to be specifically talking about how you can use AI practically in your day-to-day. Either if you want to be more productive in the office or even I'm going to give you some cool tips for you can use it in your personal life too.
And I'm just going to give you a quick background on who I am. Well, I'm Jose. I currently am working in the innovation team with an AI focus in Deloitte.
And yeah, I mean, I've started using AI since basically it came out. So I have more than approximately 1,000 hours.
And I've done several workshops of ChatGPT in IA University. And this is actually my third talk here in MindStone.
So...
What will we see today?
We're going to mainly focus and deep dive into ChatGPT, but specifically into the agent mode that recently came out, the deep research, task automation, custom instructions, and finally some bonus AI tools that I think are useful. So without further ado, let's start.
And we're going to start with agent mode. So Agent Mode was released approximately one month ago or so and right now there's lots of talk about what are agents etc. There's lots of definitions but I'm going to talk about specifically what Agent Mode means in ChatGPT.
So you can find Agent Mode when you press this plus button and you can click here. And for those that don't know, Agent Mode allows basically ChatGPT to create a virtual machine and very basically very simply to
create a web browser such as Google Chrome etc and use that web browser as if it was a human so like get the mouse move it do click here click there and what better example as a live example so I'm I have your pre-made prompt which I'm going to ask the agent if it can order some food okay and just eat And yeah, I mean, I just made up this address to Cassiana 98, and I want to ask for burgers for two people.
And it will start thinking, and as you can see, it will start to execute agent mode. Yes, of course. So, as you can see, this is the interface of Agent Mode, and I'm not touching anything.
ChatGPT is searching, it's going into Just Eat, and right here I don't have to do anything else. I already gave the instruction of where I want the food delivered, how much food I want, and it will do everything by itself. And I asked it to put it into the cart.
Of course, this is something that takes some time, so I'm not going to be here waiting for the whole agent to finish. But in the meantime, I can tell you some things that I can do. So if you want to see it display view, we can just leave it like here, or we can also change it to activity view, where we can see all the decisions and things that ChatGPT is doing.
So it's clicking on the hamburger item, et cetera, et cetera. And we can even pause it if we want to, or we can even take over. So in the case that we like specifically to choose Burger King, we can click on Burger King and then let Chateaubriand do the rest.
So yeah, as you can see, it's just doing things. Look, in this case, it's asking to put the location. It will probably put Castiana 98, exactly. And I'm not doing anything.
So as you can see, this is pretty advanced.
So there are some very good use cases for Agent Mode. It's very good at data analysis. So you can prompt it to go into Google Sheets and analyze data.
It's good also when it comes to presentations.
So I'm going to switch tabs here because this is going to take some time. And we're going to go into...
this presentation, so I basically just gave a generic and long prompt that I want Chad to create a presentation on how to give a practical talk about AI. So here it gave me an initial template after prompting it to put some content within the presentation. As you can see without me doing anything, it has created a sort of simple presentation, but at least this is better than nothing.
And if I wanted to see how this was done, as you can see here, it worked for 30 minutes. I could simply go into the history and I can see exactly what it did.
So here, for example, it went up and looked for icons. If we go a bit further back, we can see it's looking up in internet cool things that I could present. So it has the whole rounded vision.
And finally, of course, it created this presentation that actually you can even see it within ChatGPT. So that's another use case presentation.
It's useful. But I personally like it more because it can do general tasks in the internet.
I, for example, here in this case, prompted ChatGPT to go into my LinkedIn profile. I wanted to go into my last post. My last post, I announced basically that I was going to talk here.
And I told ChatGPT to say thank you to 10%. people that attended this event.
So as we can see here, firstly, of course, ChatGPT had to go into my profile of LinkedIn. So here what it does is, when you have to go into the login page of ChatGPT, it asks you to actually go in and put the details of the login of the username for LinkedIn and the password for LinkedIn. Of course, in this scenario, ChatGPT does not observe your screen, so it's not going to know your password or your login, but it's useful if you want ChatGPT to have access to some credentials.
And here, basically, you can see how powerful it is that it actually went into my profile, It went into the event section. This takes some time. But yeah, it's your reasoning within itself.
Let me just press here play. And it will go into the event section. It will click my last post. All of this is automatic.
It will go again into the event section, view event. If I move a bit forward, this is the event of today. Let's move a bit forward.
And it will select a participant that has signed up at least for the event. I don't know if Javier Corral is here or not, but that's a person. And it will click on the person. Let's continue. And basically, it will click the message button so that it can send a message to this person. So as you can see, it presses the message button. And it opens the message and directly introduces, as you can see, the Madrid AI Meetup.
I stopped it here because, of course, this was a test I did before the event. I didn't want it to perform it. But you can see how powerful it is.
Let's go back to this hamburger. As you can see, it put some burgers in... in the shopping cart and if I wanted to, I could now take over and I could put my login details, I could put my payment details and directly order it. So as you can see here, I can even press pay, etc.
Okay, so this is a bit the power that agent mode has.
Now, as I said, I'm going to go into deep research. This is also very useful, especially if you want to do research online. 1The power of deep research is that it can spend lots of minutes, half an hour, even one hour, just doing 100% research on a topic online.
So, in this case, I'm going to show you a personal case, which I found interesting, which I wanted to do analysis of my name. I mean, I don't know what the origin of my name was, my surnames, etc.
