This podcast episode elucidates the intricate landscape of artificial intelligence (AI) and its multifaceted applications in contemporary business. Alan and Bhairav engage in a profound discourse aimed at demystifying the often nebulous concept of AI, highlighting its pervasive presence in modern discourse while acknowledging the widespread uncertainty surrounding its practical implementation. They emphasize the necessity for clarity amidst a cacophony of expert opinions and grandiose claims, advocating for a pragmatic approach that prioritizes tangible solutions over mere hype. The episode further delineates the various subsets of AI, such as machine learning and neural networks, and underscores the importance of data quality in harnessing AI effectively for business purposes. Through this dialogue, we aspire to equip listeners with the insights needed to navigate the evolving AI landscape with discernment and strategic foresight.

Takeaways:

1. The episode emphasizes that artificial intelligence (AI) is not merely a tool but a complex concept encompassing various subsets such as machine learning, neural networks, and computer vision.
2. AI's current proliferation is largely attributed to advancements in chip manufacturing and data processing capabilities that enable more sophisticated computations.
3. Business owners often face the challenge of discerning the practical applications of AI amidst the overwhelming hype and grand visions presented by purported experts in the field.
4. Effective AI implementation requires a clear understanding of the specific problem it seeks to solve and the quality and relevance of the data available for training models.
5. The conversation highlights the importance of due diligence when utilizing AI tools, as reliance on their outputs without proper verification can lead to significant errors and consequences.
6. Lastly, the discourse underscores that while AI presents remarkable opportunities for efficiency and innovation, it also necessitates a cautious approach to mitigate potential risks and biases inherent in its applications.

The real world entrepreneurship podcast with bhairav patel and alan clark.

Welcome to the Real World Entrepreneurship Podcast where we tackle the practical challenges that startup founders and small business owners face and give our own unvarnished opinions.

My name is Bhairav Patel.

I'm the Managing Director of Atom Ventures and Atom CTO where we partner with small businesses and startup founders to help them define and build their tech vision.

And I'm Alan Clark.

I'm the founder of the Business Growth Partnership and we work with high potential businesses who help them unblock the blockages that are preventing them from reaching their full potential.

Hello everybody and welcome back to the Real World Entrepreneurship Podcast with me, Alan.

And on the other end of Zoom is Bara.

Where are you today?

Hello, I'm in London.

I'm back a house here in London.

I'm in Lasgon and the wonders of technology.

It feels like we're sat the same rib.

Today is a podcast that I have been busting to do for a long time.

Those of you that know us, Bharav is the tech Jew out of us and I am a bit of a side of the shore in this area and where we are right now.

Obviously AI is a buzzword and a tool that's everywhere, but I'm not sure that I really understand what it's all about.

And today I was reading something on LinkedIn.

I just paraphrase a little bit of what it says that I thought absolutely sums up where I am and I suspect a lot of people out there as well.

And it says that AI is everywhere in the headlines, but most business owners still don't know who to trust or where to start.

Far too many overlaid experts.

I love that phrase overlay.

Experts are pitching grand visions while very few are actually rolling up their sleeves to build something that solves real problems.

What we need is less faith, more reality, less natoil, more solutions.

If that resonates with you, I think the next 20 minutes or so might be quite illuminating because I want to know about the reality and the solutions and I want to learn more about what is faith and what is net oil.

I know this isn't exclusively your area of interest, but let me just start by picking your brain, getting your point of view on.

For me, what are some of the fundamentals?

And the first one is what actually is AI?

You know, it's a phrase I hear all the time.

It's a buzzword that now reminds me of.com 25 years ago, where actually back then I'm not sure that we actually knew what the heck we were talking about.

And I suspect AI is a little bit like that right now.

It means different things to different people.

So what is AI?

So I guess we did been.

We have been through this with the Sanjay on one of the podcasts there.

But I can explain it in a way, hopefully that makes sense to everybody, but also kind of ones that sticks with you.

Right.

So AI is just artificial intelligence is just a.

It's our search to become or our quest to find computers that can think like humans.

Right.

And within AI you've got multiple different subsets.

So you've got machine learning and neural networks, computer vision.

