Introduction

I was trying to figure out what to talk about today, and initially I thought I was going to give you a bit of an overview of how to use quite a few of the tools that both Nick and Beatrice have actually showed you today. And so last minute, I thought there is probably something better I can talk about.

A Personal Journey and Changing Times

And in this case, I want to talk to you about an experience that I had very recently. because it illustrates everything that's happening at the moment.

And so I titled this presentation, The Shifting Sands Beneath Our Feet, and it's an iteration of a presentation I gave earlier. When I say earlier in AI time, it's a lifetime ago, but really it's only about three weeks ago.

An Innovative App: DreamyTales.ai

And I would like to start, actually most of you can go and have a look

at an app. You can type it in your phone. It works on your phone directly, dreamytales.ai.

Have a look at it. at what it does, but basically it's very simple.

It takes the name of a child, in this case I've got a three and a half year old Lucas, and it takes some personalization elements. And so in this case, so what do I tell about him?

So loves cars, planes, and anything that moves. Recently, we built a new dock alongside his train station. Very simple thing.

Basically, this entire app, you can see here, produces children's bedtime stories. I'm just going to put the audio. So what this does, the audio is there.

It's going to take about a minute or so to generate. But basically what it does is it looks at the name of the child, the personalization elements you give it. It then creates a story.

It splits the story up into scenes. It creates imagery with the story, then uses text to speech to make it come to life, and it builds a full bedtime story for a child. Everything here has been AI produced.

So literally, not a single line of code has actually been written. Even the video itself is a video from Sora that is just on repeat, going over and over, as you can see.

You might think, why is Josh showing us this random child bedroom or bedtime story app? There is an actual reason.

A Weekend of Innovation

One, so you can actually pay to go and get the rest of the story. But that is not what this is all about.

About one week and now four days ago, I took about 40 entrepreneurs to Scotland.

So they were all founders of their respective companies. And we built 10 companies in three days.

By the end of the weekend, 50% of those companies were generating revenue from people that were not at the weekend, including this one. So this one is one that I built.

We made about 40 pounds, I think, by the end of the weekend. There was even a repeat buyer. Someone had bought two stories because they tried it once, and then the next evening, they wanted another bedtime story somehow that came through.

Rapid Prototyping and Revenue Generation

And the reason I talk about this is because of how fast things are moving as we are talking about all of this. So 1in 60 hours, we went from ideation to development to launching to finding the first paying customers.

The entire flow was live. You could actually check out and have the final product

And actually, the first version of this, I think, took about 24 hours. I think there are three people in the room here tonight that were actually on that weekend as well. So I think it was 24 hours that the first version was live with Playmobil.

I'm no longer sure if it was 24 or 30-ish hours. Anyways, pretty fast for the average startup when you think about going from ideation to live.

And this is not a fluke.

Some of you three months ago, I think, two months ago, no, three months ago, here were We'll have heard about this story as well, but I went through a similar loop recently where I wanted to build or I wanted to create an experience for all of you here to basically get connected and figure out who you should really talk to.

Challenging Established Solutions

I dropped the ball because I didn't use it tonight, but the whole idea was I wanted to build an experience where each of you knew which were the three people from tonight that you should really talk to. I ended up looking around different platforms that would offer that, found one that really worked, and I still remember it was a Thursday, I think it was towards the end of the day, got really excited, jumped on a call with the founder of the app, and decided we were going to go and use their platform.

It was doing about 85% of what I wanted, the idea being they could ingest all of the emails of people who were here, would ask you questions about who you really would want to meet, and then match you up with whoever you might want to have a conversation with. Did about 85% of what I wanted.

But they said, OK, if you sign the contract, we're happy to get it there 100% of the way. And that was four days before we were going to have our London event. So I said, OK, we're going to go and do it.

$15,000 for a year-long contract. We do these events all over the world. I was kind of

high-level math, it was okay, well, it's a little bit less than a dollar per user per year, sure, we can figure that out, I'll figure out some kind of sponsorship deal that will cover all of that. But that same evening, I sat down with my, it was date night for us, so we went to a restaurant, and I talked to her, because I got really excited about the fact that this was something we could now do. And about halfway through me talking to her about how this was going to change how we could build a community,

I stopped and suddenly in my head I went, I think I can build this. And then the Friday morning I opened Replit and two hours later I had the platform built out. And actually before that same Tuesday, which is now I think two meetups ago, we actually did use it and a bunch of people got matched up.

And that was a $15,000 contract I was about to sign that was built in two hours and was actually doing 100% of what I wanted instead of 85% of what I wanted. All of this is happening right now.