So, I just gave this a bit of a long prompt, because, of course, deep research works basically, you send a prompt, You receive some questions, some clarifying questions, and then after you respond to those questions, it starts. It doesn't ask more questions. So you need to be very specific on the specific research that you want.
And as you can see here, here starts the research, the origins history of my name. And look, as you can see, it's a very, very, very, very, very long research. And it even cites all the sources. So yeah, 21 sources, et cetera.
And it did this in, I think it was 30 minutes or so. So you can really see the power that deep research has in 14 minutes.
Yeah, another very powerful tool that not lots of people know. Also, so that you know, both deep research and agent mode, you can't use it unlimitedly. You do have certain tokens that say, I think it's approximately 40 a month, just so that you know.
Okay, so yeah, that's about the part of research.
Now, another use that not a lot of people know about is the task automation. So let's say that you wake up every day and you look at the news, right? Sometimes the news are too broad and you want specific news every day that you want to read.
So what you can do is as simple as automate this prompt. So ChatGPT prompts every, let's say, Tuesday at 10 a.m. or every day at 10 a.m.
So I can say like, I want the top five AI news. They have to be interesting news. I want you to do this every day at 10 a.m.
Okay, let's see here if the Wi-Fi is quick or not. If not, I have already a pre-made example. But yeah.
Sometimes when you want specific news, you need to let AI run for a while, maybe four minutes, five minutes. And this takes time. And you don't want to every morning wait those five minutes for the news. 1So by using this task automation, you will have those news every day at 10 a.m.
Of course, I'm just using a generic case of news, but you can do it in any shape or form. So, I mean, this is going to take some time, but I already have a pre-made example. Let's see, this is the linguistics.
So here, as you can see, I prompted chat to give me the top five A news of the day, and it gave me the top five A news of the day. And as you can see here, it says that it will do it every day at 10 a.m. Okay, so if you specific, if you prompt specifically chat to do this every day at 10 a.m., it will do it. So here you can see, yeah, it's a prompt every day at 10 a.m., you could do it twice every day, et cetera.
I think this is a very, very powerful tool use case for chat.
And finally, within ChatGPT, I'm going to talk about custom instructions. This is actually one of the features, one of the first features that came into chat, but I still see lots of people that don't use it and it's very powerful because if you think about it, ChatGPT serves hundreds of millions of users and it's a generic software, but every person using it in a slightly different way. So why don't you specify and personalize ChatGPT to, you know, for your needs and wants.
And so in order to access the custom instructions, it's as simple as you press here down left of the button, you press personalization and in this part. So you need to click this button. And the first thing that it allows you to do is to choose the personality. I personally like Chat to be efficient, quick, straight to the point.
Here's where the magic happens in the custom instructions and in the more about you So I have here and I have it here in this in this Google Docs, so I'm gonna I've separated custom instructions that I personally have into into useful and Preference okay, so something that I I like my chat to do is for it to understand what I say. Sometimes I don't give enough context of my prompt and I want Chativity to tell me that, hey, can you please expand on this area for me to give you a better answer. So this is exactly what this custom instruction is for.
Also, this whole document will be for you at the end of the session. I'm going to give you a QR code so you can download it. So don't worry about thinking about it or writing it down.
Another Another good prompt that I like to give my customer instructions is that before any task, if there's anything useful that I can give you so that you can give me a better response, please prompt me. Imagine I am asking Chad to do a SWOT analysis of my company. Well, maybe it will ask me what my customer persona is, what my objectives are, so it can give me a more specific answer rather than just a more generic answer of the SWOT analysis.
also this you know sometimes you ask information about a document and it kind of makes things up this prompt is specifically made so that it is 100 sure that if i ask something about a document it will give me information about the document and not make something up so i feel like this prompt is very useful and also um i think we already know the whole prompt engineering world and where they ask you that you need to tell the chat to act as an expert. And instead of writing it in a prompt, I just put it in the custom instructions so that this triggers every time that you ask chat to be a message. So in every response, automatically assume the role of the most relevant expert according to the topic of question.
Now, these two other prompts that I have in my custom instructions, they're more of a preference. They both basically explain the same thing that I don't want any fluff in the response. I just want straight answers, no extra text. And this is basically some interesting ways that I found that you can say it.
Okay, so yeah, I mean, this is a bit within ChatGPT, how everything works. Now, I have a couple minutes left.
I'm going to very briefly go into two AIs that I feel like are very, very powerful. One I'm pretty sure you've heard about is Nanobanana, okay? So it is very simple. If you go into Google, you search Nano Banana, you can get right on with it.
I took, for example, this random Coca-Cola ad of some bus station, and I gave basically a prompt, and I said that I wanted to be advertising today's event of MindStone. And as you can see here, in almost one minute, it created almost a perfect replica of the ad directly in MindStone, and it's almost indistinguishable.
right very useful also something cool I've probably everyone have seen it online like the picture of you with yourself when you're young you know if you put a picture when you're young and you when you're now you can create kind of a cool you know you younger was you now I don't know I feel like it's kind of cool and finally I'm gonna end up with lovable okay this is something that my colleague is maybe gonna talk about after
This is a very, very, very powerful tool that can help you do websites and softwares all with AI. Very, very useful.
And finally, yeah, I mean, that's everything for today.
Thank you a lot for your attention. So again, thank you for your time.
And if you want to download the document that I've said before or follow me on LinkedIn, you're more than welcome.