So all of these things have existed for quite a while.

Actually.

The reason that we're seeing so much more of it at the moment is because of the fact that chips and the infrastructure has kind of has gotten together.

So I don't know if you remember that we, we haven't released it yet, I don't believe.

But we did a podcast with Sanjay and you know.

Yeah, don't talk to me like it's you and Sanjay talking like technology.

Talk to me like I'm an old man that doesn't care all this stuff.

And that's really what I want to do in this podcast because I think there is a Toha out there that have got a pretty good idea of what's going on.

I think there's a much bigger toolheart.

Think AI as I do.

I think AI is chatgpt or profile.

Yeah.

So that's where it's coming.

More than that, it's a tool.

But actually I think you've maybe articulated it in what you said there and what I took from what you said there.

It's more a concept than a thing.

Yes and no.

So, yes, in the sense that, yeah, there is a grand vision that we want to be able to create artificial intelligence.

So machines that think like us.

That I guess is the.

Is the vision, if you're going to put it in any way forward.

And there's lots of ways to the.

Concept, I would say, like that's the concept of the movie.

Machines that behave like humans.

Yeah, that's the idea that triggered all of this.

Right.

And I think when you're looking at it, it's broken down into a lot of different subsets.

So you've got, like I said, things like computer.

So driving cars.

Right.

Self driving cars.

Self driving cars will use a form of AI like computer vision to take the images of what's around you and then make decisions based on it.

Right.

So should I break, Should I move faster, can I switch lanes?

All that kind of stuff, right?

So it takes all of that data and then uses a, algorithms behind it to actually then determine on what are the next actions it should, it should do similarly with machine learning, right?

So we've all, we've had that around for, for years.

Even things like weather reports, you know, you're doing like predictive analysis on the weather's coming.

You know, you're seeing a weather pattern coming from a certain area, you're modeling out how that weather pattern will then, you know, move forward.

Will it, will it grow in strength?

If it's a typhoon or whatever it is, where will it go?

What other kind of features will play a part?

So the warmth of the ocean, all of these other things that can play a part as to where the direction of, let's say, a typhoon go.

So that's all kind of machine learning models which are just algorithms crunching a lot of data to determine an outcome.

And then you have things like neural networks and what else is there?

There's a bunch of, kind of subsets of, of the umbrella, which is AI, which is trying to all combine, can get us there to like a single talking robot that we, you know, the iRobot type thing that we want to get to.

So it's more than ChatGPT.

And ChatGPT is like one small subset of overall AI that exists.

I think, I think I understood that.

And all of these other tools, because I guess that's what they all are, are tools that are using a certain type of computing, are they all using the same AI, the same engine, or are there lots of different engines that allow different computers, different systems to think.

So I guess the thing to think about there is that it's essentially a set of algorithms that run.

Now, I'm not a great mathematician, right, but if any of the listeners out there who did their master degrees and et cetera, you've got lots of different algorithms that you can run on data to produce certain outcomes, right?

So one of the big ones you do in banking is all the Monte Carlo analysis, right?

So you, you'll have a, essentially a scenario and then run some algorithms to decide on what that, what the outcomes will be.

So essentially that's it.

And so what you're doing, the factors you've got are how much data can you crunch?

What's the quality of data?

Like, what are you actually, what questions you're actually asking, right?

And do you have the right data to answer those questions?

So there's the Algorithms themselves, which I would assume you could say, I like the tools that we're using to define, to shape the format that we want to shape.

The answer that we want is done by the algorithms.

Right.

The real crunching, the hard work is done through the GPUs, right.

So the whole thing about data centers and using all the water and whether they're environmentally Nvidia going up massively in stock, that's because.

Because these are chip manufacturers that have built chips now that are allowing us to do heavier and heavier computations on them.

That's why you're seeing such a surge now in AI and the generative AI and all these LLMs, ChatGPT, etc.

The reason you're seeing so much of that proliferate at the moment is because chip manufacturing has just got better.

We're able to now to be able to crush the numbers at such a volume that we'll be able to do these cool things that are coming out there at the moment.

