Introduction

We will try to remake the presentation that Ayla made starting from the Bosnian text of the speech in 15 minutes. So we will have a presentation in 15 minutes that Ayla could have used that might be better or worse than what she put together. Let's see.

Live demo and expectations

Let me start from two important points. This is a live demo.

Things can go wrong. They might go wrong. And it's normal.

Things go wrong with AI. No, it's not a car. It's not a deterministic technology. It's not something that always works as you expect.

But the important thing is when it starts being right 80% of the time, because that's when you start saving time.

The second thing is I did wake up at 2 AM this morning to come here. So I'm not as fresh as I would like to be. So I'm putting myself out there, all right?

Tools and setup

I will use three different applications.

How many of you have heard of Gamma? Okay. Say probably 20% of the room.

How many of you have heard of SuperWhisper? Okay. Fewer.

How many of you have got the LLM, so Cloud Charge GPT, linked directly to the email?

Okay, so it looks like you might all be able to learn something new today, combined, I hope. And even if you know these things, you see how I do them, and then we discuss afterwards what you would do differently, if you do it better than me. I'm really always very happy to take any feedback on board.

Using SuperWhisper for dictation and email access

First thing we're going to do, and I'm using Super Whisper right now, which is a dictation app, you will see. I will zoom a little bit so you can read. Let's see.

Hello, do you speak Bosnian? Let's see. OK. Great.

Yes. I speak Bosnian, looks like. Great. Great.

In that case, I would like you to go through my emails. There should be one today that I received from Ayla with the text of a presentation. Can you please find that email and confirm that you have access?

Not bad. It didn't get your name right, Ayla? Yes. Good advice.

Because of the gym. Say Ajla, and then you can know.

OK. Tell me about it. Not at all.

My surname is Piano.

OK, so you see what this is doing is going through my email and finding the email that Isla has sent. So this is saving me a lot of time. I don't need to go to Gmail. I don't need to search for what Isla sent today.

I can just see that it has access to that email. And that, yes, there is a presentation about artificial intelligence in Bosnian, Croatian, Serbian. Great.

From speech text to outline with an LLM

Now I want you to imagine that you are an experienced AI researcher. And on the basis of that speech, I would like you to put together a short outline of about eight cards that I can then put on gamma.

And so it's going through the document. And it will put together the text of the presentation. What do I do?

I am losing pieces, sorry. So, all right. Doesn't look too bad.

What do we think? It's decent. Decent.

What would you change? Probably the phone.

Can we open this in a canvas, please? That I cannot do yet.

Editing and refining the outline

So what I will do, and I do this quite often, I ask Claude to open this in a side-to-side window. So now what I can do with this, I can basically edit the text in real time, still speaking to the model. So I don't need to copy and paste it anywhere. I've got it here.

Iterating on content in-context

One thing, for example. I don't like is the visual suggestions here, because I don't need them for gamma. So what I would say, OK, this is great. Can you eliminate the visual suggestions from all cards, please?

So now if it gets it right, you should see that, yes, got rid of all the visual suggestions. So it's getting rid of those. So again, it's something that I don't need to do manually.

And I tend to write a lot of work documents in this way. Depends what I need to do. But if it's something where I need to take information and move it somewhere else, then I would write them this way. If I need to be a bit more creative, then probably I would start from the blank sheet of paper.

OK. That is done.

Building slides in Gamma

What are we going to do is now to put this into gamma. Before I go into gamma, is this making sense? Do you have any questions? OK.

So you see you have got different options here. So we are going to work with the presentation. And we are going to generate it. from notes or an outline.

Okay. Might do some basic edits. We don't really need this for now. So I will get rid of this. Great.

Generating a presentation from notes

Now it's gamma. It's going through the text. By the way, this is available for free for the first few cards, at least.

This was the title, right? Something like that. So we've got the different cards here.

You see that you can choose how much text you would like. We don't want a lot. You can choose your template. Here we are going to go with something quite simple.

But you can import your own templates based on PowerPoint or a PDF. You can select your image model. You can select the image style.

Let's generate it. Let's see what happens. And so we see the presentation is being created. It's taking a bit longer to generate the images.

But it's looking good. Not bad. What do we think about this? Can we read it?

We're still generating the pictures. I would suggest going up. Yes, let's see. Not bad.

