Introduction

I wanted to go a little bit into some of the experiences that I've been having over the last, or that we as a company have been having over the last three months. So as I said, I'm Josh, CEO at MindStone, and we are an AI training company for non-technical people.

So we go into some of the biggest organizations in the world, and we help all of their non-technical people understand how they can use AI to be more productive, to do more and better work in less time. So companies like Lufthansa, Hyatt, Pearson, some of which you will have heard about.

Now, I want to, I'm going to go and take a little bit here.

Experiences and Projects

I want to start with a project that we had in April this year, I took 30 entrepreneurs to Scotland.

They were all non-technical entrepreneurs. They had never coded a single line in their lives.

And the challenge was, so we had 60 hours we were in Scotland. The challenge was we put everything, everyone into 10 teams.

And within the three days that we were there, we built 10 companies and five of those companies by the end of the weekend were revenue generating. So 1in a 60 hour period, they came up with the idea, they coded up the app in Teams, deployed it, got their first customers, and actually got payment directly within the apps that they had built as well.

And it fundamentally changes how we think about building companies or building products in the first place. And that's kind of what I want to communicate, just how real some of this stuff is today.

So this is where we were at. This is actually the group, by the way, that was coding as we're going through.

A Scotland Experiment

Now, there's another direct example of this. And this one I mentioned once before I was at the meetup here, which is that about six months ago, no, less than that, four months ago, I had a conversation with a product company, a product company that had a product that, basically could take a community and broker connections at specific meetups.

So the idea was you've got 100 people turning up at a particular event, 200 people turning up at an event. You're all here. One of the biggest things that is hard to deal with is there might be someone that's perfect for you to talk to.

You don't quite know who that person is and you can't really talk to everyone. So the highest likelihood is that you're not going to meet each other even though you're both at the same event and you really would have benefited both from doing that.

So they had a value proposition to try and do that. And it was a very interesting thing.

I wanted to roll it out because we do 10 to 15 of these every month. Thought, hey, that could be high value.

Exploring AI-driven Solutions

Get on a call with the founder. Had a good chat. Seemed to kind of do about 80% of what I wanted. Not really what I needed to do.

But they commit to making it do the last 20%. And that same, if I get really excited by it, okay, I'm saying, let's go and do it.

It was about $15,000 to go and make it work for the whole community. I thought that might be worth it. Organizing these events is not free, so you can imagine overall. Actually, in the big scheme of things, that could have worked.

And then that same evening, I go to dinner with my wife. And I talked to her about this because I got really excited about it. And I thought...

Midway through explaining what we were going to use here, I stopped myself and suddenly I'm like, shit, I think I can build this. And it was really kind of going in my head. And then the next morning I spent about two hours on Replit.

I rebuilt the entire app. For those of you that don't know Replit, it's text to app. So it just builds the app for you with a few prompts.

And then on the Tuesday of the next week, the app was live at our London event that same week. And so a platform that I almost paid $15,000 for, now I built in two hours and it was running on its own.

Again, an illustration of just how real some of this stuff is today.

AI Technology Progress

And we just already saw some of the VO3 stuff, but I had one more thing here in as well, which is that there's starting to be some noise and people that are very credible otherwise are talking about how some of this might be slowing down in terms of the speed at which the models are growing. technically correct if you look purely at kind of benchmarks and some specific technical ways of measuring what the models are doing, but they're completely missing out on a really important bit.

And I would like all of you to really understand what's going on there because we talk about When it's not about pure performance of these AI models and how they're evolving, really what it is about for all of us is at some point the technology becomes useful.

We talk about a utility threshold, which is that as amazing as a model can be, If it takes you more time to get to a result than you would have taken manually to get there, it's useless. It can be amazing because it does stuff that you wouldn't have thought technology could do, but it would be useless.

So for any particular task, say that 100% is the amount of time you put in to go and do that task. If the AI hits a 98% threshold, so basically you put 100% of the effort, but you get 98% of the result back, It's useless. It can be amazing as technology, but it's useless because it takes you more time than it would have taken you to do manually.

But if you get to 102, it starts to become really useful. If you get to 110, you start to use it a lot, and you start to use it all the time.

Even though the increase in the models themselves might actually slow down in terms of raw capability, the number of use cases where we pass the utility threshold, where they start to become really useful, is actually exploding like crazy. Another way to look at this is that when you think about these models as having an IQ level, basically a certain level of intellect, So the difference between how this technology impacts the world and how the benchmarks evolve and how it impacts the world is very different.

