AI News flash from Google IO and Apple WWDC 2024

Summary

In this episode, Michael and Ralf discuss the latest news from Microsoft, Google, and Apple regarding AI. They start with Microsoft's decision to focus on enterprise and discontinue support for custom GPTs in Co-Pilot. Then, they move on to Google's Gemini for Workspace, which offers summarized documents and meeting recaps. They also discuss Gemini 1.5 Pro and Flash, which are improved models with multi-modality and long context understanding. Finally, they mention Google's Veo, a generative video model, and Project Astra, a real-time AI assistant. In this conversation, Ralf and Michael discuss the latest announcements from Google I/O and Apple's WWDC. They cover topics such as Google's LaMDA, Google AIO, and LearnLM, as well as Apple's Apple Intelligence and the integration of ChatGPT 4o. They also touch on the new iPad calculator and the upcoming guest, Henk Boelman. Overall, they provide insights into the advancements in AI and technology from both companies.

Takeaways

• Microsoft is discontinuing support for custom GPTs in Co-Pilot and focusing on enterprise
• Google's Gemini for Workspace offers summarized documents and meeting recaps
• Gemini 1.5 Pro and Flash have improved models with multi-modality and long context understanding
• Google's Vio is a generative video model and Project Astra is a real-time AI assistant Google's LaMDA and Google AIO showcase the advancements in natural language understanding and AI capabilities.
• LearnLM from Google aims to support students in their learning by providing additional resources and answers to educational questions.
• Apple's Apple Intelligence integrates AI capabilities into Siri and other Apple products, allowing for more interactive and personalized experiences.
• The integration of ChetGPT 4.0 into Apple's ecosystem offers developers the opportunity to build on AI capabilities.
• The addition of a calculator app on the iPad and other math-related features demonstrate Apple's focus on education and productivity.

The upcoming guest, Henk Boelman, will provide insights into AI and the global AI community.

Titles

Google's Veo: Generative Video Model
Project Astra: Real-Time AI Assistant
LearnLM: Supporting Students with AI in Education
Developing on AI Capabilities with ChatGPT 4.0

Chapters

00:00 Introduction and Overview
05:21 Gemini for Workspace: Summarized Documents and Meeting Recaps
10:28 Gemini 1.5 Pro and Flash: Improved Models with Multi-Modality
39:00 Project Astra: Real-Time AI Assistant
41:23 Google's LaMDA and Google AIO
45:46 LearnLM: AI Support for Education
53:09 Introducing Apple Intelligence
55:43 Creating Personalized Assistants with Gems
59:14 Apple's Integration of JetGPT 4.0
01:04:50 Enhancing Education and Productivity with the iPad Calculator
01:08:25 Upcoming Guest: Hank Bowman on AI and the Global Community

Links

•  https://www.windowscentral.com/software-apps/copilot-pros-gpt-builder-bites-the-microsoft-graveyard-dust-as-the-company-prioritizes-refinement-of-core-experiences 
• Google I/O 2024 (io.google) 
• Introducing Apple Intelligence for iPhone, iPad, and Mac - Apple 

AI, Microsoft Build, OpenAI, language models, AI development tools, hardware advancements, Google Gemini, technology development

Transcript

Michael (00:17.742)
Hello and welcome to our latest episode of Decode AI. My name is Michael and I'm here together with Ralf. Welcome Ralf.

Ralf (00:26.887)
Hello folks.

Michael (00:29.87)
Today we have a really, really special episode because we talk about latest news of different companies, huge tech companies actually talking about AI and some interesting stuff. Sorry if it's maybe not so detailed packed today because we have some limitations like they are not talking about the actual date, for example, something like that.

Ralf (00:58.983)
Yeah, release dates are really missing. So it's like can be released in summer, may appear in Germany next year. So it is not that detailed as we were with our build news. Yeah, but we have to start first with, I don't know what kind of news it is, but it is the news. So we have to switch first to Microsoft, I guess.

Michael (01:24.046)
Yeah, today we jump from Microsoft to Google to Apple. So the huge tech companies, I would say for the regular AI stuff, besides the leading tech company like OpenAI that you heard about the last episode, just check that out. It will be interesting what they have some news. And we start with a bad news or maybe a good news.

Ralf (01:29.639)
Yeah.

Ralf (01:53.767)
I really cannot estimate whether it is good or bad. So Microsoft is stating that they want to focus on their product and they want to focus on enterprise and stuff regarding quality and they're built in copilots as well. So the announcement is that beginning from 10th of June, they will disregard any custom GPT.

Michael (01:53.934)
What do you think?

Same.

Ralf (02:21.639)
which is deployed to Co -Pilot Pro. And you won't have access to it anymore since up from that date. And this is, I really cannot, I mean, it's July. So 10th of July, you won't be able to edit any GPTs you've custom created. And they'll also start deleting until 14th of July, any GPTs.

custom GPTs as well as the collected data with that custom GPTs. And we're not really knowing what they are going to do with that or what does the way. So what Microsoft says is that they continuing to elevate, evaluate their strategy for consumer co -pilot extensibility and are prioritizing core product experiences.

while remaining committed to developer opportunities. To this end, they are shifting their focus on GPTs to commercial and enterprise scenarios and are stopping GPT efforts in consumer pilot. I really cannot negotiate with myself whether that's a good or a bad news. I mean, I'm sorry for those who put efforts into it.

But this is what they decided to do. And that's it. What we will say today about Microsoft, I guess. Maybe it'll come up in some comparisons here, but those are the news we wanted to share with you.

Michael (03:57.966)
Yeah, and I cannot decide if it's good or bad on either, but yeah, it's, I think it makes sense from the technology perspective, like from the business and Microsoft side to reduce the effort for the private or more personal area. So you can still use Copilot Pro, which is a personal license.

