Introduction: Live Demo Overview

Let's get started. So I promised it's going to be a live demo.

So I think I could do a few things and I would like to show a few tools and a few techniques that could be useful.

What do you think?

I think two main things I had in mind.

Scenario Setup: From Research to Presentation by Tomorrow

One is doing some research and a presentation. So I will do two things.

Have you had the chance of using perplexity at all? How many of you have used perplexity?

Okay, gamma? Okay, so that could be useful.

How many of you have integrated Claude in Gmail? No, Claude, the Claude model, okay.

So it looks like a good combination. We will all learn something new today.

I want you to think about, I mean, almost play a scenario of 6 PM. The day is ending. And we need to do a presentation for the next day and prepare the day tomorrow. So how can we do it with AI tools in the best possible way?

I want you to keep in the back of your mind and what I will try to show you. Now, let's imagine we need to do this presentation.

So, I will actually, who has used Napkin before? Napkin AI? Okay.

Choosing the Market to Analyze

So let's choose a market, a market that has a lot of public information online so that we can actually research with AI tools. Any ideas? Could be pretty much anything.

Let's imagine we need to prepare the presentation. It could be for the senior leadership. And let's imagine that it's on a specific market.

Let's pick a market, could be anything, gyms, restaurants. Food, what specifically?

Restaurants, let's do restaurants.

Defining Presentation Goals and Key Questions

Okay, what do we want to know about the restaurant market? What do we want in our presentation? Let's imagine we maybe need to present it to the leadership to review the market to invest.

Maybe we want to know main alternatives.

Using Napkin to Turn Notes Into Visuals

What is the most review on internet okay anything else maybe growth and trends now I want to show you what napkin can do just on the basis of this text

it will produce an infographic or a design. You see it's generating. Not bad.

If you don't have a visual intelligence like me, this is very useful. And you can export this as a PowerPoint, and you can play with it.

Interesting. We have got some features. Now, what can we do from this?

Tooling Deep Dive: Voice, Prompts, and Research

I'm going to go and charge GPT.

Super Whisper for Fast Dictation

How many of you have used Super Whisper? Have you heard about Super Whisper?

So it's a dictation app that works better than most embedded functionalities in existing apps.

And you see that whenever I put the space bar, it will actually start recording what I'm saying. And so this is useful for two ways. It makes communication faster, but it also changes the way you interact with these tools. See, you tend to reveal a lot more when you speak.

So... Imagine... Of course, it's not working. You know, this is a live demo. Things go wrong, unfortunately.

Let's give it another try. Otherwise, we do it the old-fashioned way. Okay.

Crafting Consultant-Grade Research Prompts

Imagine you are an expert in the restaurant market. I'm here with my team. We have produced a list of different factors that we want to include in a presentation for our management. Can you write a research prompt to the level of a Bain consultant?

So here I'm asking the model to take this list of factors and to come up with something that we can use on perplexity to do the research. I will add that I want this to be 10 pages, just for the interest of time. Yes?

Can you do the super whisper? Yes. I can show you it works a lot better.

I mean, two reasons. It works better, but also I can use it on every app through a shortcut. And so you're just building the habit of actually... you actually build a habit of just clicking, I think it's option and space bar, and start dictating that.

And then there is that psychological reasons of, but that's more of voice in general. Any questions? It's curious.

nobody would infer it from your tone and you can train it so they respect it because the whole idea the whole idea is that it's not i mean it would be a very good typist from the old from the from 70 years ago where you're actually It is so good that it would infer where punctuation should go. You can just focus on the flow.

Yes? Yes. Yes, I have it in English.

Working in Canvas and Scoping to the EU

Now, we've got our prompt. How many of you have used the canvas before? Half and half, so I will probably show it.

So the canvas basically is this two-side window where you can continue to chat to the model on the left and you can make changes to the text on the right.

I would normally spend some time reviewing this. In the interest of time, can you, that stopped again. Not great.

1Can you focus on the EU only please. And so what it will do, it will go through the brief and it will make sure that it only fits into, we only research the EU specifically.

It's taking a bit longer because I'm using the thinking model, which is the reasoning model of ChartGPT, which is a bit more accurate, but would take a bit too long in this case. Okay. Switch the model.

Can you only... Look at the EU, please. And now it's only looking at, it's changing the old text so that it would only focus on the EU.