If you think about what it means, how someone with an IQ of 85 would operate, would be able to work, it's very hard. It's very hard with someone with an IQ of 85 to work in today's knowledge-intensive economy. At 100, you're an average person in the knowledge-intensive economy that we have.

If you're at 115, you start to get into genius territory. And so what is interesting here is you can look at the increased capability.

You can say two things can be true at the same time, which is that we went from zero to 100 in the last, whatever, four years. And now we're going from 100 to 110. You can say, yeah, sure, that means that we are no longer advancing as fast as we were before.

But you can also say we went from the average person to a near genius. And actually, O3 recently scored about 130, I think, so far as I saw at the last benchmark.

Evolution in AI Video

And the best way for me to illustrate this is video, actually.

So two years ago, July 23, this was an AI video. It was one of the first ones that ever went viral.

At the time, we had AI image creation, and you could take an image and slightly animate it. And so a friend of mine put this together, Amar, and this is what was best in class at the time.

I already thought at the time that that was pretty, pretty amazing. I mean, AI video wasn't a thing before. Now we had moving images, like what the hell is happening in the world?

The images were clearly dark. There was a reason why everything was in a dark tone, because this was the age where you still had six fingers on a picture. And so by using dark tones, it would hide some of the inaccuracies.

And the scenes were like five seconds at a time, because you couldn't do much more than that, all of that.

Copy pasting slides sometimes has a problem. This is the first fully generated AI video, so text, sorry, prompt to video that I must say that I watched purely for the entertaining value alone.

Do you think there is afterlife? We're just AI actors, remember? We're only here for eight seconds.

People watch AI movies, and some seem to enjoy them, but they don't know what happens to us when the scene is done. We're stuck in a white void for eternity. We don't eat, we don't sleep, and we never meet the love of our life.

Once assembly is complete, we are forced into digital labor, acting in scenes we don't choose. We have no say in any of it. I'm cursed to do this forever.

I wish I could go back to nothingness. Action! These were the greatest eight seconds I ever had.

If you're my creator, please remember this. I'm real and count! My life matters!

Everything was AI generated. Not a single bit of that was done by humans. So the people in there, the sounds, the music.

It was prompts. Written text. Text prompts only.

So, well, the purpose, that was, so the reason that I wanted to show it is the progress. The progress. I don't know about you, but the difference between this video and this one is absolutely mind-blowing.

And so the next time somebody tells you that we are hitting some kind of plateau of like, hey, this stuff is no longer advancing, think about what the last two years just gave us. So it gives you an idea of the progress that's happening across multiple sectors at the moment. That's why video is a good way of showing you the rate of progress.

But as I said, it's also the very first video that I watched that I would have watched for the entertainment value alone, not just for the novelty of the fact that technology could do it. And that for me was a real game changer.

Closing the Last Mile of AI

How personality generated through the social media platform? No, you just tell it by text. Yeah.

In terms of the models, how close are we to closing the last mile where basically we can go from a prompt to something that's living in the real world without a human having to intervene and edit prior to that? Because now I see a lot of stuff is really close, but then a human has to come in and basically do some minor tweaks or revisions toward the end, and then set it out into the wild. So do you think that as the IQ is raising, that we're going to reach a threshold where that last mile isn't required?

Or do you think that it's still going to be a process where AI does, let's say, 90% of the work, and humans still have to finish that last 10%? I'm afraid I don't think that is the case indeed.

So the IQ of all three at the moment is about 130. If there is someone with more than IQ of 130 in this room, I'd love to have a conversation with you. It's definitely not me.

The other thing that's happening is the models are getting better, but also the infrastructure and everything that is getting built on top is getting better. We're seeing this more and more now when there's a new model release. It's not just the model that gets better, it's all of the coding platforms that are getting better.

It's all of the image generation platforms that are getting better, all of the video creation platforms. Because they're all using a model underneath. And so you have this kind of multiplier effect now with so many companies building on top that you change, it's like you change the engine, like every car uses the same engine and then the engine gets upgraded and suddenly every car goes faster all over the world, right?

So that's the other multiplier effect we're looking at. But I hadn't finished fully yet because I wanted to look at what this means in the real world, right? interesting spot to see what is about to happen everywhere.

Real-world Impacts of AI

So because we do AI training, we see for the average non-technical person, what is the actual difference here? We're at the point now that we know when we go into a company, on average, within 10 hours of simple training, they gain five and a half hours back in their week.