20 bucks a month. So you can still use Copilot, but support all the other ways to integrate some stuff. It maybe makes sense to focus on the business need where they can get more money out of it. So yeah.

Ralf (04:44.819)
Well, to me, it's cool that they are stating that they want to focus on development experiences. So that's really touching my heart.

Michael (04:53.358)
But yes, that's it for today about Microsoft. We got really bigger news about Google and Apple. So let's start with Google. Talking about the Google I .O., which is quite a few weeks ago. So it's not brand new.

but still some really, really interesting use cases and scenarios they were talking about. So, the first of all is we got new models, improved models from Gemini. And that sounds interesting, but...

Yeah, let's start with Gemini for Workspaces or Workspace. So I think this is something we have seen with the integration of Copilot into Microsoft products. So it's pretty similar from the idea to summarize and get recaps from Google Meet meetings. And,

Gemini will integrate in the whole suite to support you with summarized documents, emails, and help you with getting into the information of the meeting itself. So it's getting details via the recap. And it felt like, yes.

That's something you have to do if you want to integrate AI technology into your suite. But on the other side, it's actually the same as we already got.

Michael (07:03.79)
I'm, it's, how can I put this? I am, I'm a Microsoft MVP and I don't want to, I don't want to talk about the other companies like I'm Microsoft minded and it's all bullshit. That's not the way it's, I just feel and the most parts are pretty much the same when we talk about the integration into, into desktop and, and.

Ralf (07:06.695)
You

Michael (07:32.398)
of course, mobile apps, but it feels like the first step we currently have from whatever company is we start with summarizing content because that's easy, easier for AI. And then we have a foundation and we can spin up and maybe get some new fancy stuff. And we have some fancy stuff from Google. So it's just about.

Gemini for Workspace right now.

Ralf (08:04.455)
I mean, first of all, Gemini itself has evolved, right? It is now in version 1 .5 and they claim it as well as a modal LLM now. And they as well claim it isn't anymore in LLM. So that confuses me 100%. And I got my point. And yes, it feels like that...

Michael (08:13.07)
Yes.

Michael (08:26.506)
Yeah.

Ralf (08:33.671)
I mean, Microsoft was first to come up with this co -pilot solutions and they building it into their office products. And as we all know, Workspace isn't anything else than a counterpart of Microsoft Office product, providing it in a more hybrid way, I would say. So the Google Workspace was the first workspace working on all devices, almost seamless.

And now they are adding a same like support, which is co -pilot giving to us. And yes, you're right. So it feels like not a copy, but they want to provide same services with a few, maybe with a few enhanced capabilities, which I cannot say yet, because I haven't had the touch on it. I hadn't had the time to explore it a lot.

but it is like, it feels like so. So, what came up to my mind is that they are giving a lot into search engine power here. So they are like pulling AI into their search engine so that, the results of this search engine will be somehow explained and more, detailed and summarized as you're talking about, workspaces as well.

They offer now the AI helpers as they say across the workspace like in Gmail and G -Drive, Docs, Photo and so on. And yeah, for sure it feels like it is now almost the same. So the products are going to be even again.

Michael (10:28.718)
Yeah, and that's good to hear. And Google is delivering these capabilities on very different ways. You mentioned the Gemini 1 .5 model, and they separate this into two different kind of models, which are the Pro and the Flash. Not light, it's a flash in this case.

But the Flash is actually the lighter, lightweight model, which has more speed and is more efficient. But still, both models can now handle multimedia. Sorry, multi -modality is the right term. I just remember myself about multimedia because that's actually the way. What I think that's the current use case. You put some video, some audio into it.

Ralf (11:14.951)
Yeah.

Michael (11:27.182)
and you get some something out of it. You can use it not only for text, that's a point actually. And it's got more understanding for longer text. So long context is a term here. So they get, you can get more with the new models. So I think it's, I'm bad with numbers, I'm bad with names and just.

good in talking about some high level stuff. So please forgive me. But I think they are on a very same level as OpenAI about the understanding of long text and the capabilities we had in the last episode from OpenAI talking about 4 .0, GPT 4 .0.

It felt pretty similar and it's interesting and there we have a big difference between Google and Microsoft. In the first place, Microsoft is integrating it with a help and with all the backend power from OpenAI in the background. But Google with its own AI model can integrate it much better and get

more out of the capabilities in the first place. So I haven't had any experience so far. I don't have any experience so far. But yeah, it's interesting to see if they are maybe able to put more, I don't know, change my documents by putting in some audio files and let it sound like...

this summer sound or whatever, create a poem like this matching to this melody, whatever. I don't know yet. But this is maybe something you have bigger opportunities with the current integration and the new model in this place.

Ralf (13:47.463)
Yeah, one thing we have to highlight is that if you want to get your hands on Gmini 1 .5 Pro, it costs you 20 bucks because you have to subscribe to Gmini Advanced. If you don't do, you won't get access to Gmini 1 .5 Pro. Okay. This is something I was really hard to figure out, but here you go. So.

Michael (14:00.846)
Yes.

Ralf (14:18.087)
as well as you have to pay for like other pro services from OpenAI or Microsoft, like Copilot Pro as well is worth 20 bucks a month, is it?

Michael (14:32.75)
Yeah, and it's for consumers. So if you compare it on a document suite level, I would say, it's cheaper actually, as Microsoft. And that's also something which can help and decide some companies to take a look into Google.

Michael (15:02.286)
All right.

