Introduction

Right, so I'm going to try and make sure that you can interrupt me at any given time for questions about what you're trying to learn here tonight, because it's all about knowledge sharing. I'll do a short introduction of myself.

Early Fascination with Computers and the Internet

So as a young kid, my dad, who was in the Australian Navy, that's where I was born, he gave us a computer. We just moved to the Netherlands, and I started programming as a five -year -old.

and thankfully no one laughed at me for being a nerd because people just didn't know what a computer was at the time so that helped and then when I went to study and I went to a business school in the Netherlands this new thing had been pretty much introduced into the computer room

called the internet and it blew me away I was like I knew that was my future this is going to change the world this is going to change how we relate with each other which has been perfectly true

because when I argue with my children it's always about screen time or other things that have to do with online subscriptions.

From Networking to Early AI

So the internet was where I started working and for 15 years I built internet technology, faster networks, better applications, stuff we now call cloud and even stuff we now call AI when it wasn't named AI yet.

This is self -healing networks which is built on the same type of technology where you now see prediction and machine learning that we have

that predicts the storm that we've got on tomorrow so if you didn't check your app yet there's a storm coming tomorrow and AI has made our weather prediction so much better but we now continue on with natural language and it's just fantastic where it's going and a little bit dangerous or

Teaching Productivity and Building Tools

maybe even scary but that'll be you more addressing it so what did I do in the meantime I do a lot of training teaching people how to manage their email personal workflow productivity but always applicable to the computer so I teach a lot of people how to do inbox zero the more professional way of

managing your email still today a very popular training I made an app at some point that's about seven years ago called brain toss it's probably the laziest app on the planet it's because I'm lazy

it only does one thing email yourself but that's a method of capturing my intentions and making sure I drop it into my inbox and I also have an inbox

Lessons from Coaching and Learning

for the Amsterdam Cricket Academy which we started just when COVID hit so this is us trying to learn help children learn better how to play a game like cricket and I found that there's actually a lot of similarities and how we try to learn things about AI and children learn things about cricket

it because it's just our brain learning new stuff and repetition is very important in both so just

Dotting AI and Today’s Demos

get going and do it so now a founder of dotting AI which is mainly about how do we as humans learn and how can AI assist in that so we are a partner in discussing that with companies and organizations but also building tools around it and I do a lot of lectures around the Netherlands so

So, wrong one. And this is actually the wrong one. Apologies. I just changed it around.

Exploring Atlas

Anyone tried out Atlas yet? Atlas? I'll explain that.

Yesterday, indeed. Yeah.

And this is what it looks like. And just because I'm playing around with it, which I can highly recommend.

So, here I've got to look for a mind stone. There we go.

So that was the same slide.

Understanding Modern AI: Probabilistic vs Deterministic

So a couple of words that I'd like to think about before I do the demo. And this will apply to both sessions that I will talk about, both the vibe coding and just the demonstration of different tools.

The new type of software that we're all getting used to now, Copilot, ChatGPT, all the language models, all the image generators, have something in common which we call probabilistic.

this means is that they're actually not determining what's exactly true they are guessing what we as humans want to hear or see and that's a very important difference in what we are used to to get as a result from a computer what is

deterministic it means that when you expect from a calculator or a computer and you say 1 plus 1 you know the answer is always 2 that's not the case with

chat GPT and with other large language models or machines that do generative AI, what they do is they pretty much guess based on a lot of data what you probably want to see.

A Watch at 10 Past 10: Training Data Bias

There's one beautiful example is that when you ask an image generator on the internet to create an image of a watch at 3 o 'clock, you only get a watch 10 past 10. Does anyone, can anyone guess why that's the case?

So it's been trained on images and that's the image it's seen the most because that's the prettiest watch face and all webshops on the internet have images of so watches on 10 past 10 So it's not it doesn't understand that it's a watch it doesn't understand that what 10 past 10 is it just knows based on the data

This is what people have always always been looking at at the internet And that's why I can also guess when you say my pet is a you're probably gonna say dog or a cat that's based on all the content on the internet, and that's what we call probabilistic.

It is guessing what you want to hear next, and there's actually a little bit of error margin there deliberately built in to make it more natural.

And this is very important when you start to use AI tools, that you know it's not going to be right all of the time, it's actually going to be wrong quite often.

It will just try to please you to get a result, it will try to do that fairly quickly because it wants to save energy really but it's been programmed to serve you and that means it's something very different than a computer program which is going to tell you

no you're wrong one plus one is actually two. So that's an important difference.

Prompting Well: How to Work with AI

Now when I have when I've been working on my own workflow and when I train other people on how to use AI we talk about prompting quite a lot.

Prompting is really the currency of using AI. It is a new way of dealing with how do I get results from artificial intelligence.