So you've got one aspect which is the infrastructure, which is essentially the computers and servers that you need to run on, which are getting bigger and better and able to crunch bigger numbers.

And then you've got the algorithm which have always been there.

Right.

They've been there since the days of the Greeks that we're using to crunch those numbers.

So do I need lots of data, specific data about my particular problem before I can use AI as a business tool?

So generally, businesses.

One of the issues with small and early stage businesses is retained lots of very much data.

Yes.

And so you do the large language models that you see out there, like ChatGPT, Anthropics, Claude and all that lot.

Right.

They're using vast amounts of data from, let's say they scoured the Internet.

Right.

And they're taking huge amounts of data from there and they're churning that.

From a business perspective, you don't really necessarily want that.

Right.

It's not only like a sledgehammer to crack a nut, but it's also the wrong type of hammer to crack that nut because essentially it's just a.

It's a tool that's too big for its purpose, which is your purpose.

Right.

What you really want is something that's.

Well, there is kind of prolific now is more small language models which aren't trained on as much data, but are trained on specific sets of data which are very relevant to what you're trying to do.

So a good example would be in the legal sphere, if you're a, a large law firm that has written tens of thousands of hundreds of thousands of contracts.

You can use all of the contracts you've written, put them in and train a model on them in order for you to write more effective contracts in the future.

Because you can always say here the contracts that we've written.

And then you could feed it with outcomes or any case law or anything else that was decided on by these contracts and then use that in a more specific way.

Well, effectively that's like mining your large amount of intellectual property that your business has grown over the years and then using that to power your own tool that presumably is bespoke to your organization because only you have access to that data.

Yes.

Because you don't need a model that is trained on Middle Eastern recipes to figure out whether your contract is the correct or not.

Right.

Your legal contracts.

Right.

That's.

So that's, that's.

You have to use in chat GBT to look at your, you know, employment contract.

Doesn't make any sense.

Right.

So I'll give you a good example.

Recently I was working on a project where they have, it's a solar energy business that has a lot of contracts, right.

And so what you wanted to do was to extract information from those contracts so that you could, like if you've got, you know, 100 page contracts, there's a lot of clauses in there that you know, you need to surface and monitor whether it's SLAs or whether it's, you know, penalty clauses or whatever, that you know your own duties and responsibilities, you want to surface that up.

And it's very difficult if it's all sitting in documents.

Either you have to hire people, a team of people to literally note down what's in the contracts and then keep an eye on them.

Or you can use AI to extract the information, store it in a database and then people can search on it.

Right.

Or you can send alerts.

So you know, when I created a, the models already pre exist, so there was a model that you can get through Power PowerApps.

Microsoft had this as a single power automate and you can put a minimum of 10 documents through for it to recognize various clauses.

And so the way it works is that let's say I've got a document, a legal document with various clauses, right?

Severability terms, liability, etc.

I can put them into the model, tag various data that I'm interested in.

So in this example, it might be the address of the client or it might be the amount I need to pay and the frequency I need to pay.

So you put them in, you train them model to say, this is the piece of information which says fees and it comes in this part of the document.

So you just kind of tag it and then you do that 10 times.

And it's all probabilistic.

Right.

So you gotta remember all of this is about probability.

You then train the model to say there's 10 documents that are quite similar.

This is all the data I want to get out of it and I've tagged them.

Then what happens is the model trains itself.

It says, okay.

It starts to look at the things that you've tagged and say, okay, yeah, okay, this looks like a fee because it's next to a currency symbol and it's in a wider context that talks about fees.

Right.

So it understands that.

Then you come along, put a brand new document through it and then what it does is it essentially does some probability analysis to say, okay, I believe that this field is a currency or sorry, is a fee.

And I'm doing that based on probability of what I've learned from the other documents I put through it.

And so when you test it out, it'll give you a probability score of 90% because it thinks it's 90% accurate that this is it.

Or it might be 95%.

No, it's generally 100%.

But then it might give you a score of 40% and you might see actually it's not a currency, but it's actually the size of the thing and you'll correct it to say, actually no, this is the size of the plot of land or whatever it is.