Ralf (15:03.047)
So as well as on Google Workspace, you have to subscribe to a special service to get more AI help in your Google Workspace. We'll add some links on our show notes so that you can get a little bit more oversight about this. Yes, so we have spoken a lot about what is about Flash. So what is that Flash?

what you've mentioned already.

Michael (15:35.886)
This is a lightweight version of the Pro and this is the way to work with the new capabilities like multi -modality and long context understanding but on a faster and more efficient way. So it's...

It's still not a small language model. I wouldn't go so far, but it's a quicker way to integrate it and get your answers. It has a, I think a 10th of the trained information, but it's still...

SS Gemini 1 .5 Pro. And it's still able to do almost the same. It's just you get more created by the model itself.

Ralf (16:53.479)
Okay.

So, Jiminy Flash is like an enhanced model with the purpose to be built as a faster and more cost efficient model. Would I say so? Okay. And so, what I've heard about Flash is that it addresses developers' feedback and that they were claiming that they need low latencies.

Michael (17:11.854)
Yes. Yes.

Ralf (17:25.543)
as well as lower costs. So with Flash, Google is trying to admire or to resolve that request as well as they are providing it then. OK. There are some limits for Gmini Flash, as I've seen as well. So it is limited to 1 ,000 RPM, so requests per minute.

And but the request per day limit was removed.

Ralf (18:04.327)
What else can I say about Google Flash?

Ralf (18:14.343)
Nothing specific. It has mobile support. So Google AI Studio has a mobile support with a dark and light UI scheme, which is for developers really cool, I would say. Yeah, there were other topics at Google as well. So as Microsoft announced on their build event,

as well as Google did. So they are more and more taking into security and are advancing any security issues. So regarding privacy concerns, Google showcased a demo of AI -based call scam detections during the I .O. That means when you're an Android user,

or you somehow added Google's GminI or AI capabilities to your software regarding or doing phone calls, you can now detect scam calls here. And they also updated security features. So they have new security and privacy protections for Android.

including on -device live threat detection and better safeguards against malicious applications, which is pretty cool, I would say. What do you think, Michael?

Michael (19:46.006)
Absolutely. And security is key nowadays. So whenever you have any discussion about AI, you always have a discussion about security, data protection, and both is really, really close to each other. So it's important to highlight this, not to just assume or to...

think it's maybe something everyone already knows because we are good or we are, I don't know. So it's something you have to highlight and it's really, really important to get this into any announcement into especially new products to ensure it's on point and you don't...

scare the possible customers and users in this case.

Ralf (20:46.855)
Yeah.

Yeah, that's true. Then there was something about Gemini Nano. Did you see or hear anything about it?

Michael (20:59.47)
No, I think I missed that one.

Ralf (21:02.823)
Okay, so it seems that is the newest thing they want to shoot out coming soon, they say, with promising exciting features for mobile devices. So a more optimized version of Gimini OnePoint Flash or Gimini OnePoint 5 Pro, so that they can utilize it on a smartphone, as we both know that they are the development of

NPUs for smartphones is ongoing or the support of the general support for Gen .ai is going on smartphones as well. So might come together with that. I'm also looking about the comparison of Gemini 1 .5 Pro and Flash, what we were talking about.

So the costs from Gmini 1 .5 Pro additionally to the 20 bucks, you're going to have $7 per million tokens while you have a token window available with a baseline of 1 million tokens and you can spread it up to 2 million to a waitlist. And...

The Gmini 1 .5 flash has the same token window, but the cost starts per 35 cents per million tokens instead of 7 dollars, which is

Michael (22:40.334)
That's a big difference.

Ralf (22:41.767)
big difference. Both supporting text images and videos in their multimodal. I'm not seeing here any like...

So it seems to be limited to text images and videos.

Ralf (23:10.183)
No. No, no, no. Just wanted to drill into it and figure it out. So you were referring about something like long context. What's exactly the meaning for this?

Michael (23:10.286)
That's interesting.

Okay.

Michael (23:28.142)
For example, if you have a...

Michael (23:33.742)
a long document.

you can continue to work with it. So it's.

It's not cut it on a very short, short, it's kind of definition context. And you have to run it maybe on different parts. But that's a.

Yeah, that's actually the improvement you feel and you work with large data.

Ralf (24:22.055)
May I jump to your side for short?

Michael (24:27.214)
sure if you have more details.

Ralf (24:29.159)
Yeah, so we've spoken already about the window token, the token window, which is now one million tokens. And every phrase you fill in the chat or you give to a Gemini or ChetGPT or whatever is split up into tokens. So the longer the...

conversation is going on or the documents are, or the bigger the documents are, the more tokens you'll have. And with that blowed up to 1 million tokens, the context can stay longer, which means when you have a discussion or let's say a conversation to Gemini, you can, it can now refer back to 1 million tokens.

So what exactly happens within a conversation is to keep the context, the whole conversation is per request, send it to the LLM that it can have the context. Cause otherwise, if you don't do so, the context will be lost. And the next response will just refer to the question you were asking.

If that question has a relation to the former context or the former conversation, it won't be able to answer to that. And with that, this is the meaning of the longer context or the larger context here. It can now handle up to one million tokens.

I need to dig a little bit into it to explore what exactly can be one million tokens in a conversation. But I guess this is what this is going to be like. What do you think, Michael? Is that explanation the correct one for the largest context or long context?

Michael (26:37.838)
no.

Ralf (26:39.559)
Okay.

Michael (26:42.478)
Of course, that's a better description for my generic understanding. So yes, that's a good one. To leave a little bit the Gemini 1 .5 area, I would like to highlight something we have seen from OpenAI already, which is called VIO.

Ralf (27:11.463)
Mm -hmm.