And prompting is nothing else than instructing in natural language. So you use your normal language like we speak with our grandmother and with our children and we tell the computer this is what I want as a result and the word prompting is associated with that.

Just like in Google Google, the phrase was a query, you query a search engine, and now we prompt a generative AI model.

I've done a lot of research on how we can get better at prompting. And there's two things.

Practice Builds Intuition

One, do it, just do it a lot. The more you prompt, the more intuition you will build on what works well in your applicable program. And I don't really don't care whether that's chat GPT, Gemini, Copilot.

that's the same discussion as, ah, do you drive a Peugeot? Well, I have a Citroen and I think a Citroen drives better. It's not about the car, it's about the driver.

So you make your mileage and you will get better and you build something which is quite important. You build something

which is an intuition of how AI can hallucinate, can simply lie to you when it just wants to please you it can give you an answer you weren't waiting for it can it can go quirky sometimes and that's something you need to just get experience in so it's about the mileage a second thing is about

Brief Like a Pro: Clarity, Context, and Role

the prompting that I don't think we need to worry about it too much it's changing so quickly that all the tricks there's like 15 different acronyms for prompting in a certain way and I'm and I'm I always try to make it simple and we already know how to prompt because when we were in school it's exactly what the teacher did to us.

When you tell a 12 year old to do an assignment you are prompting it. You're telling it what is the objective, what is the background, this is what you need to do, you need to do it in this format and we expect it by tomorrow. It's called homework.

Someone we prompt really well is an expensive freelancer. I don't know if you've been in the position of instructing a freelancer. That's a great way of thinking about, all right,

I'm not just going to ask for a consultant. I'm going to ask for a UX consultant and I'm going to talk your language and I'll give you a very explicit briefing of what you need.

And as soon as you become more explicit the AI tool does not need to give you a generic answer it can give you an explicit answer so those are the trips the trips

Context Over Tricks

that I use and that's prompting versus context the more context you give the AI the better it can decide from this still where you want to go and give you a better quality answer and that's in my workflow so what I did recently any

Any questions on this? Because I'm going to get back to what I know and what I don't know in the second part, in the vibe coding.

All good? Okay.

A Practical Workflow with Multiple AI Tools

So what I did recently, and this is, I'm going to take you on a little journey on how I use different tools.

And I think I may have skipped one slide that I do believe is important. I want to show you this slide here.

is try to do a job application. I'm actually looking myself for a part -time job, only two days a week, of a vision that I had recently.

I don't know whether the company exists where I can do the job. But what I do know is that AI can help me.

If I have a lot of question marks in my head, AI can help me get a good answer and steer me in the right direction.

When to Use AI—and When Not To

so when would I recommend you use AI? A when you need to brainstorm very good when you need to

practice or when you're stuck so if you're stuck in a certain way that you need to go and create something AI can help you out it can help you out get unstuck but you will probably need multiple

tools and the important caveat here is when you shouldn't use it and that is when you need the truth. And that's a very important and a very good addition.

And the truth is something that you do not also get from your assistant or your freelancer.

The truth is a very interesting discussion to have.

Truth, Fact‑Checking, and Second Opinions

So what I've learned to respect is that I actually get a result from AI and we internally we use happiness factor. How happy are you with the result?

And if you need the truth you need to do your own fact -checking just like you do with a 12 year old student or with a colleague who hasn't doesn't have the same domain knowledge as you so do your fact -checking you can actually use AI

quite well to help you do fact -checking I often get a second opinion from Claude when chat GPT pretends to do something or the other way around and that helps me discover stuff that I didn't even know it was lying about.

So use, you do some fact -checking with Google, do some fact -checking with resources you do trust, and know Booklem is a great thorough way of getting through it.

But

Human-in-the-Loop: The Tiki‑Taka Method

what you've just explained, the way you do that, and that's a style of working with AI tools that I highly recommend you, it feels clumsy, but know when you embrace it, it actually makes the human in the loop something natural.

The The terminology human in the loop means you do your fact checking at the end or you do a quality check halfway.

And what I personally do is I call it tiki -taka because it's a famous technique. Johan Cruyff introduced it at Barcelona. The ball is moving a lot to and from and to and from and back and forth.

And this is where you are. And that's exactly how I use AI tools. I will use AI tools never to expect an end -to -end result,

just to go hey can you do this for me and then they will run off and at the end they've magically done all the work while I sleep or I have a watch TV no I actually do a lot of copy and paste copy and paste copy and paste I often copy

and paste into something called a markdown file which I highly recommend looking into how you can use markdown files as another currency a way to capture some information that you fact -check and then you reuse that

markdown file which is just a text file you could use a word document or an excel spreadsheet but I use markdown files because tools that use AI love markdown files dot md very simple file and

it has my specification of my project so tiki taka is the style and the project I'm going to

Project Showcase: Conflict Prevention AI

demonstrate for you just quickly is this job search and I'm still in the job search so it's not really a job it's a project or a long -term project but it's something that I'm highly interested in because it just keeps me up keeps me awake at night and I call it the conflict prevention AI so just two

two minutes on the background. How many times? Five minutes. Yeah, that's perfect.

