:29:25.361 --> 00:29:28.163
And that's really a tremendous game changer here.
00:29:28.163 --> 00:29:30.733
it's really, I'm looking forward to will come there.
00:29:30.733 --> 00:29:35.375
I'm still avoiding to wear that glasses or going to buy one of them.
00:29:35.375 --> 00:29:43.786
The one thing is that, I mean, would you be afraid now if someone standing up in front of you and wearing some glasses?
00:29:43.786 --> 00:29:46.886
What will be the indicator that you're now on filming?
00:29:46.886 --> 00:29:48.557
mean, will there be one?
00:29:48.557 --> 00:29:58.984
Is that then becoming a common and usual situation or how do we I'm not sure what we'll do people who cannot afford something like that.
00:29:58.984 --> 00:30:01.359
And really that feels here.
00:30:01.359 --> 00:30:04.256
I don't have a good feeling for this at the moment.
00:30:04.256 --> 00:30:11.622
Yeah, and I think that's definitely a point we have to consider about this very specific topic.
00:30:11.622 --> 00:30:22.451
One point is every time analyzing AI helping us in our daily life, which is impressive, I would say, from a technical perspective.
00:30:22.451 --> 00:30:24.563
But the other end is the...
00:30:24.563 --> 00:30:27.325
technology, put it into glasses.
00:30:27.325 --> 00:30:29.917
And then you raised some critical questions.
00:30:29.917 --> 00:30:30.941
I totally agree.
00:30:30.941 --> 00:30:34.192
But I also see some opportunities, right?
00:30:34.192 --> 00:30:38.204
So where shadow is always light or something like that.
00:30:38.204 --> 00:30:38.721
Mm-hmm.
00:30:38.721 --> 00:30:39.261
anyways.
00:30:39.261 --> 00:30:54.722
So if you think about someone, the old use cases we have seen a lot from Microsoft, other VR headset companies who produce VR headsets uh or AR headsets.
00:30:54.722 --> 00:30:56.954
Now, Apple is different.
00:30:56.954 --> 00:30:59.156
So the focus group for Apple is...
00:30:59.156 --> 00:31:00.919
They're stopping the program, aren't they?
00:31:00.919 --> 00:31:02.099
and customers.
00:31:02.099 --> 00:31:03.801
So that's what I want to say.
00:31:03.801 --> 00:31:06.403
there are rumors they will stop that.
00:31:06.403 --> 00:31:11.018
On the other hand, there are some rumors they will develop glasses as well.
00:31:11.018 --> 00:31:13.800
And then with AI.
00:31:13.800 --> 00:31:15.702
my gosh.
00:31:15.702 --> 00:31:16.257
Anyway.
00:31:16.257 --> 00:31:18.282
you mean Apple intelligence there.
00:31:18.282 --> 00:31:21.164
Unfortunately, I mean Apple Intelligence in this case.
00:31:21.164 --> 00:31:35.498
But the story behind that is actually we have seen, and I forgot the name, for Microsoft there's some AR glasses, are some AR headsets actually, not glasses, available since quite a while.
00:31:35.498 --> 00:31:36.915
It's HoloLens.
00:31:36.915 --> 00:31:38.857
The point is the name is HoloLens.
00:31:38.857 --> 00:31:54.261
And the idea behind that is you get something in a manufacturing area, for example, who can see the blueprints in the room to work with the blueprints or get data analyzed on a very quick and easy way compared with manuals.
00:31:54.261 --> 00:31:57.355
But you still have this huge headset, right?
00:31:57.355 --> 00:32:02.122
And now Google is putting this into glasses and you give someone in...
00:32:02.122 --> 00:32:08.921
I don't know, logistics, manufacturing, healthcare, something intelligent right before the eyes.
00:32:10.742 --> 00:32:12.144
That's interesting.
00:32:12.144 --> 00:32:13.395
That's really interesting.
00:32:13.395 --> 00:32:23.438
I totally agree with all the personal and private areas we have on a daily basis, like walking, going outside, shopping for groceries.
00:32:23.438 --> 00:32:26.825
This is something, well, I don't like that either.
00:32:26.825 --> 00:32:31.064
But for some business perspective, that's good.
00:32:31.064 --> 00:32:31.689
I like that.
00:32:31.689 --> 00:32:37.384
Yeah, I'm still having issues with it.