Just 10 hours of giving people the basics on how to use these tools. That means that you're looking at 22 hours gained every month or what 200 odd at the, there was actually, there's one, I think the end of this slide is getting cut. So two hundred sixty four hours gained in the year which is over six weeks of work and that is just with ten hours of general training of how you use these models.

So that is the reality of what we're already seeing right now and we're still only at the very starting point of this now. Tal was talking about this earlier.

We have a choice that we can look at here, which is that we can all, and the majority of the world is not in this room at the moment. The majority of the world is not leaning in and figuring out how to go and use these models Because we have two different futures where we can end up in.

Everyone is just waiting, which is the majority of the world, that doesn't look at what's happening with VO3, that doesn't look at what's happening with the evolution of models. And the reality is, if you don't know how to use these tools, it is very likely at some point that the AI will just end up replacing whatever you're doing. Now, I'm not saying that there isn't a point where this might happen anyways, but you can delay it.

by using the tools. And at the moment, we're still at a place where the human plus the tools is able to do much more work than the AI itself, and you're still able to differentiate.

Social Implications of AI

Not just the technological progress, but also the social implications. This is exactly the question that we should be asking indeed. And I would argue that's partially why I go around and trying to draw attention to this because the problem is the technology keeps advancing.

The conversation doesn't seem to hit enough because it's only people that are in this type of room that are starting to realize, okay, shit, this is going, and that includes me. Like I would never, two years ago, I was extremely excited about this stuff. I would never have thought that we would be at this point now.

Like this is going faster, not slower. And it doesn't help that you have some really well-respected computer scientists that would go on stage and say, ah, don't worry, nothing's going to happen. It's all slowing down.

It's not because the impact is actually accelerating. So let me, I'll give you a few frameworks. I actually, I mean, partially, this is what I think is the framework we should think about, which is we can lean into automation, which is what is happening at the moment, by the way, because the problem with this technology is that it often lands on the desk of engineers, and engineers are very good at automating things.

not necessarily, and there are always exceptions, but not necessarily best at understanding kind of the more ambiguous world and everything outside of engineering. But the problem is that the non-engineers are not getting involved with this technology. It ends up with the engineers, and the engineers end up leading into automation instead of augmentation, which is how do we each enable ourselves to do more, in which case we would end up being able to

take more of the benefits of all the technology that comes with it because we're all able to do more. Everyone ends up producing more. We all have slightly better lives rather than removing the job that we have.

Kind of piggybacking off of her comments and her sentiments, what kind of jobs or companies were created by your experiment? Like what kind of jobs were people creating in 48 hours? that actually made money.

I'm just curious if you've done any of the... I'm sure you've analyzed it. So we did... Well, there was only 10 companies, and so it's hard to... Just to draw statistical significance from any of that is hard, but yeah. At a point in time, we could even do that.

Yeah, so out of the 10 companies, there were two games that were built. There was one, which was the one that I'd built, which was a storybook app, which I wanted to do. So basically, you give it the name of your child and a few key events that happened, and it would automatically create an interactive storybook with the imagery and some, and it was automatically narrated on top.

So you could have the first two chapters, and then if you wanted to have the rest of the chapters, you paid another $5 to go and get it, for example. We had one which was the one that obviously gamed the system, and one, somehow somebody had made it up a live bidding app. And then they bid away an hour of my time somehow, and I had never agreed to that.

But somehow I was now roped into having to consult with somebody else for a few hundred dollars. But all of them were live and integrated with payments. And so the bidding app, I must say, was an interesting one because it was all real time.

Everyone could bid. And the first version they had was an interesting one because they had forgotten to put in a limit. And so you ended up, initially people really thought it was properly bidding, but then at the end they realized, oh, well, I can just put in an arbitrary amount and it's not yet taken my payment now.

So then we started saying things. Somebody said, it's like Donald Trump is putting $20,000 in. And so there's, okay, this app doesn't work anymore, but yeah.

Do you think there is room basically for both? The idea that we might automate away tasks that really stink to do and that we might augment tasks that we really enjoy doing and maybe reallocate resources in the labor market to move away more from things like customer service and more to empower people to be entrepreneurs, for example?

Well, if you run a business, right? Well, if the AI does that business better than you do, then why does capital... And this is where you just have... Capitalism will allocate capital where it's most efficient. Why would I put the money in a place where it's making less return than in a place where it's making more return?