So I've done a lot of work in Claude to do this in co -pilot, et cetera.

It doesn't matter as long as you find a chatbot that you're comfortable with that speaks your language.

And I've noticed that's personal.

Some people prefer Claude over ChatGPT. Not an issue. It's personal.

All right. So what am I trying to do here?

Reading Weak Signals to Prevent Conflict

I'm trying to work out, I'm brainstorming with ChatGPT. I believe that AI has the ability in the future if we start working on it well to read from all the words that are printed on social media and other places on the internet

to read weak signals that are signaling hey if this keeps going on sometime in the future we will have an aggressive conflict here right these are because we know because we can actually learn from the conflicts that we've had in the past and what was written before those conflicts that I mean large machine

learning should be able to to start to predict listen here in this region the way the people are talking to each other right here we might have a conflict in the future so if we could build that we can let AI a give a signal hello there's

trouble about to start but also to give advice based on everything that we know about conflict mediation, you should consider ABCD as a plan to resolve this conflict. Now, there's a lot of organizations on the planet that work on conflict mediation or

resolution or data gathering about conflicts, but I haven't found one yet that's really diving into how can AI do this. So if anyone here knows an organization like that, talk to me afterwards. I'm very curious.

Prompting as Extended Thinking

I've done quite a bit of research on it. But my initial research with the ChatGPT, and what you will see, this is how prompting really works in my opinion. It's not a quick resolution.

It takes hours and hours of to and fro, and then I pause, and I go and have lunch, and I come back, and I can have more hours of prompting, refining it, refining it, refining it. And in the end, when I asked it for a vision paper, it was no more than one or two pages

with a lot of going to and fro. But what was happening all of that time, AI was helping me think. Normally, this would cost me weeks if I had to do it with real people. Find the right expert, talk to them, and they would inspire my brain for the next step.

After I did that, I was thinking about, right, I need to put this story somewhere so that I can share it to go and find me the job.

From Vision Paper to Landing Page with Gamma

And I chose Gamma. Why did I choose Gamma?

Gamma app is probably one of the easiest slide creators, PowerPoint substitutes, that uses AI. So what you just do is you give it a quick prompt and it will show you a slide pack or a landing page. They also do landing pages.

So I turn it into a landing page. And now that I have a landing page and it generated all the images, I didn't have to do much. I just gave it my vision paper and now suddenly ten minutes later. I have a website about that vision and

Then I thought well hold on. Is this the right one? Yeah

Gathering Feedback with Formless

I'm gonna do a new formless. I need to tell my My network that I'm trying to look for this So I'm just I'm going to create a form to share with my network

just like you would do with Google Forms or Typeform. This is run by the same company, Typeform. It's a Spanish company. It's called Formless, and it's an AI form creator,

and it lets people talk to the form rather than type the form and enter fields. So I'm going to come up with a prompt. I'm just going to say,

please give me a form to get feedback from my network and other stakeholders that will allow them to comment who in my network I have to go and contact to get closer to landing a job for the vision paper on conflict resolution AI or conflict prevention AI I should say

this is that we would like to help do that but we haven't found the right organization yet. So I'm just asking for a couple of simple questions how they can inspire me to find the right people.

I'm just speaking out loud as much context as I can about what I would like the form to do. I'll just enter our company name because it always asks that and I'm going to hit generate.

So now I mean this is this is one of the most simple AI tools that I could imagine because it's done.

I can now share the link to this forum to people in my network and ask them to share their wisdom in a conversational way. It has changed the tool into a conversation that I can even speak. I'll do that quickly if it's working.

Don't change your browser one night before you do a demo. Now, all right, I'm actually going to stop it right there because I haven't done one thing I haven't given this form the context yet

so if I were to upload the URL here to my paper and just the document of my vision then this tool would understand the vision we're talking about and it will then generate way better questions as soon as people hit send of this I get

five or six different emails for my network helping me go and find the right job and I could just add this to the to the website so formless highly recommended as a specialty tool where you can learn to see how a different

user interface than chat GPT can turn something that you speak into a nice little product questions on formulas it's I do a lot of yeah that's a problem still today but let's see what tomorrow brings that's one so people have

probably already forgotten what it was like a year ago standalone two years ago it has improved so much these problems will gradually become less and less and gradually is going a lot quicker than technology before so I'm what so the

Managing Conversational Drift and Resets

question was what about and this is in the context of your discussion with it with the thread, you will start to make, if you make a small error at the beginning of your conversation, that will just keep continuing on.