Running the Research in Perplexity Labs

What do I do with this?

I take it pretty much as it is and I put it into perplexity. And I put it pretty much as it is

Now, I will use... Thank you for spotting that. And I will use that labs feature on perplexity.

Have you used labs before? So what you will see, you have used it.

Generating Multi-Format Assets

Basically, it would create different assets for you. So it just doesn't give you the text or the information as a bulk, but you can take different bits and pieces that could be useful for different things, a chart, an Excel. It would also give you the code for the data.

What do we do while this is running?

We prepare the day for tomorrow. So first I show you how that works, and then we try to rebuild it together.

Planning Tomorrow with an AI Assistant

So I've got my personal assistant here. All I would do is prepare my day for tomorrow. 1So you will see it will now go through my email, through my calendar, it will look through my meetings, and it will then give me for each meeting a briefing of what I should know to be as prepared as I can possibly be.

I think the internet is a bit slow. Let me see. Is this running? Yes, this is going. Yeah.

Connecting Data Sources Reliably

You see, these are all the different connectors that you can link to your cloud. You can also do this with ChatGPT, but cloud generally works a bit better. So this means that your cloud can directly access information from each of these sources. Which becomes very useful because it means that you don't need to copy paste the information that it's no messy PDFs.

No copy pasting So you see it has found two meetings Let me show you they're actually legitimate meetings I need to prepare for What is it doing It's giving me a daily overview. And then, for each of the meetings, it's preparing me a briefing, going through my emails and reminding me of what I actually need to know for that meeting.

It's coming again. So it's basically saving me having to go through the emails and finding out what's going on. It's putting all of that information together.

On Reliability and Hallucinations

You might ask, is it reliable? Does it hallucinate?

On this, given the instructions that it's starting from and the fact that the source has been very clearly specified, it is very reliable. hallucinations typically happen when you are asking something that the model cannot do given the information that you have provided. In this case, there is a clear ask, there is a clear source, so the model is not tempted to invent things to please me, which is something that often happens.

And so this means that

it's a lot easier for me to prepare my day because I've got everything in there and I don't need to go through my emails and do it as I guess I would do I would typically do before this whole AI thing happened now let's see if perplexity is done perplexity is still going the internet is not our friend so let me

From Research to Slides: Custom GPTs and Gamma

I actually did, I do have a report on the restaurant market. So we can look at this later. What I want to show you is how we can create a presentation from this very quickly.

restaurant I should have it No, I'm not finding it. So, this is still going.

In the meantime, another thing I want to show you is, which I think it's useful, is the Fireflies integration. So now, Fireflies is a note taker. All my notes are in here.

So if I want to, I can take that information, and then I can use it, for example, to integrate it in a proposal. I can use it to produce a follow-up email. The fact that I have that, and I do not need to go on Fireflies, take that information, and then make use of it.

So for example, can you go through my Firefly transcripts. And can you find my latest meeting with Henry Allegra? There were a lot, right?

Yes. So you see, it's all misspelled, but I would hopefully get what I wanted to do. It's going through my Fireflies transcripts. It's looking for that meeting.

And once I found that information, then I could write a follow-up email. I could make use of that information for whatever thing could be useful for me.

Did we invite fireflies? Maybe we didn't. This is misspelled. Let's see.

Are there any questions on this? Any curiosities?

Yes. Yes.

So I'm using Claude because it's a bit more, the integration has been around for longer and it tends to work a bit better. And the integration tends to work a bit better. CharGPT has recently introduced it, but it's a bit fiddly.

All right, we have a meeting. We're over.

Yes, with the email you can. You might struggle a little bit with outside services. But you can do email integration, Google Doc integration, you can also build custom GPTs, which is what we are about to do once this is done.

Yes, it took a bit longer. I think the internet was a bit overloaded. But this is really a thing of beauty.

We have got the report. And we have these different assets. So we have got all these different tables that we can export to slides or copy, paste, or play with. And we also have the code to reproduce them.

Now, we're not going to use any of this. I'm just going to take the boring PDF.

What Are Custom GPTs (Gems)?

How many of you have used custom GPTs?

Custom GPTs, they're called gems. The way to think about it is as basically a pre-programmed version of an LLM that has a prompt embedded in it.

So if I just now take my report, It's just a report that we have produced. And I say, prepare an outline for a presentation of 10 pages.