Michael (27:11.854)
I'm not sure if I pronounced it correctly, but it's a video, generative video model. So they are, with video, you are able to create your own videos just per description. And that's a, that's amazing. So this is, if you have something as a source,

And you just use it as a source, as a description source to create an

a video, I'm looking for something, moving pictures. That's the wrong description, but you get a video out of it. It sounds like something small, but you still have the challenges with different aspects. If you have a person, you have maybe different angles, you have different structures, you have to keep the form to keep it almost the same person.

If you have moving areas around it, you have to work with it and keep it the same frame by frame. And that's a huge step with Vuo in this case. There was something that blew me away.

Ralf (28:36.263)
Yeah, I like... sorry.

I guessed you were ready to let me jump in. Sorry. Go ahead.

Michael (28:49.006)
That was the last phrase in this place.

Ralf (28:51.207)
something.

Michael (28:56.142)
You can actually, if you don't want to, let me talk about a little bit more. Right now it's limited to 1080p, but I think that's good enough for the first part. So when you start with the description, and we have seen the demo case of a pretty short prompt of a specific scenery.

Ralf (29:11.975)
I mean.

Michael (29:26.446)
and a video coming out of it just with capabilities matching pretty good, of course, because it was a demo on public stage. It was a kind of prepared, I think. And you can just use it as a material. You can just...

Ralf (29:43.687)
For sure.

Michael (29:53.326)
You can decide the length and you can maybe just also some part was to extend the video just by click about a couple of seconds or even make it longer and still keeping the content. So it's there just in a second. That's interesting. Yes.

Ralf (30:16.103)
Yeah, the videos can be like over a minute, right? And then...

Yeah, you can use phrases like cinematic style or having like so the big issue here is or the big benefit here is that you can use your natural language and view has the capability to understand it right. Now, so you can make use of words you know from like.

a common telling of give me a time lapse or produce an aerial shot of a landscape and you just specify the landscape and it can direct that video for you and we'll output it with a resolution of up to 1080p which is pretty good, high definition I would say and that's awesome.

So I've also seen a couple of those videos coming out or delivered by Vue, which is not so bad. So yes, it is as well a copy of Sora from OpenAI, but made by Google this time.

Michael (31:40.046)
Yes, and you cannot... To start with, text prompt is not the only way. You can put in some pictures, you can put in some videos and you get a new video out of it. And that's... As you said, it's pretty close to Sora. But we have a valid competitor. So in the first place, just...

Ralf (31:53.543)
Mm -hmm.

Michael (32:07.726)
in my head before this keynote, Google started something with AI and they failed with BART pretty hard. And now I see it's really close and they are on the same or almost the same level and they are really, really good on improving their own technology, improving their own models, developing, making the next step.

the next logical step for the as a set foundation where you can build more capabilities, you can create more content right out of something you give the model. So yeah, it's interesting to see Google has a high invest in this case and they are improving a lot over the last months.

or they improved a lot over the last months.

Ralf (33:08.871)
Yeah, but we have to say that Vue is not generally available for the public and it's still in test. I'm sorry. We have to say.

Michael (33:15.726)
Yeah.

Michael (33:21.71)
Sora is not available as well, so...

Ralf (33:27.463)
It wasn't, it wasn't, I wasn't taking care of what is better or not. I was just looking forward to share that it is still not generally available. So when you listen to this podcast at this time, it is still in testing. That's it.

Michael (33:36.366)
Yeah, it's.

Michael (33:43.214)
Yes.

Yes. Yes. And that's kind of annoying, right? So whenever we talk about some fancy new stuff, usually we just got a showcase on a keynote and a description available later, soon, at the end of the year, fall 2025, whatever. And you just think...

or I think, I don't know how it's on your end or on your end, dear listener. Every time I see those brilliant use cases, I think, hey, cool, let's give it a try. Let me try it out and give me some opportunities to take it to my own challenges and see how good it is on the real ideas I have at something I would like to test. And...

Yeah, that's a first world problem, I think. But yeah, that's interesting to see. I think the direction is really interesting. And Google is heading a really good way from Gemini to the different capabilities like Vio. And I've seen another use case, but I don't remember the name. I'm not sure if it's an agent.

called Project Astra. So maybe I mixed it up, but I have this use case in my head where someone is going around with Google device, Gemini is running, or the agent, I'm just, I'm not sure about that. And pointing to different places, walking around and the

Michael (35:45.166)
Camera is getting all the stuff around the area. Then switching to Google Glasses and walking around doing some fancy stuff like there are two puppies. One is a... What's the English? English... Like a teddy bear or something like that to get next to a dog. A stuffed pet. Thank you.

Ralf (36:09.927)
stuffed pet.

Michael (36:15.15)
And this is, and then they say, okay, create a fancy band name for, for this duo here. And, they got an answer and it's okay. And the next part is, well, I forgot my, my coffee. And then, you got an answer from Google glasses. You left it on the other table, just on the left side of you.

next to the keyboard or next to the Apple or whatever. So something you not asked or gave the AI to remember to do something with, but still you got the information because the AI recorded everything and was able to kind of remember those things. So that

is something what blew me away. It was amazing and scary to be honest, but in the first place, this could help some people if you have some diseases like you cannot remember everything and you can use such kind of technology for good, for bad as well, but...

Ralf (37:40.135)
You

Michael (37:40.494)
There are some huge capabilities and things about not only generating some videos, some pictures, some texts, summarizing something, get a recap or whatever. That I thought, wow, that's a use case that makes sense and feels kind of natural instead of now we create a scenario you may have in your daily work or in a daily life.

But now, this is something which happens maybe quite often.

Ralf (38:17.511)
So when I get you right, you're telling me that project Astra can be like a real -time AI assistant to me.

Michael (38:29.838)
Yes, please.