I can see then that if we are an established business with lots of relevant information, be that sales data or written records or plans and executional detail or contracts, that I can then use that to build very scalable solutions.

Exactly.

For my business.

It's also reasonable to say though that if you don't have lots of existing information, I. E. You're a startup, so you probably have not much more than a pad of paper and a pen shop, that one can still use that publicly available information, that stuff that OpenAI and everybody seems to be scraping now to train their AI models because it's a whole lot better than nothing.

Yes.

So there's a few ways around it.

Right.

What you can do is rather than trying to go and steal everybody's data, what you can do is there's a couple of things.

There's a lot of pre trained models out there, so you can go out and a lot of the infrastructure providers like Azure or Google, GCP and Amazon, they provide you with off the shelf models.

Even Meta, Meta's got Llama, which has got pre trained models for different things, right, that are trained on a certain area.

So for example, you might be able to go out there, find a model that's trained on four legged creatures, right?

Then you say, okay, so this model is recognized, can recognize images of four legged creatures.

What you then do is say, okay, I want to recognize Friesian cows, right?

And then you take that model which is already trained to look at four legged animals, and then feed it your data, which is all going to be freezing cows.

So then it learns better, which is a freezing cow, which is not a freezing cow.

So then you're, you're accelerating your development by taking something off the shelf and then just training it through that.

You can also there's other techniques where you can have a model off the shelf which is not necessarily trained on your data, but it can use your data as a reference point.

So that's the whole rag thing.

The rag.

It's another technique where you can actually have a model that doesn't necessarily need to be trained, but is entrained on, let's say interpretation or semantics or something like that.

And then you can use your data whilst it's using it.

So you don't actually have to train the model, you just use the model as a kind of a tool on your own information.

On to my next question.

Now, one of the great things about programs was that by definition they did what they were told and only what they were told to do.

So they were unbiased in their analysis.

Is AI actually unbiased or because we're encouraging it to think, can that draw a skew and it think the wrong thing without me realizing?

Yes, because this is the whole hallucination and bias.

So the one big thing that they talk about when everything came out right was the bias on image creation where everyone was white because all he was trained on were white images, right?

So then he didn't really understand how to represent.

So there's that bit, but that's because it hasn't been trained on the full subset.

And also if you think about it, most the Internet is English.

So now they're coming out with like Hindi models and Tamil models and all that stuff.

Right?

But until we can get all of that together, that's, I mean it's happening now, but it'll be biased.

I think what you're seeing is AI result are biased, inherently biased.

Only if you're using something well, yeah, exactly.

In that sense, yes, but it's everybody biased.

It's kind of entrained on everything, everywhere, forever.

No, I want an interesting part here which leads me on to my next question, which is, does AI think like a human does?

Two elements to that, two elements to that is no matter how unbiased we think we are, we all have inherent bias.

But it's a.

We may be so unbiased that we are wisely unbiased.

Yeah, go on.

So I think we're going to think of it slightly differently because it's all probability right at the end of the day.

So the bias is something that, let's say if we're thinking that it's culturally biased, that's something we're inferring on the, on the results.

What it is, is.

And there's a great article on this actually, because it's essentially the way that everything is trained is probabilistic.

That's why when you put the same question into a chatgpt, you'll get.

And you put the same question in twice, you might get different answers.

Right?

Because it's doing a probabilistic search.

So it's coming back and saying, okay, I think that this is the answer.

It's all bits and bytes.

It's not actually understanding anything that is written.

What it's doing is it's saying, I'm taking it back into bits and bytes, saying, I think that this is this and it's meant that.

And so therefore I'm going to reply with this.

So AI doesn't think it has lots of sizes.

It looks at information as it all upstairs.

And this is the answer that is most likely to be accurate based on information that it has available.

Yes, because, you know, it doesn't think in the way of the human things.

Right.

In that sense because we will take whatever cultural influences and you know, experience and whatever else that we've had through school and parents and friends and use that experience in order to come up with something.

So I don't need to be worrying about my laptop being an amazing with me and making my life hard?

No, I mean, they do, they have been giving him personalities recently, right.

So they've been instructing them with, you know, to be nice and fresh and happy and all things like that.