00:32:37.384 --> 00:32:46.513
I don't know the name of the film with Tom Cruise where Ball is claiming the next suicide or murderer.
00:32:46.513 --> 00:32:50.356
um Minority report.
00:32:50.483 --> 00:33:01.773
And that was so scary on how they determined who was passing by the the walls and is showing up then the dedicated marketing.
00:33:01.773 --> 00:33:03.955
No, I don't want to think about that.
00:33:03.955 --> 00:33:05.557
Let's go over the next topic.
00:33:05.557 --> 00:33:11.805
we had already Astra is becoming Gemini Live and real life assistance.
00:33:11.805 --> 00:33:16.137
So next topic is about VO3 and flow.
00:33:16.137 --> 00:33:20.622
So VO3, you remember we had it a few, a couple of minutes before.
00:33:20.622 --> 00:33:37.621
So where it is going to produce video for you and now also audio and is allowing you to create videos with talking peoples where the dialogue will match to the faces moving, including sound effects and so on and so forth.
00:33:37.621 --> 00:33:39.355
So really cool thing.
00:33:39.355 --> 00:33:42.277
I don't know if we need to go deeper.
00:33:42.277 --> 00:33:42.836
to that.
00:33:42.836 --> 00:33:50.289
So you already talked about the democratization there and that it will be a disruptor on the media market.
00:33:50.289 --> 00:33:52.692
So I'm completely with you there.
00:33:53.128 --> 00:33:56.541
I just want to the flow part.
00:33:56.541 --> 00:33:57.701
There's a...
00:33:57.701 --> 00:33:58.624
I call it product.
00:33:58.624 --> 00:34:00.345
I'm not sure if it's really a product.
00:34:00.345 --> 00:34:02.646
It's something Google named definitely.
00:34:02.646 --> 00:34:04.118
So I think it's a product.
00:34:04.118 --> 00:34:12.146
Anyways, flow is something for easy editing and working with the data you created with VO3.
00:34:12.146 --> 00:34:18.684
So that's something which can help you to improve the already great video content you get from VO3.
00:34:18.684 --> 00:34:20.079
Yes, absolutely.
00:34:20.079 --> 00:34:29.911
Are you open to spend 200 or 300 dollars to get your hands on Vio and Flow?
00:34:29.911 --> 00:34:36.103
Um, when I got it right, you can have view with that 20 bucks or 22 bucks.
00:34:36.103 --> 00:34:36.684
Really?
00:34:36.684 --> 00:34:38.119
Sorry, please be patient.
00:34:38.119 --> 00:34:39.373
I have to click something.
00:34:39.373 --> 00:34:43.516
you can click on try and flow and then create with flow.
00:34:43.516 --> 00:34:51.882
And then you get the smart message that flow is currently not available in your region.
00:34:51.882 --> 00:34:53.764
Okay, then I just have to change my...
00:34:53.764 --> 00:34:55.177
It's not a problem.
00:34:55.177 --> 00:35:01.925
It's not like Apple Intelligence, which is blocking everything because you have a lot of data where I actually am.
00:35:01.925 --> 00:35:04.139
It's more about where my browser is.
00:35:04.139 --> 00:35:05.532
So it's just Google.
00:35:05.532 --> 00:35:07.719
Okay, so when we look at Google I.O.
00:35:07.719 --> 00:35:09.521
Well, impressive, I would say.
00:35:09.521 --> 00:35:20.710
from my point of view, Google is something coming back and is on the AI perspective and maybe the whole ecosystem about mobile devices.
00:35:20.710 --> 00:35:29.579
Really something you have to consider and put into your strategic consideration if you talk about AI using AI.
00:35:29.579 --> 00:35:33.224
and to use it when it's integrated in everything.
00:35:33.224 --> 00:35:39.849
yeah, user-centric, we may get a new era of Google in the next couple of months.
00:35:39.849 --> 00:35:42.974
I wouldn't say years, definitely a shorter period of time.
00:35:42.974 --> 00:35:43.813
Absolutely.
00:35:43.813 --> 00:35:52.123
it's really emphasis on multimodal capabilities, real-time assistance, user empowerment through AI tools.
00:35:52.123 --> 00:35:57.628
That's reflecting a future where technology adapts seamlessly to human needs and context.