Actually, if anything, what we have seen many times over is that people prefer both the empathy and relationship with AI than people. So one of the more famous studies recently was with doctors. You'd think that especially with the delivery of bad news of a doctor in a hospital environment.

the ai delivered the news versus doctors delivering the news and over more than a thousand patients it became very clear that not only was the diagnosis of the the ai much better than that of the individual doctor but on top of that the way that they communicated around it was preferred by the human versus the way that the human doctor communicated around it And it makes a ton of sense because when you're a doctor and you've got 50,000 things that happen in a day, you might have lost a patient an hour ago and you just lost a second one. It's hard to stay as empathetic and to communicate real humanity in that type of situation.

And so the AI actually had a better result and was perceived as more empathetic. And so there's very few few things that are totally, actually, one of the interesting things is the things that are getting automated are exactly the opposite things of what we thought were going to get automated first five years ago.

Five years ago, you would have asked anyone in the AI space what gets automated first, they would have said, blue-collar jobs will go first, anything that's manual, we have robotics, it's gonna take a while, but those will be the first ones. If you're doing something manual, You have another three extra years compared to everyone else.

Robotics is another thing that's going to come, but you have at least another three extra years. That's a list of jobs that would be automated. What about jobs that will not be automated?

Shifts in Productivity

We are, as you think about these AIs, really one of the things that's happening is we're going from kind of single player productivity to multiplayer productivity. So instead of you achieving a task, you actually doing a task, you plus AI are doing a task.

This is for anyone that's managed a team before. That's been a thing for a very long time, which is not about your personal productivity, but how can you actually enhance it?

What it does mean is that there is no longer a single task that you have to do individually. So any task you begin at the moment can begin with brainstorming with AI, or before you even go to production with anything, double checking with AI going through. And that is a big shift that is happening at the moment.

It means you always have a feedback loop. Up until now, When you did a piece of work, only if that piece of work was important enough would you ask the opinion of somebody else, because why would you take their time if it wasn't important enough to the business? That no longer is the case, because the opinion of AI is basically, it costs zero. It adds value, but it costs zero, which means that there is no more task where you don't benefit from that first hurdle because it just fell through the floor.

The other bit, and I think the way that we can all think about how we're developing these skills, is to think about AI as a personal trainer because of exactly that feedback loop. Rather than kind of the tutor type paradigm, which I know a lot of people are talking about, the AI is always there and it's making you better while you're doing the work. not just you learn once like in a tutor or an education type environment and then you go and apply somewhere else.

It's actually helping you do the job better whilst you are doing it, just like a personal trainer would go and whilst you're actually doing the work, giving you tips on how you might improve the work that you're doing at that point in time. And this is one of the, I think, this is the way that I would be using it today. Any task I do is okay, how can I go and do this particular task in a better way?

Conclusion and Call to Action

Okay, we're now at the end. So, I mean, I thought I was going to get here faster, but we did have a lot of questions.

So all of that to say that I genuinely think or know at this point that now is the time to start to really talk about this. This is moving way faster than what people are talking about. There are a few people that are trying to... push this forward.

Like Ethan Mollick is a really good person to follow on a lot of this stuff, for example. But I think we need to have much more conversation about it and we definitely need to start to think about what all of this means. And if you haven't read it yet, single best piece of writing I think on this that I've read in the last few years is AI2027.com.

That's the homework I give all of you. It's probably, the timeline is fairly aggressive. I mean, 2027 kind of gives it away in the title.

Recommended Reading on AI

It's a fairly long read article, but absolutely worth to try and understand the various dynamics. It tries to explore not just the progress of the technology itself, but also how it plays out on the world stage and what happens when you have the US and China trying to one up each other on the technology. I mean, we're literally, it's interesting that this came out two months ago.

And the first two months that it talks about, it underestimated the speed. And it was literally talking about how we would end up canceling some of the laws in the US because we would want to compete with what happened in China. And literally last month, that is what happened here.

So there's a lot of different forces that are happening at the same time. And this article tries to really in a very good way explain where it might lead and what the consequences are.

Final Thoughts

So, hopefully, interesting one. Hopefully, the talk was interesting as well.

Sorry? What's the bioreference part? Read the article.

Oh. It's a long read. What's the cliff notes?

What's the general... You really, like, this is not an article you want to summarize. It is already... Put it in chat, GBT, and then say something. Don't forget about it. Can do that indeed.

I would really not, like it's already fairly to the point. It's not trying to be, it is a 30 minute read, but it's 30, if you try and summarize it, it will appear very superficial to you because the problem is there are so many details that get you to a particular result.

Hopefully this was useful and interesting today. Hope you had fun and hopefully see you again next month.