Improving slides with agentic edits

Now, I want to show you a few things that we can do. Well, I think the presentation is not bad. It's working well. There seem to be all the key elements.

I want to see, what could we change? This slide, let's see. Can we make this a bit more visual?

So this is an agentic function. You can speak on the right, and it would make changes to the slide. Let's see how that works.

Internet is being a bit slow. It did add some infographics. Interesting.

But it also changed the pyramid. Can you keep the pyramid in the AI capabilities slide, please.

Styling visuals to match a brand

Now something that I want to show you is that you can actually decide also the style of the different images. So let's go back to our cloud output. Now let's say Let's do things properly.

Deriving a visual style from the website

So I'm going to go, Ayla has a website here. 1So we are going to ask to come up with a description of the style of her website that then we can use for the images. Great, now I want an image style for my presentation. And I want you to describe in four or five sentences the style of this website so that I can use it as the theme for my presentation.

I need to copy paste this. Oops. OK, took the text in the wrong place. Now we are .

So of course, I could go through the website and do it myself. But given that I'm asking one LLM to describe something for another LLM, these methods should probably be more effective.

Sophisticated, minimalist aesthetic with generous white space and clean typography. OK, interesting. OK, let's take this.

Applying custom image styles in Gamma

Now we go back to gamma. And we take one image. OK. Take this, which I don't really like.

And we put our new style from the website. And we see if it's more similar. I hope it will be.

In the meantime, I also want to show something else. That's another thing which I find very useful. Let me save this image.

You see it's, again, a picture of Isla. OK. So you see, change the image. And now it looks like something which is probably more aligned with the original website. Yeah, so I think that worked.

Transforming photos into artwork

Last thing I want to show you, imagine we want to replace this with the picture we just took. But then we actually want this to be more of a, a painting, not just a photograph.

So can you make this a pop art painting? So I'm asking. This is just my own personal preference. Pop art could be something else.

But I'm asking to transform this picture into something which looks more like a painting, which might be consistent with my objectives. How about an impressionistic artwork from the 19th century?

Here I'm really pushing it. Let's see if it works. But it's quite flexible in terms of interface and the sort of things that you would be able to ask.

Not bad. All right.

Q&A

I hope this was interesting. Any questions? Any curiosities on gamma, the connection of Claude to the LLM, Super Whisper? Anything you're curious to know?

Yes? Is my data safe when I ask Claude to analyze my dream maker?

Data privacy and model training settings

So it depends on what you mean by safe, and it depends on what account you have. Generally speaking, if you have a paid account, regardless of the type of account that you have, you are able to ensure that your data, if you want, are not used for training purposes. So there should be, where do I look for it, in cloud.

I don't think it applies to me because I have a pro account. This happens by default in the pro account that many enterprises use. But you are able to decide not to use your data for training the cloud models, which means that they should be handled safely. in that sense that there is no risk that they are used to contribute to strengthening the models for other users. So they will not be part of the training process.

Exporting and code generation

I saw a hand there. Could you export what you just made as code? As code? From gamma? Yeah. Could you export it to a code?

That's a very good question. And I'm not technical myself, so I don't know the answer. I think I can jump in. So these are the presentation making.

BAM also offers the website presentation. So I'm pretty sure in that case, when you're building the website, that can be exported. With presentations, I think you can only export Yeah, if we look there, that's a good point. Thank you, Ayla. If we look there, look on chat, let's see.

You can post. Yeah, I think for presentations, you would only be able, you have these options. But something that we can try and do, is this going to work? Okay, here we are stretching it. Let's see.

What coding language would you want this to be? Well, I mean, I would be more comfortable with something React-based, but whatever. OK, let's see.

Now, what I'm trying to do is see if we can get from the one slide of the PDF to some source code to reproduce it. Let's have a look. This is a version of the presentation. I would want you to take slide one and write some JavaScript that is able to reproduce it on a website. I've never done this before, so it might not work.

It is doing something. I don't know if it's going to do the thing that we actually want it to do, not being technical. So these things typically work.

Or at the very worst case scenario, it would give you a starting point. A starting point you can work on. From my limited knowledge, it looks like something that's picking up the right elements, at least. All right.

Conclusion and handoff

Thank you very much, everyone. It was a pleasure.

Andre, on to you. Let's set you up.