Groundbreaking Advances in AI

And if you think it's slowing down, there are a few more things that I literally just now kind of pulled from my Twitter feed. But recently, this is one of the things that is now becoming possible with AI.

I've been waiting for this for 40 years. Dolphin Gemma is the first LLM trained to try to understand dolphin language.

I'm a research scientist at Google DeepMind. Dolphin Gemma has Denise's data and sort of encapsulates a lot of the knowledge and experience she has in it. But it's also small enough we can train it with more data as we get it.

Denise has the world's largest collection of dolphin vocalizations. Dolphin Gemma will input sounds. Once a dolphin starts doing a vocalization like a whistle, it can try to complete the end of it.

When you're doing a Google search, right, it's finishing your sentence, right? So feeding dolphin sounds into an AI model like Dolphin Gemma will give us a really good look at if there are patterns, subtleties that humans can't pick out.

We can actually keep on fine-tuning the model as we go and hopefully get better and better understanding of what the dolphins are producing. So they're smart, but language is still the last barrier.

Dolphins can recognize themselves in the mirrors. They use tools. If dolphins have language, then they probably also have culture.

We do not know if animals have words. What do they talk about?

You're going to understand what priorities they have. The goal would be to someday speak dolphin. And we're really trying to crack the code.

AI and Communication with Dolphins

Who would have thought that GPTs, so these large language models, end up actually being a similar breakthrough that helps us actually talk to animals.

This is not like made up stuff. This is happening right in front of us at the moment.

AI-Generated Video Content

And then another one that I have here, which was also this week. I don't know if anyone saw, but the ability to now produce, this is entirely, 100% AI produced. There's no sound yet with it.

So the big innovation here was that we can now do prompt to video with respect of a timeline in a story. So this is a minute and some of a story that actually is consistent on a timeline going through. Now, I know that this is a cartoon, and so it's easier to do than a realistic movie.

But any kind of cutting edge cartoon about six months ago would not have come near to anything like this. Some inconsistencies.

Lutra AI: A Tool for Integration and Automation

And then earlier today, has anyone used Lutra AI yet? Lutra AI. That wasn't on the bingo card.

So I think I can go back up to 16 now. Lutra AI is now finally able to do something that I tried to build about nine months ago, which is the idea that you have an AI assistant, and the AI assistant can use a bunch of tools to actually execute on what you're looking at.

So here, I'm going to say, can you write me a tweet? Actually. Can you send out a tweet talking about how the MindStone AI meetup in London really delivered based on my on the information you can find about the meetup.

When is it? We are the 15th of April, 2025, in my Gmail.

So what this tool is able to do, if you've heard of Operator, you might have heard of Manus, which do similar things. But basically, it is able to take a task. It hooks up to various tools. So in this case, I've actually hooked it up to my email.

I've hooked it up to my Twitter account. All of this I didn't do until about four hours ago, because I didn't really know about this tool up until that point.

But you can see it is looking at different emails. It's found five different emails. I don't know which emails it's going to find about this particular event. I hope this experiment goes well, because it's actually looking at my email right now.

It is then reasoning. So I need to extract useful information from these emails about the meetup. OK, let's figure out what it does.

Oh, so it looks like it has picked up maybe the speakers, and it is going to talk. Apologies in advance, Nick and Beatrice, in case it is messing up any of the tweet that is about to go out. Okay, so it is extracting the Meetup details.

I'm discovering this with all of you at the same time, by the way, because this is hadn't done before. And now it might actually get stuck in a loop because it's doing the same thing over and over again. OK.

So somehow it's found two different locations, and it's found different speakers that are not here. It's found me, but Anita Pereira is not speaking tonight.

Now it's actually tweeting, tweeting what it found. So I need to go to my Twitter account now and maybe delete this tweet.

Let's see where we get to. Actually, sorry. Wait, where do we go? Ah, there. What is the tweet?

Mindstone AI Meetup in London today really delivered amazing talks by Anita Pereira. So I don't know if you are in the audience, but I don't think you have been speaking. Well, there we go, yes.

So did we talk about you speaking tonight? No, so I don't know how that happened. So somehow it's picked up one of my emails that we must have spoken about tonight somehow in a different way on cutting edge practical applications, impact of AI on work, life, society, fantastic insights in the future, great networking, and so on.

The tweet here has been cut because it was doing more than 180 characters, which I think it has actually talked about. So here, I think it talks about it here somewhere. Well, you can actually see the tweet itself.