Ralf (38:30.279)
Which is, which is like hanging out with me, getting all the information from what I'm talking, what I'm looking to, where I'm going. And in, in case of I'm using Google workspace also about my calendar, my emails and stuff all around. And I can just like, by, by simply asking or, or, commanding, steer, Astra to support me. Like.

Maybe Astra will remember me to appointments and that I have to go to get in time there. Or as you said, where do I left my keys or something like that? And Astra is like remembering it and gives me the answer. So that would be a real time AI assistant.

Michael (39:01.422)
Yes.

Michael (39:20.59)
Yes. Yes.

Michael (39:26.414)
Yes.

Ralf (39:27.719)
as far as I know, Astra is still a project and, nothing to be, available soon, but yes, I'm with you. It is really an awesome, and scary, tool here. Cause I mean, the discussion about, okay, we have to go back a little bit to Microsoft because Microsoft announced a tool which will.

Michael (39:32.622)
Yeah.

Ralf (39:56.999)
Copy everything happening on your computer, you remember?

Michael (40:02.35)
Yes.

Ralf (40:03.207)
So the name is...

Michael (40:06.19)
Recall.

Ralf (40:07.335)
Recall! It gives you the possibility to recall every action you've made on your computer. So, and I mean, this is limited to the actions on your computer. And with Astra, we're going to, I mean, have it the whole time with us.

I mean, the rumors about recall were really, really heavy duty, right? So especially from Germany or the EU, there were a lot of concerns regarding GDPR and stuff. And now we're going far wider, far further to...

Michael (40:32.462)
Yes.

Ralf (40:57.031)
to let AI into our life with that real -time AI assistant. I mean, that's really... Yes, something we have to think of. We may need to learn that this is the way the future is going to be. I'm not sure. I mean...

Michael (41:21.166)
There are some movies out there and some games about what's happening if you are so related and dependent to something like an NNI, like with those capabilities. Yeah, I agree. And as you said, it's a project. Yes. And usually what happens if a project is really successful at Google, the responsible person,

Ralf (41:23.207)
As some.

Michael (41:50.926)
we'll do another job in a couple of months and the project dies. Sorry for that, but we had this with very successful projects and products actually. But the idea behind that, and if it's just something another company picks up and go the extra mile to bring more features, more secure capabilities into this, that's...

Ralf (42:04.327)
Yeah.

Michael (42:21.102)
That's a way I think it sounds brilliant. And of course, it's Google. If you think about some security concerns, you hear about Microsoft, Facebook, Google and...

Yeah, that was the first three I would say that you hear about.

Ralf (42:46.727)
OpenAI, you have to name OpenAI in that case.

Michael (42:49.774)
It's not in everybody's heads. So I think it's... Yeah. Anyhow. So let's see about this part as well. So let's see how security will be, how the actual idea behind that will be if it's something they are just checking out today and in five years we have a product out of it or maybe sooner or never.

I just saw this use case and this use case was something what blew me away as I said. Yeah, because for me always the interaction with AI technology, with a computer, with any device and intelligent device, so -called intelligent device.

Ralf (43:26.727)
Did it really? So maybe I can surprise you today.

Michael (43:48.11)
It's just on your prompt. It's just with the information you gave to it on purpose. And you have to work with, you have, I forgot to mention something, then the information is wrong. You have to put in a document to get the right context into something. And if you have an assistant like this helping you over the day and...

you're getting into something like, I forgot something to add, I forgot something to use, or I just forgot where it is. It's so on another level, I would say that it feels more natural, like a second mind to remember something.

Ralf (44:46.407)
Was that a question to me?

Michael (44:49.646)
No, it was more.

Ralf (44:51.847)
Okay, the question to Astra. Got it.

Michael (44:54.382)
It's more, yeah, on a, no, I'm missing the English words. On a high level to may have something to think about theoretically possible. On a two. Yeah.

Ralf (45:12.551)
It reminds you on stuff in any case. It can take over some organization for you, for instance. Yep. Still, it feels spooky to me.

Michael (45:33.934)
Yes, absolutely. But do you remember?

Ralf (45:35.303)
Well, we can switch to another thing, which Google also came out on Google AIO, which is LearnLM.

Ralf (45:46.311)
So LearnLM is for sure a product out of the family of Google's AI models. And it is based upon LLM as well. For sure, Jimmy and I was built in to be an expert in subjects. And with LearnLM, they are going to have the idea to support students and...

to learn or for their learning by when you do a search on anything that they support you with images, videos and whatever. So you might have experienced that circle to search on Android devices, which gives you search.

search request and it gives you results about that search. You now can do the same like on stuff which is like a school lesson or on a special mathematic stuff or physics or whatever problem and LearnLM will help to solve the question about that.

Therefore, it's going to give you whatever, a video, some answers, like a summarization of explanations, highlighting the sources and stuff. And it is a little bit like Microsoft's Orca math, because it has the ability to answer.

mathematical requests as well. Google says it out benchmarks that Orca math AI model. So it says it is with math better in benchmarks than GPT -4. So LLM is specifically fine -tuned to only respond and find answers based on educational research.

Ralf (48:10.047)
That's cool. So with other words, LearnLM is not having the capability to support you to plan a trip or to find a restaurant. And this is very specific, though. And Google says they are working with educators and the pilot program on Google Classroom so that they can use LearnLM to simplify.

lesson planning and learning. So I'm pretty much looking into some of these ideas, how they're going to develop in the future. AI assisted learning seems to be, yeah, is there still the interest of a student to learn if they can ask AI for everything? I mean, on my opinion, you shall have the

the better understanding of everything. And I would love to see that AI is supporting you to understand, but you have to understand and you have to have knowledge. It is not that you just can utilize AI to make everything for you. Then we're going to go for to be a silly civilization in the future. Yeah, so that was the thing I wanted to throw in.