But no, it's not really, you know, it's not upset with you if you don't say thank you or please.

Yeah, I was actually, I was quite convinced the other day.

My computer, my laptop was in a mood and I said, we had always next thing to adopt AI into my business.

How much coding skill do I need?

Is it a coding exercise or how do I go about doing it in the most simple conceptual terms?

So a lot depends on what problem you're trying to solve, right?

So the question is what do you need AI for?

So for example, if you are a, if you're a business that does a lot of interpretation of documents and data, right?

And so let's say for example, you need to, again going back to the contract example, you need to interpret lots of contracts.

You can use AI to help you as a tool.

So you could just subscribe to one of the platforms that are out there, you know, Gemini, Perplexity, whatever it is, and upload the document saying help me find all of the, you know, give it some context and then say find me some of the clauses that you know are interesting because of X, Y and Z.

That's easy enough to do and enough to implement.

There's lots and lots of other small tools that are coming out there to, to make it easier for you to, you know, manage emails, etc.

Etc.

But if you're looking to do something bigger then you've got to really sit down and think what am I using AI for?

Do I need to build it?

Do I need to buy it?

Well, I think this brings us back to my sort of initial point which is there are businesses who are seizing AI and have been for a very long time and using at the heart of their business.

But the majority of the people who will ever run a business, they will ever listen to our podcast.

Who will ever exist in the real world?

I suspect they don't need complex solicit.

I suspect that most things that they need should be relatively easy to do.

And a little bit like going back to dot com and the Internet when that was a broad concept that no mess, the technology understood well.

But your average banker didn't understand what was going on other than it was really exciting and it could do lots of things.

All we had to do was give that a little bit of time to mature and all of a sudden there was websites that were useful and then there was things like online banking that meant I didn't have to go to the bank anymore.

That meant there were all of these Internet based tools let me in my life easier and I didn't have to invent them.

So yes.

So I'm not going to say it's like the dot com boom.

I think there is a lot more hype than there probably was.

I mean I know, the 90s, the hype was huge, right?

But actually hype wore out and our lives have changed because of it, right?

It's massively changed the way that we exist.

How AI plays out in the next 20 years is a very, very interesting.

Conversation because of, I think all I'm trying to make.

I don't want to overdo it, but I was trying to draw a parallel to a time in my life where a technology emerged that changed everything, like the Internet.

You know, I'm so old.

I remember before the Internet was anything other than connecting a few academic institutions.

And then within five years, it exploded.

And I remember the first time I really heard about AI was four years ago.

Maybe it was a very senior job representation.

We may often say, I've just come back from California, there is a tidal wave coming like we have never seen before.

And it is a rolling out of AI into the thought normal world.

And my God, was he right?

And it's true.

And look, for example, things like meeting notes, right?

Meeting minutes, you know, if you subscribe to teams or even zoom, right now it's got an AI companion and you can all of the notes come out perfectly right?

And so no longer do you have to write meeting minute notes.

And we're all pretty bad at writing those notes to begin with.

So that's helped expose a lot things of right, in the sense that you now have a set of notes that you can take actions from.

You can actually go back and understand what you talked about.

We don't have to do all of those meeting minutes anymore.

And I think that's a great productivity tip.

But that's nothing.

I know another very simple one is I've got all my cot to my customer take for the last nine years on my CRM.

I can stick that into chat GPT and ask it to analyze who are the growers, who are the followers and are there any characteristics and why they fall off.

The kind of exercise that was literally impossible in years gone by without dedicating teams of people and your small business to try and look at patterns that you couldn't see.

And that's great, but that's not a big why you spend billions and billions of dollars, right?

The reason that you spend billions and billions of dollars the way matter spends billions, billions of dollars is because they want to be able to do things like personalize the ads directly for Alan, right?

So they'll say, okay, Alan, we know what Alan's doing.

He's looking at.

And so therefore we'll create on the fly Personal ads directly to you.

That's how we'll get our money, the marketing money.

I understand that matter.

I'm thinking about people who are not in the business of AI, but they're in the business of what am I looking at?

I better be careful not to give away any secrets here.