I'm on the wrong one. Wait a second. That's what I wanted to choose. That's the one I use for my social posting.

or even here, I can just, because I've pre-programmed it so that I can click. Didn't work, of course. So normally, I would need to prompt, say, imagine you are, or give it some sort of persona or identity, explain what I want, some instructions on how I want the slides to look like.

Now, all I need to do is basically have one line in there. and it will come up with a text that I can just export.

You think this could be useful? Something that you would use?

You might wonder, why do you just have the bullets in there? Because now, we'll take this, once it's done.

Building Slides in Gamma from an Outline

Now we go on gamma, which is probably my favorite AI tool.

Got that wrong. Why is this not working now? Let's try again. Copy. Nope. Okay.

So what Gamma does, just on the basis of this, it would start compiling the presentation. And you can choose different... text options and choose different templates.

You can import your own templates. You can have some AI generated images. You can choose the model and also the style. You could have your own style.

So we are just going to keep this boring for now. The internet is not being friendly with us. It's normally a lot faster.

But, we see that just on the back of that prop, it started to put together a decently looking presentation. Is it perfect? No. We can probably make a lot of changes.

Is it better than what I would have been able to do in five minutes? Probably yes. Thank you.

Can you include brand identity there? Yeah. You can. And the cool thing about it is that you can actually speak to it and ask for changes.

Of course, if Super Whisper doesn't crash like it's doing right now. Nope, I'm not lucky.

Can you make this a bit more visual? And so it's going to, I hope, add some infographics here. Let's see.

Where were we? Oh, it's looking at the entire presentation. Why, yes. They'd make this a bit more visual.

And you can chat to it and ask for changes on specific fields. And all in all, it's a less fiddly way of playing with slides.

One last thing I wanted to show you.

Napkin vs. Gamma: When to Use Which

Napkin is really good at just taking a little bit of text and coming up with one slide. Napkin is best if you just have like half a page and you want it to summarize in an infographic. It's very good at that.

Gamma, you give an entire outline, and it would come up with a presentation. But you could also give the text for a proposal. You could ask for a web page.

You could use them together, I guess. You could put a napkin into gamma. If I were to use them together, if you want to combine them, you can definitely copy and paste it.

I can show you.

No, it wouldn't work in the same way.

Behind the Scenes: Building the Assistant and Projects

One thing I wanted to show you is actually how this whole assistant was built.

Claude Projects as Long-Form System Prompts

Have you used Cloud Projects at all? So it's the same idea of the custom GPTs. So you go inside, and this is basically a long prompt.

Look at this. It's basically saying, OK, follow these five steps.

Go over my calendar. Look at the external meetings. Do some research on each external meeting. Create briefing notes.

And the way I created this was actually chatting through the model. I want something that does more or less this. Can you come up with a description so that I can put in as a project description?

And you see that this gamma builder was actually built in the same way. There's like a long prompt inside. It's basically telling... the model, how it should come up with these slides.

Formatting Conventions for Slide Generation

And it's even saying include these three dashes because that's how gamma would separate the slides. So I can just really copy and paste it.

Conclusion and Q&A

I hope this was useful and it gives you a flavor of what we do and the approach we take.

bumps along the road, Super Whisperer not working, and the internet being a bit slow, but these tools are not perfect. And we do not want to convince you that they are perfect.

We just want to show how you can still make quite a lot of work out of this. You can save quite a lot of time out of these tools that will probably remain imperfect for a while.

Happy to take any questions.

Data Flows, Privacy, and Enterprise Constraints

so you want me so you want me to show you the data flows So at a high level Yes So I've not I've not used I've not used open AI for I've only used open AI for the I like charge EPT for this deck building.

So what has been transferred? For the day planning, I've used Anthropic. I've used Cloud. But it's not because so.

Yes. I mean, is Cloud accessing my emails? Yes. Do I trust them?

Yes. I mean, as much as I would trust Google to hold my email and my information. So I think they did look through. They do have access to my emails in this way. Because I'm uploading my emails onto the model. .

Because the obstacle to this kind of working is the cyber security policies and many companies locked inside Microsoft Word only being allowed to use Copilot and you are Asking, can I make an automation that reads through my emails from Anthropic and the IT administrator? Ah, never. It's like, it's very smart.

Every time I introduce a new tool, I have to get approval from IT. It takes a week, sometimes. I mean, we have seen it with cloud, right? A few years ago. I think it would just take some time for enterprises to adjust.