Michael (49:35.246)
I agree.

Yeah, and I totally agree. And I think that's, that's maybe something you can get out of this new way of learning just by, for example, using it to get more additional questions and like a, like a, like a teacher in your pocket. So if you, if you have an issue on your homeworks, for example,

And you can interact with something like this AI that can help you to understand the benefits of whatever you try to learn. And you may have the better results. And the may was on purpose, not on a phrase part in this sentence.

I hear about some students just talking about, now I've learned everything the teacher told us. And I will forget it in just two weeks, but for the exam, I am prepared. So they learned everything. But they just, they just learned the context, but the content, but not the context. So they miss.

the actual reason why something happens like it is. And if you have someone who can answer the question or give you more examples to maybe solve a mathematical issue and you don't understand the formula, you don't understand the reason why this or that happens, you can go for it and use it to support you instead of just learn the answers.

Michael (51:30.542)
because you have a cheat sheet or something like that or something. Yeah.

Ralf (51:35.495)
Yeah, so ideally that will help students understand complex information better. Got it. And it fills a gap, right? So we have a gap of teacher and...

Michael (51:41.518)
Yes.

Ralf (51:51.943)
As well as our classrooms are kindly growing. So we have like 30 students within a classroom or even more. And it is tough for a teacher to get to all of them around to answer their specific questions. And here you're right. LearnLM or something similar can jump in to give that information at that moment when the student has the question and explain it maybe on the same level, like a teacher can.

Yeah, that's cool. Cool. As well, this use case is there and it can be very, very useful and supporting to keep our knowledge and to keep people educated as well.

Yeah, you're right. Yeah, I was a bit looking from the dark side from the first moment to it.

Michael (52:44.11)
today.

Michael (52:48.366)
Hehehehe

And as always, I'm looking from the bright side. That's another perspective. I learned something new. I talk about the bright side. Okay. I'm looking at the time and I'll ask you, we were talking about Google quite a while.

Ralf (52:53.767)
Yeah.

Ralf (52:59.423)
Cool.

Ralf (53:09.191)
We're not ready yet to stop talking about Google. So we have one thing we have to mention. So we have to talk about gems.

Michael (53:15.406)
What did I miss?

Michael (53:19.534)
yeah, I've seen that and yeah, yeah, yeah, yeah, yeah, let's go ahead, let's go ahead.

Ralf (53:22.647)
You

Ralf (53:26.247)
So you can create your own personal assistant with Gems. Cool, I always mentioning Gems. And yeah, it feels like a custom copilot.

I don't know how do you feel or think about gems.

Michael (53:47.982)
without any hands -on experience. It sounds good, like a personal assistant for a specific topic that can support you. But yeah, it sounds interesting. I'm not sure. It feels like Google tried a little bit too hard to sell it, from my point of view. So I'm a bit...

Ralf (54:15.239)
Okay.

Michael (54:19.47)
I feel a little bit, I feel a bit resistance on my end to say, yeah, that's a good one.

Ralf (54:31.815)
So gems is something you can set up for yourself and it's personalized and it can be like a virtual companion for you. Like a gym buddy, a coding partner, or you have a creative writing guide or something like that. And you can set it up very easily on your own.

Michael (54:32.014)
What are your thoughts?

Ralf (54:59.719)
And Google states that they are designed to supercharge your creativity and productivity. And you can set them up on the Gemini website.

Michael (55:06.638)
Yes, I like those phrases.

Ralf (55:13.319)
To be honest, I've never visited Gemini since 1 .5 is out. I do have to explore it a little bit more. Okay, yeah, though we are ready with all that stuff came from Google I .O. I do not feel that we have something left. But there is another company and that company came up this week with its...

WWDC, so the Worldwide Developer Conference. They claim the name for it. And that's pretty exciting for you, isn't it? So you are a Google and Apple fanboy and that must be a bright topic for you, isn't it?

Michael (55:53.102)
Hehehehe

Michael (56:05.422)
It's, there are two, at least two traditional events as an Apple fanboy. You will watch at least the keynote. The first one is, when they announced the new iPhones. And the second one is, the WWDC. And the WWDC is something they have in place, rolling out every year since.

ages. And they use it to announce the new capabilities of the operating systems. And if you think on the ecosystem of Apple, they put an own operating system on every device, like iPad has its own OS.

Ralf (57:03.079)
Specified OS for every device. Or dedicated.

Michael (57:07.886)
Yeah, but once again, it's in quotes, I would say. So the iPadOS is very close to the iOS, which is on iPhone. TVOS, there's now for the Apple Vision Pro, the Vision Pro or Vision OS, and you have the Mac OS. So I have...

I missed the watchOS and I think that's it. So this is, but that's something you can see every year. Where is Apple investing in new capabilities? There are some fancy stuff, which is maybe already known from other companies. This is, if you have a broader, if you have a wider.

view on other products, like comparing Android with iPhones. You are just thinking, okay, now Apple is doing the same as Android does for quite some time, like moving icons on the iPhone.

Ralf (58:21.383)
Yeah.

I was really surprised by that announcement. I wasn't aware of that you couldn't do that until now.

Michael (58:31.598)
That's a good one, right? So from my point of view, it's kind of ridiculous. So as well as a fanboy. So the first hour or first half of the WWDC keynote was just, -huh, -huh. Okay. Yeah.

Ralf (58:53.063)
Hahaha

Ralf (58:57.703)
Okay, okay, okay, got it, got it.