I'm doing a project now where I've got some presentations on Wednesday.

One of them is out about how to spot when structures are about to fail.

Another one is Josh.

I'm trying to remember what the other ones are anyway, but these are not AI technology projects.

Every one of them said we will use AI within our business.

We should not demand should allow us to analyze and understand the markets we're in and do deliver things that were otherwise unavailable.

And they're not spending millions, they might be spending tens or maybe £100,000 in that area, but getting an enormous return compared to where they would have been 10 years ago when they said we just can't do this.

And that's the point.

Right.

So there's two things in there.

There's one where it will be life changing.

So a lot of the stuff that you're seeing right now in health and medicine coming out now with different diagnoses, different tests, even with the ability to perform certain surgeries, right.

That, that is the big stuff.

That's.

And again, things that people like Metro doing, etc.

Those are the big, the big bits when it comes to business.

It's always about understanding where, you know, how can you use AI to increase efficiencies, increase your revenue, go to market quicker, new products and services, those kinds of things.

Right.

And those aren't necessarily big spend items, but it could just be ways of connecting different areas within your business that weren't able to be connected before or to outreach to clients in a way that you wouldn't be able to do it cost effectively before.

Right.

You might have to hire someone in the US to do it, but now you can actually use AI tools to do that for you.

Right.

And I think that's it.

That's where it's not necessarily life changing, but it becomes just more automation, which is what tech has been evolving into over the last 20 or 30 years anyway.

Right.

But having said that, we're still talking about digital transformation.

But.

Well, towards that another thing, is it fair to say that any job site for an analyst in IT is in a bit of trouble?

Yes, I think so.

It's one of these ones where it's quite an unfortunate one simply because you need to get to the analyst step in order to get the experience to go and do the big thinking, right?

And if you haven't gone through it.

So there's two ways of looking at it, right?

There.

There are certain tasks that analysts did which are really just terrible anyway.

That was a big chunk of their time, but now speeds up and I think the.

So I'll give you an example.

I did paralegal when I was much younger, right?

And part of the job was proofreading documents for spelling errors and, you know, grammatical mistakes and things like that.

You don't need that anymore.

We've had spell checker for years, right?

So that, that kind of bit is gone.

But also what that means is that you could spend more time thinking about the content rather than the form.

So ideally what you want is you.

You.

You'll still want analysts, but you want analysts to.

To do more efficient and to be able to get the data sets and do work with those data sets more efficiently and much and much deeper.

But you may not need as many analysts as you have.

Really dumb question.

Is spellshare an example of ea?

No, I think that just is a dictionary.

It basically takes a.

It looks at words and essentially does a comparison.

Right?

The reason I asked that was because spell searching now seems to be getting a whole lot better at telling me the words that I meant to use rather than the words that I put in that it was, lets just say, imperfect.

And that leads on to my last two questions, which is, that's wonderful when it makes a suggestion and it's valued and it helps.

What happens if AI makes a mistake?

What happens?

Who's responsible when AI gets it wrong?

Is it the person that developed the AI or is it the person that used it?

Or is that expensive?

Court.

So this is the thing which is quite interesting and I had a little discussion with a lawyer on this about AI and the law.

I think there's a couple of things to separate here.

Firstly, who noticed it?

Will you notice the mistake?

That's the first question, right?

And a lot of people start relying on AI in a very lazy way where they'll just put something into ChatGPT, spin it out and then go for it.

Right?

And that is something where you really need to start thinking about checking what is actually coming out, because you wouldn't necessarily do this in any other circumstance.

But for some reason people think ChatGPT or any of these LLMs are flawless, which we know they're not.

So first thing is make sure you have your checks and balances to ensure you're figuring out what the mistakes are.

Right.

Don't blindly just embed it into your system and let it go.

Second thing is the fine line of whether.

If it comes out with one plus one and says it's five, whose fault is that?

Well, yeah, to some extent, it's going to be the fault of the people creating the language model, right?

The one that you're relying on, whether they'll take liability for it.

Probably not, because what they're going to say is, it's a tool.