All of that to say that it has gone through multiple steps here. It looked at what I was asking for, went through my emails, figured out what the emails were about, composed a tweet, and then actually tweeted it out.

And so I am going to wait. People are already retweeting it. Wait.

I might delete this tweet just to be sure this was an experiment. All of that to say just how fast this is moving. We are now starting to see these tools that are able to leverage all the other tools that we are already interacting with on a daily basis.

Reframing Perspectives on AI

And this brings me to one of the more important points that I think we should all focus on, which is that We do a lot of this work.

So we do AI training for non-technical people in some of the biggest organizations in the world. And I am starting, I love all this stuff. I mean, anyone that knows me here, literally the entire purpose of these events is because I love it so much.

Because you can do so many things. But I am starting to be a little bit disappointed how much everyone is leading into automation and very few people are actually leading into augmentation. 1These models can allow every person in this room to do so much more than what we are capable of today.

Both Beatrice and Nick gave amazing examples of how you can use the technology to do more. But somehow, because for years and years and years we have talked about AI as automation, it is the only thing that people think about when we talk about AI.

And so AI is now becoming a bigger thing. But every company we talk to somehow instantly thinks, OK, so how can we go and make half of the company obsolete? Or how can we go and automate every single task that's there? And I think that is a real opportunity. But that is a real opportunity so long as you're also leaning in on the other side, which is how do you make sure that the people that are actually there are able to leverage the tools in a way that they can do three times more than they were able to do before.

And this technology actually delivers more on the augmentation side than it delivers on the automation side. And I think people are really forgetting that.

From Deterministic to Non-Deterministic AI

I was talking earlier today about how Previous AI was really about a deterministic environment. Basically, you have a set of rules. Whatever you put in, you go through those rules and you get the same thing out.

What this technology is great at is everything that's non-deterministic. So all the places where we used to be able to leverage technology, this technology is not great at. but all the places where we never thought about applying technology, where ambiguity was the main thing that made it difficult to employ the technology, now this technology is really great at, because you don't get the same thing. You get much more nuance out of it.

And so that means that the type of use cases you can look at, the type of use cases we all can leverage, are very different. And I have the impression that because we have 20 years of thinking about AI, and this is now being called AI in the same vein, it's almost as if automatically people default to the types of use cases they've always talked about.

From Single Player to Multiplayer Productivity

Now, one of the biggest things that I think each of you can really start to think about how you start to incorporate it, the biggest shift here is that we're basically going from what I would call single player productivity to multiplayer productivity. It is the idea that up until now, it cost you something to get the opinion of somebody else on whatever piece of work you were doing.

The CMO that Nick was talking about earlier today really would have... Well, I mean, you either have to have a really good reason to go and talk to, or you pay a lot of money to have someone with that type of expertise to give you feedback on whatever problem you might have. That cost has just fallen through the floor. It's zero at this point.

At least, I'm not saying that you get the exact same quality, but you get... Whatever quality benchmark you put on this, and you get it at a cost of zero, which means that your entire approach to how you think about work in the first place, the cost of getting somebody else's opinion is now, or something else's opinion, is now zero.

There is no more reason for you to do anything that doesn't have a second pair of eyes on it. And that changes fundamentally how you start almost anything you work on.

Because previously you'd start it yourself, get it to a certain point, and only once it became important enough would you get a second, third, fourth, however many people's opinions. And the more important the task, the more opinions you would get. Now that first opinion should be on literally the first second that you have anything you're producing.

AI as a Personal Trainer for Work

And the thought framework that I would like to put forward is to think about AI as in the same framework as a personal trainer, which is that at any point you are working, you now have a 24-7 personal trainer that helps you with whatever task you are putting forward. Elite sports people have their personal trainer with them all the time, all the time they are performing and the time that they are training. And the same thing is now becoming true from a cognitive perspective on when we do work.

And that means you think differently about that feedback loop, and AI can be extremely useful around that. And when you use that effectively, it means that feedback on your work is now instant.

Conclusion and Call to Action

This is the end.

This is the first time I think I've done a little bit of a cry for lean in, please. And there's a bit of homework for all of you.

I would invite you to go and read this this piece of writing, AI2027.com. If you haven't read it yet, it is probably the most significant piece of writing that I have read in the last few years.

It is eye-opening, it is a little bit scary, but it is also the first time that I've actually seen written down in a coherent way some of the frameworks that I believe that we are living through right now.

And it's interesting because it's on a timeline of 2027. We are talking two years from now.

We're not trying, nobody's trying to figure out what 10 years from now looks like. It's just what the next two years might look like.

Hopefully this was helpful.