Michael (58:59.63)
Nice benefit, yeah. But then... And I use almost every product of them. So I have almost everything of those OS except the Vision Pro in my household. And I know my daily use. Yes, and maybe it's on my list, on my bucket list, but maybe...

Ralf (59:14.631)
Which is coming in June to Germany, I guess.

Ralf (59:21.735)
Maybe, -huh.

Michael (59:24.622)
It's a July, actually July this year. But anyways, but the second half was, in the first half, they just announced feature by feature by feature by feature by feature. Small. Nice. And then they were talking about AI and they reclaimed AI as Apple intelligence. And I said, Whoa, that's ballsy.

Ralf (59:27.207)
sorry.

Ralf (59:45.319)
Yeah.

Ralf (59:50.371)
I was wondering that they don't name it IAI.

Michael (59:58.254)
I thought it's really, really ballsy to reclaim the whole phrase AI as now it's Apple Intelligence. But yes, they did it. And what should I say? You remember about the first feedback about Gemini for workspaces? They copy the thing Microsoft does with OpenAI support on their products.

on those products. Yeah, Apple is doing the same. Kind of, kind of. So if you live in the Apple ecosystem, you have a voice assistant named Siri. And Siri is good for turning lights on and off and asking some really basic questions. And now...

Everyone was hoping for AI support into Siri and yes, it's coming into Siri. Siri gets a huge leap. But the first step they showed was from Apple intelligence was you can create recaps. You can create rewrite or you can rewrite your text. You can, you can prove the language.

You can check the tone and the basic stuff, I would say. So the fundamental stuff, if you start with AI integrated in your product. And as I said, it's okay. That's a good one. You have to do this. It's necessary. Otherwise, you have some issues on the market. But yeah, it's impressive on other levels.

Let's go this way because they do it on the device. They do it on the device.

Ralf (01:02:04.358)
Partially, they just do it partially on the device. So it is not.

Michael (01:02:09.518)
Yeah, and that's something. Let me try to recap this phrase. The most of the parts will be done on the device. And if it's not enough, they extend the security boundaries of the device to a private cloud or something like that.

Ralf (01:02:30.183)
to the Apple Cloud.

Michael (01:02:31.95)
Yeah, but it's not Apple Cloud. It's something where nothing is stored. Nothing is...

Ralf (01:02:39.655)
They claim that nothing is stored permanently.

Michael (01:02:44.43)
Yes, they claimed it. And they told us they use external auditors to verify this. So they claimed this is proved by external auditors. I don't know the right...

descriptions they used. But yes, it was something not related to Apple to verify this is true. And yeah, the first part.

Ralf (01:03:29.607)
So now let me summarize a little bit because we are going to deflate it a little. So we have now artificial intelligence by Apple, which is named Apple Intelligence. And you're claiming that Siri will get like a chat GPT intelligence into it so that you can have conversations with Siri, which wasn't.

wasn't the capability of Siri before. So you claim that Siri was just able to switch on off lights, do simple tasks. And now that'll become two conversations. So you're going to have something like your own Star Trek computer or whatever you name it.

Michael (01:03:59.918)
Yes.

Ralf (01:04:22.535)
And then we can have now capabilities within our tooling on Apple devices. So that supports for writing and stuff. And as well as we have the capability to let Siri or whatever device application it is to create AI images within a message, right? And it's an iMessage.

where you can now create AI -made images, right?

Michael (01:04:56.462)
Yeah, let's don't squeeze it to just some side notes here. You have some capabilities for Siri, which are pretty new. I'm not sure if any other device or vendor is currently doing the same because Siri is able to interact with your device. So you can ask.

Ralf (01:05:00.935)
BLEH!

Michael (01:05:23.086)
Siri to create something, to do something, not only change some settings, but do something really, really from a bigger part. And you get on the device directly. So it's not like just telling you how you can set something up, but it's actually done on the device.

Ralf (01:05:47.367)
Mm -hmm.

Michael (01:05:50.094)
And you also get some information on the device from different sources. So it's a kind of the technology in co -pilot where you get multiple sources together and ask for example, to get a summary of a specific topic or something like that. And you get the sources, everything together. In this case, you get some sources like email, iMessages, notes.

And everything together, for example, to get a quicker answer. You don't have to look anything up anymore. You can create short answers in iMessages based on the context. So you get a pre -written answer just by selecting two, three options asked by Apple intelligence. I'm not sure if it's Siri or.

how the integration actually is, but you get details from the whole device. So everything what's on the device, what's known from and accessible for Apple intelligence is able to interact with Siri or the AI capabilities itself. So...

It sounds really powerful, but it sounds really powerful if you are just living in the environment in the ecosystem, in Apple ecosystem. I would say in almost every part, like really using the mail app instead of, I don't know, Google or Outlook or whatever, an external app. So party app, whatever.

also for every node. You use the ecosystem, but the fancy part is there's an API, so developers can use and build on those Apple intelligence features as well. So it's not locked just for Apple products. You can hope, at least.

Michael (01:08:13.198)
Some developers will use the SDK for it and create some new stuff and give it also to this new assistant.

Ralf (01:08:25.351)
cool so we can build our own Jarvis here.

Michael (01:08:30.542)
Yes, yes, for example. I'm not sure if a kind of something like Microsoft, I live in almost every part of my work in the Microsoft ecosystem, this Outlook on my iPhone, Outlook on my Mac, Teams, of course. If Microsoft will use it to have a competitor using their data,

Ralf (01:08:32.487)
Cool.

Michael (01:08:59.822)
in those applications comparing directly with Copilot? Let's see. I doubt that. But yeah, the general idea behind that, except the silly part of naming this, all the stuff like Apple intelligence, that drives me nuts, is from my point of view,

Ralf (01:09:06.887)
I don't know. I don't know. You have doubts?