And they have a lot of disclaimers when you're using the tool that it can hallucinate, that it can make mistakes, and it's up to you eventually to check what is there.

So if you're taking someone else's work and then applying it into your own business, it's your responsibility then to make sure that what you're applying is correct.

And you do the same thing with a calculator.

Right.

Like if you had a calculator and one plus one equaled 11, you'd think, okay, maybe I missed the plus sign.

Right.

You'd do something about it.

Right.

And you'd make sure that you wouldn't blindly just go, and.

My last ration is AI safe?

Am I in danger of having my business that I have not just loved and grown and developed, and now I'm going to seize this AI opportunity?

Could it be the thing that ruins all that I have done?

Yes.

And I think that's not wrong.

I relied on computers, not people.

I think the answer is a simple yes, simply because people are relying too much on AI to do the things for them.

So a great example at the moment, especially in our world, there's lots of entrepreneurs using AI to create technology.

Right.

And there's loads of stories out there, and I've come across quite a few of them where they're building their MVPs using AI.

So cursor or lovable, and then going beyond MVP and then just spending all their time building, learning how to code.

I mean, there's.

How many people you come across, oh, I learned how to code.

So did you really just learn how to prompt an LLM, Right, to produce something?

Like, it's like me saying, I know how to build a house.

I put four.

Like, you know, I put a bunch of planks in the.

In the ground and then put, like, one across it.

And then, yeah, that I've got a nice house.

It might look like a house and serve like a house, but wind will come and knock it over pretty quickly.

Right, yeah.

And I think this is the key.

So a lot of what people are doing is they're relying on AI because it's AI and they're just assuming that what it does is correct.

That's the risk.

A lot of it's you not double checking what's going on and you not actually incorporating it into a wider process that makes sure that the outcome's correct.

Is this a new phenomenon?

No, I think people have a tool available to me or a tool, a facility available to me.

It will do what I tell it to do and on my head.

Beer.

You know, you love your analogy of building a house.

Well, I love to build a house.

I know how to talk.

I could say to a builder, build me a house.

I want you to build it over there.

I want you to make four rooms and put a roof on it.

You might do exactly what I asked.

And I go, where's electric?

We never asked before that.

Where's the plumbing?

Well, you never asked me for that.

It's sensing.

Well, you didn't ask me to put foundations in.

Personally, my view on it, for what it's worth, is, yes, it's not perfect.

Yes, there is risk, but I have a feeling that it's a lot less risk than getting humans to do a lot of it.

It's actually built on rationale.

And why are large language models so.

Language models.

Because there's a hell of a lot of data going in them to come up with the answer.

As you said earlier on, really we're looking at probability of being wrong.

You know, if we rely on it to answer the most complex micro issue.

Well, there's not a lot of data, is there?

But if we're using it to help us to address questions and challenges and analysis for what we have to.

I think she's been really helpful, but thanks a million.

I got to ask a lot of questions that have been plaguing me.

I understood most of the answers.

I think don't ask you.

Did it actually make sense to you?

Yes, but I think I'm still finding my.

And I will for a very long time find, be finding my way through this.

My perspective on so many of these things is you need to be aware of it, but go carefully and make sure everything you're relying on, you've done your due diligence.

And if you truly believe that spending a thousand tons on the eye is going to transform your business into a global powerhouse, you're probably a little bit mad.

But it goes back to what we've always been saying on this podcast, right?

It's risk and roi.

We've all said this across.

Do the roi.

Is there ROI in implementing this and what are the risks and then off you go.

Brilliant.

Well, thanks for joining Bharav and I on this episode.

We do have more on AI but unfortunately our social spiritual has lost his voice which doesn't make him the ideal podcast guest.

But as soon as we have voice back online, audience configured a bit of AI that can take his brainwaves and turn them into audio.

We'll be back to see you soon.

Thanks a lot guys.

Bye for now.

Many thanks for taking the time to listen to the podcast today.

If you like what you hear, please leave a review or subscribe.

You can find us on Apple Podcast, Google, Spotify, all the main podcasting sites and if you'd like to learn more from us, you can Contact Alan, he's alanbgp.co.uk and you can find me at infoatomcto.com.