It's kind of more delicate. Then Cloud. Cloud computing is just moving data database from there to here, from there. But there was a lot of tension, right, also around Cloud.

But actually, using LLNs, can you really trust that they are not training the system with your email? Yes, totally. Once you have a professional account or a Plus account on ChargeGPT, that happens by default. Even on the free version, you can turn off that feature. With the free, you can turn it off. I don't have a free account with me. But if you go on Settings, and it's actually on improving the model for everyone, if you take that off, then the models are not being trained on your data.

You're saying whether they do it or not. Yes, I mean, there is also what is the actual risk of data leakage? So what is the actual risk that that particular email with that particular information is found and used for whatever purpose? There is also, I mean, that element of what's the objective risk of things going wrong.

I mean, every company has different, different, different. There are scientific papers that say that you can reverse engineer an LLM even if it was impossible for discussions about privacy. Yes?

Costs and Plans Needed

Quick question. Considering all the activities you performed from the prompting all the way to the PPT generation, were they all made versions of the tools? Do they all need to be meant by the Nutella membership or will they all start to move?

So you can do all of this. All I've done today, you can probably do it for 30 euros a month. 30 euros a month, because you need to be able to have custom GPTs. And that's available on the Plus version of ChatGPT. That's 20 euros.

I think the old Cloud integration should work also on the free version, I believe. And Gamma is 10 euros a month. Free. for free I was struggling I did an update before I was struggling a little bit today but I mean I've done one hour one hour long demos without any problems before yeah Pretty much.

You also get a few cards for free. I think they are a partner of ours. If you go on our platform, we're still negotiating the discounts, but you should get a good discount. You should get at least a month free. That's what we're negotiating.

Recommended Workflow: Outline First, Then Design

Is there a recommendation of steps and tools? Let's say, before preparing a presentation, is it good to follow how you started, let's say, with the prompting on how we gave you and stuff like that? I would try to get the outline in good shape and to have the information that you want. Once you have the outline, then you can move to gamma.

And I would do the same with proposals, by the way. So for proposals, I have a project. I have a project here. with all our sales assets inside in a good format. And then I would basically be able to say, draft a proposal, and it would just draft a proposal on the back of these calls.

Because remember, it's linked to my Fireflies. It's on my Fireflies, my note taker is there. So what happened, Claude can know what happened in the call. So I can basically say, write a proposal on the back of this conversation I've had with this client.

Designing Your Own Multi-Tool Workflow

I found a workflow that I wanted to tell you. Actually, this is something you can, I mean, beside all this example, this is something you can reason with AI. You know, if you want to have a workflow and you say, these are all the AI tools I feel comfortable with, some are free, some are paid, you can play with this.

So I've done that last year, something I've done thousands of times in my life is doing from A to Z research to presentation, something that would take six months to do. And I say, let's build a workflow redesign, you know, to get higher quality, top-notch quality research, all these steps that you have seen and so on. And it suggested 20 steps with like 40. And I said, well, hold on, man.

I don't have access to all this one. I have access to this 12 one. I feel extremely comfortable with this one. So can we do that? And eventually, in really less than one hour, I end up with a seven step using six, seven AIs.

And what was interesting in each step, a bit like Stefano did, it was using three or four AIs for steps. Say you take the output of this, that becomes the input, and then you should work on this, very similar to what is done.

And what is interesting, when you have your own workflow, when you have upgrades and updates on the tools, you know, I say I remember it was like a long time ago, so there was a... I think that was not yet the 3.5 of Claude. And I say, well, now Claude is doing this. And now I have the deep research. And he changed the entire workflow. And he said, you can work like that.

So you can use all these kind of tools and put that in the logic and reason with the ISA. What would be the use case based on my job? And what would be the sequence?

Reverse Prompting Technique for Scoping Projects

And I think the way I would normally do this is through reverse prompting. So have you heard of reverse prompting? No.

You would throw a bunch of information, whatever is going through your mind about how you should do a project. And then you would say, ask me one question at a time. And let's try to get to this output. So basically, the AI would be sorting through that information, and step by step, you would get to something that would be usable for you.

If we have some time later, we can try it on whatever problem we want. It's normally something we do with executives. When we do this, we do these executive sessions and then we ask them, all right, now choose a problem and use this technique.