Michael (01:09:30.414)
Someone expected just a chat GPT integration into the ecosystem and Apple is doing nothing on their own. And they got proof wrong. Now Apple has its own AI technology and is involving. Hold on a second. I see your face. Sorry listeners, you don't see the face, but, and it's not.

based on OpenAI. But, and now that's coming to the face, I'm currently seeing from Ralph, there will be an integration actually of JetGPT 4 .0. So there are two parts here. Siri.

And the Apple intelligence is not the integration of a JetGPT 4 .0, but there will be a JetGPT 4 .0 integration. And it's, I don't know, they called it AI for the rest of us. That sounds like we are late to the party, but there's still something there.

Ralf (01:10:44.423)
but in English.

Ralf (01:10:52.135)
I'm not sure the summerization I've watched about that is that they claim that they've built it into their products, which is upcoming. And I'm not sure if Apple intelligence isn't made out of one of those components delivered by OpenAI regarding the time they had to build that up.

Michael (01:11:17.134)
But it's not mentioned.

Ralf (01:11:20.455)
I'm not sure, I haven't had a look into the code at my end.

Michael (01:11:22.062)
The first, I've watched the keynote, of course, as an Apple fanboy until 1am because I was not able to watch it live. Anyhow, they claimed the OpenAI integration on the very end after they pointed to Apple Intelligent for 30 minutes straight.

So they really split it up to two different parts. The first part was Apple intelligence. And the second part was a minor part. It's not half half. It's not 50 -50. It was kind of 90 -10, 90 % Apple intelligence and integration into the OS. And the other one was, yes, we also have a JetGPT 4 .0.

And an interesting one from my point of view is if you have a subscription with OpenAI and use ChetGPT for maybe via app or another ways, or maybe any other integration, if you have a subscription, you can use it and get the benefits out of it. If you don't have any...

subscriptions, you can still use the model and it will not the data that you give to chat GPT -4 .0 will not be used to train the model. So they made a new deal, I would say, to keep it more private what you do. And you can still use the capabilities of.

Ralf (01:13:05.319)
Hmm.

Michael (01:13:20.462)
chapter CPT 4 .0.

Ralf (01:13:22.247)
Okay, cool. Yeah, definitely. Okay, so not so big news, but big news, I would say. And I mean, at the end of the day, one thing came up to me, which I wasn't aware of. The iPad calculator is there.

Michael (01:13:24.142)
That's an interesting one.

Michael (01:13:46.366)
You

Ralf (01:13:48.743)
Sorry, I have to use it.

Michael (01:13:50.862)
As I said, ridiculous, ridiculous until you see the magic that you can do with some math. I think it's not the calculator itself, but there are some that there's a new app math, something. And you can draw a graph out of a formula expression. Yeah. And you can change something in real life. You can.

Ralf (01:14:04.614)
Yes.

Ralf (01:14:12.807)
expression.

Michael (01:14:20.75)
That's really fancy if you do a lot with math or you are in the education part and you are still learning something about math. That's interesting.

Ralf (01:14:31.239)
Yes, I found it cool. So I was, it has to do something with like learn LM because it is not without that. So I wanted to mention that they're doing as well cool stuff here.

Michael (01:14:39.662)
Probably.

Michael (01:14:47.598)
But the first was really the new app on iPad is the calculator. And we used the extra space for history. And I thought, of course. Gosh, are you kidding me? Anyway, they made it a cool thing at the end. But the first part was, yeah, as I said.

Ralf (01:14:52.839)
calculator. Yeah.

Ralf (01:15:11.527)
Yeah, yeah, yeah. You gave me a good feeling about this. I was really the same. I was like, okay, now they have a calculator. Even in the DDR, they had a calculator. I mean, it's not that fancy stuff. What are you doing there? I wasn't aware of that this was missing on an iPad.

Michael (01:15:22.894)
you

Michael (01:15:28.622)
Yes.

Michael (01:15:35.342)
because no one is using a calculator on their iPad.

Ralf (01:15:38.463)
Yeah, they are more creative. Okay, so that's a wrap of all the most important events we had in the past. We do not have to create an extra episode for Apple. I was here.

Michael (01:15:43.214)
Absolutely, absolutely.

Michael (01:15:58.35)
because you shrink everything about apple in this place. I'm a fanboy. I want to celebrate this.

Ralf (01:16:01.575)
We...

This is cool. You can, you can. You celebrated it a lot. I mean, about half an hour. And I'm sorry, but I had a doubt that this would fill up a whole episode from from Decode AI. As in the past, folks out there listening to us to jiggling about different stuff here, we will add some important show notes.

and links. No, we will not add important show notes. We will add some links to our show notes so that you can have a read about all the stuff we were talking about. And if you have any questions, please don't hesitate to contact us on any way, which happened by the way already. And I'm thankful for that, that you are reaching out to us that gives us the feeling that we are doing something which is demanded by you.

Thanks for that. And Michael, it was a pleasure talking to you this time again. And we're going to see us in about two weeks. And we have a guest then there, I guess.

Michael (01:17:16.238)
Yes, we do have a very special guest actually, who's named Hank Bowman. I'm not sure if it is...

Ralf (01:17:25.895)
He's a Dutch guy, it's okay if you say Bollemans.

Michael (01:17:29.198)
Okay, talking about AI and global community. So stay tuned, stay interested to hear about this in our next episode. And once again, thank you for listening. Hear us.

Michael (01:17:51.854)
And we will hear us on the next episode. So, see you then!

Ralf (01:17:56.551)
Stay tuned, stay interested, listen up. Here we go. Bye bye, take care of all. Thanks for listening.