00:35:57.628 --> 00:36:03.753
So I'm keen to go and to see what's going on there in the next few months and years.
00:36:03.753 --> 00:36:06.088
Yeah, brings us to the next topic then.
00:36:06.088 --> 00:36:09.335
More or less a funny story if we want.
00:36:09.335 --> 00:36:16.572
Well, we can say it's a funny story, but I think some security persons will not agree with us.
00:36:16.572 --> 00:36:21.385
So the headline is, AI agents outperform human hackers.
00:36:21.385 --> 00:36:28.043
And I know from a couple of companies that the new issue is not only...
00:36:28.043 --> 00:36:43.315
something like phishing, only something like social engineering by itself, but it's using powerful tools like AI to improve the hacking capabilities and...
00:36:43.315 --> 00:36:52.838
Well, the recent cybersecurity competitions showcase that AI agents can really outperform human hackers.
00:36:52.838 --> 00:36:54.614
Elite human hackers.
00:36:54.614 --> 00:36:57.106
Elite human hackers, yes.
00:36:57.106 --> 00:37:07.788
So now the script kitty, which is able to write something to start a DDoS attack on 127.0.0.1.
00:37:07.788 --> 00:37:20.059
Please put something in the comments if you put 1337 in the comments if you got the joke.
00:37:20.059 --> 00:37:21.240
That's another one.
00:37:21.240 --> 00:37:23.577
Anyways, that's incredible, right?
00:37:23.577 --> 00:37:24.637
Yeah, that's right.
00:37:24.637 --> 00:37:36.349
These AI agents use reinforcement learning and generative, advisable models and continuously adapting their hacking strategies and techniques against the goal.
00:37:36.349 --> 00:37:44.644
I mean, imagine this is a digital cat and mouse game where machines are fighting against this and that will...
00:37:44.644 --> 00:37:46.976
evolve on the speed of machines.
00:37:46.976 --> 00:37:48.277
that's really incredible.
00:37:48.277 --> 00:37:50.561
it's really making me afraid.
00:37:50.561 --> 00:38:05.755
one or two episodes ago, we were joking that 512 kilobytes encrypted stuff will no more be safe in the far future when we're looking at quantum computing.
00:38:05.755 --> 00:38:25.224
And now we're talking about agents who are hacking and breaking into systems and when it comes to humans interacting with systems there is a fault tolerance built in and I mean We don't need to bring that further, but I mean That's scary, actually.
00:38:25.373 --> 00:38:28.503
We are both Microsoft MVP's, right?
00:38:28.503 --> 00:38:38.914
I don't want to put it all to Microsoft, but to emphasize the cloud security instances we have from different environments, I would say.
00:38:38.914 --> 00:38:41.005
But I know it from the Microsoft perspective.
00:38:41.005 --> 00:39:00.382
There are so many different ways how Microsoft improves the security, learn from different tenants, environments, and brings everything together to secure your environment, and also relies on big data on AI supported security.
00:39:00.382 --> 00:39:09.293
And I think you need something like that in the near future, not only this typical, we have a firewall, we are secure.
00:39:09.293 --> 00:39:15.952
This is nothing we will help people have in the near future and it will help us efficiently now.
00:39:15.952 --> 00:39:33.123
Yeah, it'll be very dynamic scenario with an interplay of AI defenders versus AI attackers and technologically will be agility and continuous adoption, which will become very crucial there.
00:39:33.123 --> 00:39:36.143
And that's really not something you can prepare for.
00:39:36.143 --> 00:39:41.925
Yeah, definitely something you have to keep an eye on, because this will develop.
00:39:41.925 --> 00:39:52.052
And even if you're capable to disconnect your network from the internet, you're not safe looking towards a LIDAR T or other technologies where you can run in LLM locally.
00:39:52.052 --> 00:39:56.119
So that brings additional, efforts into security you have to do.
00:39:56.119 --> 00:40:04.666
yeah, it's very, very dangerous, but it's funny that human hackers, human elite hackers are not that good as AI has become already.
00:40:04.666 --> 00:40:12.451
And I'm not sure if it is a matter of time what an AI versus a human needs to explore code and stuff, but can be.
00:40:12.451 --> 00:40:30.702
So, and then something other strange happened when we look at all that AI stuff and the use of Chetchi PT and where it is used already that was in use that we that there were fake court cases.