No different than me discussing it with a human being.

My colleague has misheard something in the first time and we get the number wrong, so always check the end results.

But what works really well for me, if you notice that it's getting quirky, I go a little bit back and you can either branch in Chatupati, for instance, instance, or actually I go back and I say, can you summarize it for me?

I grab the summary and I start a new chat with the summary. And that feels bad as a prompter because you just invested energy, but it actually works quite well.

It's just let's have a five minute break and we'll start again together. You do that by just grabbing the result, starting a new chat and then redoing it and you freshen up the context.

On Trust, Truth, and Changing Search Interfaces

So in short, always be aware that it can be wrong. It will be wrong. it's been warned everywhere but you need to double check that.

I think it's a great bridge for Julia to take over because that's more about this question about and why do we trust ChatGPT more than we would trust a human.

The daughter of, actually I'll look up a small video. This is what happens when you search with Atlas. So now you're asking a mix of what's a search engine and what's a chat GPT tool to give you an answer and it actually comes back with a lot less information than we're used to with Google but it does it

right what I would look I'm not going to show the video it actually showed the video the first time I did that so lesson one you do the exactly the same search in a browser which is run by a large language model exactly the same question you get a different result the second time that's that's good to understand and to remember.

But if you want to find a nice story about the truth and how a new generation doesn't really care that much about the truth, the little interview with Keanu Reeves explaining that he was explaining the Matrix to a younger friend of his daughter and then she's like, so what?

What do we care about the Matrix? Don't you care about the truth? No. And he thought it was awesome. A shift in how we look at the truth.

Not my point.

Vibe Coding: Programming with AI Assistance

I'm going to talk vibe coding very quickly because otherwise the pizza will be getting cold.

I'm not going to demonstrate it, the demonstration really takes too long, but I want to mention that I've got this app that we built using front -end developers and a back -end developer a couple of years ago.

It needs redoing because we're having trouble in the back end and I thought why don't I give it a shot to vibe code it over summer.

I was in my camper van in the UK going from cricket event to cricket event, plenty of time in between to vibe code. code.

Anyone not aware what vibe coding means? Has anyone not heard? Alright, so just quickly,

What Vibe Coding Is—and Isn’t

vibe coding is a trendy terminology, a new way of computer programming if you're not able to computer program.

And the only successful vibe coders are people who are able to program to code.

So this is writing code by telling AI to write it for you.

And this will be big because good coders have become so much better at getting software reviewed and producing new software and adding value because they can now integrate natural language into their workflow

but I would like to warn to you if your expectations is oh I can write software just like that well be ready to first of all run into problems I started vibe coding with this machine

Local Tooling Pitfalls and Setup

and I tried the different CLI -type -based coding, Cursor, Windsor, WobDev. I've pretty much tried them all.

They will start installing software on your Mac, in my case. Soon I ran out of space. How did I find out that I was running out of space?

My team's screen share wasn't working during a very important presentation to the government of the Netherlands. 70 people online.

And just because I started coding again, I had trouble in something on my production machine.

You'll run out of space by now everything that I've coded is on an SSD card and tomorrow I'm going to order a new Mac because vibe coding you do need a professional machine.

Vibe Coding Lite: Prototype in the Cloud

Unless you go vibe coding lite and that's what I recommend for all of you to learn.

Vibe coding lite means you get a cloud service something like lovable or even just in Claude you do a Claude artifact much worth trying and you explain what you want the software to do and you will get a result that's presentable that will help you and just from scratch you

can buy the mobile app or a web app and my favorites are lovable or clawed artifacts and it's great not so much to build software for the market but what

it's really good for is to build prototypes most like like minimal viable products and something that you can use with your colleagues or with whoever you're doing this with to discuss what it should look like.

Prototype First, Brief Better

If I were to build software I would spend a whole night evening first prototyping with vibe coding something in either lovable or in Claude and what the end result is

probably not working but that is I can brief a software developer in half an an hour using that where if I were to write the document it would take me two days and the software developer writes reads the document and he's like all right I've only know half the context than what I would have achieved if I were to do a prototype in in vibe coding style so that's just

to manage your expectations a little bit what's going to what you're going to run into into when

Q&A and Wrap‑Up

when we when we do vibe coding anyone else here has experience with vibe coding yep doing it as we speak that's the lovely thing you tell it to go and do it and then you can watch the screen

and so what I recommend is that we have a lukewarm pizza that you can talk to us about vibe coding we'll do the Q &A on vibe coding over a slice of pizza maybe yeah thanks