00:40:30.702 --> 00:40:34.326
and that produced an illegal drama with AI.
00:40:34.326 --> 00:40:47.583
So controversially, Chetri Pichy recently became famous for inventing 129 fake course, court cases, where legal professionals, are on the edge about AI's reliability.
00:40:47.583 --> 00:40:55.077
So Another story which is really needs some points and has some points and needs some.
00:40:55.077 --> 00:40:57.507
view angles to be reviewed, Michael.
00:40:57.507 --> 00:41:01.539
Yeah, well, it's definitely another level of hallucination, I would say.
00:41:01.539 --> 00:41:06.833
This is something when AI gives you something which is a whole story, right?
00:41:06.833 --> 00:41:08.471
It's not a single fact.
00:41:08.471 --> 00:41:17.061
It's absolutely from the start until everything which was discussed there and the results.
00:41:17.061 --> 00:41:17.992
Well, that's...
00:41:17.992 --> 00:41:28.027
That's interesting and brings us to the still necessary clarification and verification we need if we work with AI.
00:41:28.027 --> 00:41:37.641
From my personal point of view, it's really interesting because I had a customer and the legal department was asking, hey, can we do something like that with Co-Pilot?
00:41:37.641 --> 00:41:47.376
And we said, there are some AI companies using real data for this and using real cases and court files and something like that.
00:41:47.376 --> 00:41:51.019
And they are better for those scenarios.
00:41:51.019 --> 00:42:00.166
But well, reading this, I'm not sure if it's really, well, yeah, the models are trained, they're specified, and then we talk about chat.
00:42:00.367 --> 00:42:02.190
Here is Chetjipiti.
00:42:02.871 --> 00:42:05.452
I thought, I got it, I got it.
00:42:05.452 --> 00:42:09.215
But in this case, Chetch GPT itself, well...
00:42:09.215 --> 00:42:12.557
Yeah, so that really brings a point up, right?
00:42:12.557 --> 00:42:18.371
So it's about retrieval and synthesizing data and how that's going to be done.
00:42:18.371 --> 00:42:26.509
And do you need to verify that or not to have plausible, I mean, cases, but fake references.
00:42:26.509 --> 00:42:27.021
So.
00:42:27.021 --> 00:42:27.434
Yes.
00:42:27.434 --> 00:42:27.885
Good.
00:42:27.885 --> 00:42:36.614
Imagine the effect of that when one got or have to go to jail because of one of that fake references.
00:42:36.614 --> 00:42:38.485
I mean, that's pretty awkward.
00:42:38.485 --> 00:42:46.934
So it means verification and prove the result of AI is still a thing and it's nothing you can skip off.
00:42:46.934 --> 00:42:51.452
mean, trust is somehow building up, not on my page, but on others.
00:42:51.452 --> 00:42:53.423
who are using it.
00:42:53.423 --> 00:42:58.503
And I have to say really, and I'm really, really emphasizing that all the time.
00:42:58.503 --> 00:43:02.894
So double check everything what AI is doing and is prompting you.
00:43:02.894 --> 00:43:08.565
And yeah, one more thing when we are at that stage already.
00:43:08.565 --> 00:43:17.527
So when we talk about reasoning and multi-step thoughts and all that stuff, please do me a favor and don't misunderstand.
00:43:17.527 --> 00:43:19.978
thought with human thinking.
00:43:19.978 --> 00:43:22.731
That doesn't have to be anything of that.
00:43:22.731 --> 00:43:33.048
It is just the most likely probability, which can also be false positive in a step, but the result can be at least the correct answer.
00:43:33.048 --> 00:43:44.436
So when you force an AI to think multi-steps and to make it transparent to you and you're going through that steps, Don't think, don't understand it as human thinking.
00:43:44.436 --> 00:43:45.737
Please treat it as well.
00:43:45.737 --> 00:43:50.648
It can be false in between, but the result is correct and so on and so forth.
00:43:50.648 --> 00:43:53.079
So it has nothing to do with thinking.
00:43:53.079 --> 00:43:58.230
is just the most probable correct answer for that step.
00:43:58.230 --> 00:44:06.521
Understand it more or less like an in-between prompt challenge as anything else.
00:44:06.521 --> 00:44:07.818
Well, well set.
00:44:07.818 --> 00:44:08.701
Thanks, I'm sorry.
00:44:08.701 --> 00:44:10.574
Yeah, absolutely true.
00:44:10.574 --> 00:44:11.838
I think that's...
00:44:11.838 --> 00:44:13.030
We can end this episode.
00:44:13.030 --> 00:44:13.940
See you next time.
00:44:13.940 --> 00:44:14.577
No, just...
00:44:14.577 --> 00:44:21.688
No, no, no, let's do the recap stuff, Yeah, just let's have a quick recap for our latest episode.
00:44:21.688 --> 00:44:26.621
So you can, if you just skip to the end and want to hear the jokes, right?
00:44:26.621 --> 00:44:28.974
Then you can go back and see what you have missed.
00:44:28.974 --> 00:44:29.666
Much aux.
00:44:29.666 --> 00:44:33.262
All right, so the first topic was Claude Four.
00:44:33.262 --> 00:44:44.514
That's latest AI model promising even stricter safety standards, safety of course in becoming the new battleground for AI innovation.
00:44:44.514 --> 00:44:52.809
Yes, then we had Google's Vio3 making waves, allowing users to generate millions of AI videos in just days.
00:44:52.809 --> 00:44:59.478
It's the democratization of AI, democratization of AI-powered media production.
00:44:59.478 --> 00:45:05.768
Then Google dropped a new app to run local AI models locally.
00:45:05.768 --> 00:45:10.159
And yeah, that's still crazy.
00:45:10.159 --> 00:45:21.922
Anyways, then we have major announcements about Google products, new Google products and integrations from Google IO 2025.
00:45:21.922 --> 00:45:24.164
And yeah, I think...
00:45:24.164 --> 00:45:31.735
Vio3 is one of the most important and Gemini Live are one of the biggest ones from my personal perspective.
00:45:31.735 --> 00:45:33.510
And then we have...
00:45:33.510 --> 00:45:36.007
XR smart classes is your biggest point, man.
00:45:36.007 --> 00:45:40.717
Maybe, maybe I already wear glasses, so it's maybe just a minor upgrade.
00:45:40.717 --> 00:45:44.164
If you see me with new glasses, you have to be afraid.
00:45:44.164 --> 00:45:46.117
yeah, I'm already afraid.
00:45:46.117 --> 00:45:52.150
I don't know yet about your glasses, Yep, last but not least.
00:45:52.150 --> 00:46:01.887
We have that, we have heard that AI agents are more into cyber crime than into cyber security by outperforming human elite hackers.
00:46:01.887 --> 00:46:08.353
And that they showcased on a competition that AI agents are outperforming like human hackers.
00:46:08.353 --> 00:46:16.849
And that means that we're entering a new era where AI isn't just, just protective and it's proactively offensive.
00:46:16.849 --> 00:46:23.797
And then we had a chat GPT's fake court cases where, where the legal AI drama was ongoing.
00:46:23.797 --> 00:46:33.657
means that chat GPT recently became famous for inventing 129 court cases, which were fake as reference for a court case.
00:46:33.657 --> 00:46:39.262
So where legal professionals are on the edge of AI's reliability.
00:46:39.262 --> 00:46:39.847
Michael.
00:46:39.847 --> 00:46:41.768
Well, that was a lot, I would say.
00:46:41.768 --> 00:46:43.570
A lot of interesting stuff.
00:46:43.570 --> 00:46:49.246
And I like the point that we take out some news and talk about this.
00:46:49.246 --> 00:46:52.643
And there's way too many news to everything.
00:46:52.643 --> 00:46:56.746
I like that we can talk a little bit about the stuff behind that.
00:46:56.746 --> 00:47:00.650
But for now, I would like to close our today's session.
00:47:00.650 --> 00:47:04.875
Shall I try again with this outro part?
00:47:05.418 --> 00:47:16.150
Yeah, well, I, yeah, we, we, we've a lot about groundbreaking announcements and yeah, you can go for the outro and maybe you like it the way we did it the past.
00:47:16.634 --> 00:47:18.815
So then you may go on with that.
00:47:18.815 --> 00:47:23.079
uh tuned, stay interested.
00:47:24.001 --> 00:47:27.452
Here we go, bye bye, take care all, thanks for listening.
00:47:27.452 --> 00:47:27.981
Bye.