Generative AI Is the Travel Industry’s Future, Get Used to It

Something shifted in the last two weeks on the zeitgeist about the use of artificial intelligence in our daily personal and professional lives.

The launch of the first large-scale, general purpose chatbot using OpenAI‘s GPT3 AI engine on November 30 has reenergized the whole tech industry all at once. I wrote a story on it which will give you a good sense why.

To get an understanding of why there is so much buzz about Generative AI – the sub-sector with larger AI world which includes creation of text, images, audio and video – and what this means for our daily lives, for the travel industry and even travelers, I talked to the best expert analyst and writer on it I know, David Mattin. He writes an excellent newsletter called New World Same Humans on trends, technology, and our shared future and has been doing a deep dive into Generative AI all this year with his writings.

This is a fascinating conversation you would want to listen to from start to finish, to understand the implications of it for our industry and indeed our daily lived reality.

Subscribe

Apple Podcasts | Spotify | Overcast | Pocket Casts | Google Podcasts | Amazon | RSS

Episode Transcript

Ali: Welcome to the podcast, David. David Mattin, who I’ve known for many years. I used to know him when he was running trends and insight for TrendWatching, which is a trend watching consultancy called TrendWatching that we used to be good friends with. I’ve known the company for a while and since then he has started, he since left and started one new newsletter which David if you want to talk about, and in which you’ve been writing a lot about AI and its effect and a particular sub area of AI that we’re going to talk about today. What it means for the travel industry and what it means for content creation of which is a huge part of the travel industry as well. So talk about what you’re doing now. You’re the futurist. I don’t know if you’d like to be called that because I know a lot of folks don’t like to be called that.

Mattin: Hey Rafat, thanks for having me on. The newsletter is called New World Same Humans and it’s a newsletter about trends, technology, and our shared future and it really is underpinned by this idea that so much of the human story, our history, but also what’s ahead of us, our shared future, is fueled by this collision between a changing world, often emerging technologies and fundamental human needs, this eternal shared nature we have that doesn’t change, and it’s in the collision of those two things, often in the collision of a new technology and a fundamental human need that our future emerges, that the human story emerges out of that. So New World Same Humans takes up a lot of my time these days and I love writing it. It’s been an incredible journey writing this newsletter and building this audience, was born in the pandemic.

And exactly as you say, this year I have been obsessed by, and the newsletter has borderline been taken over by this incredible story we’ve seen around generative AI, large language models, GPT-3, we’re going to get into it all. We are amid a revolution when it comes to artificial intelligence and its impacts and its use cases. And just documenting that has been such fun and I hope useful across the last year or so, because you really see that story playing out, this collision between this amazing new powerful technology and fundamental human needs like creativity and convenience and knowledge and so on.

Ali: And so generative AI is a sub-sector of the larger AI. It’s called generative because it generates output based on input.

Mattin: Exactly, yeah. So generative AI is a kind of AI, it’s a subset of AI. And it’s AI fueled by, founded in essentially something called transformer models and Google Engineers released a paper, the first paper, the invention moment about transformer models back in 2017. And a transformer model is essentially a new kind of neural network that spots patterns in sequential data. It attends to the important part, the important parts of and spots patterns between elements of sequential data, data where the sequence of the elements is important. For example, the words in a sentence, that’s an example of sequential data.

So it attends to, it has ways, it uses incredibly complex mathematics essentially to attend to the important elements in that sequence of data and build a model of the relationships between those important elements. Obviously it’s hugely technical stuff, but it’s building a model of the relationships that underpin what we humans call meaning. So it can start essentially to simulate, produce a simulation of meaning. And one of the consequences of this is that these transformer models allow us to build large language models that have these incredible linguistic, these incredible languaging abilities that have set the world on fire recently with tools such as GPT-3.

Ali: And so generative AI, it seems the word only came into certainly my lexicon this year and lots of investment is going into it. Google bought a company called DeepMind a while ago. I don’t know if it’s technically generative AI, but it’s in the AI space that apparently from what I understand they use in different parts of the business, it also is used a bit in the consumer facing part on the search engine that you see if you are on the top of search, if you search for something, they will give you an answer immediately or short form of an answer. And there’s a lot of investment. It’s interesting that venture investments happen in waves and it was Web3 and crypto last year. Obviously this year has gone completely bust on that, or at least second half of the year has gone completely busted on that part. And it looks like generative AI is getting tons of investment on it. So why should we care?

Mattin: Yeah, I think there are a set of really powerful reasons why we should care. You are right about DeepMind. So for example, they’ve used a transformer model to solve the protein folding problem. This iconic problem in the life sciences about the prediction of the structure of proteins, and proteins are large chain complex molecules that are essential to life, that underpin all life. And if you are able to predict how a protein will fold and its end state structure, that is hugely powerful and it will unlock massive advances in the life sciences and in medicine. So DeepMind did that. The protein folding problem is essentially solved. Now, we always believed that AI would be hugely powerful and hugely consequential when it came to solving those kinds of problems, scientific problems, problems that are about crunching huge amounts of data and information and so on.

What we didn’t expect perhaps, and certainly not to the extent that is playing out now, is that AI would intercept with another set of fundamental human needs and impulses around creativity and art and the human needs around emotion and meaning, because just as you said, what’s particular about generative AI, what’s particular for example about these large language models is you can give them a prompt and then they will generate, they will output language, they will output relevant text that didn’t exist before. This isn’t about them going to a database and finding units of text and bringing them back to you-.

Ali: Which is how search works.

Mattin: Exactly. Which is how the current model of search works. It is founded in going to a huge data set and bringing a relevant part of it back to you. Generative AI is doing something totally different. It’s generating an entirely novel output but a relevant one based on some input you’ve given. So you can say to GPT-3 and we can talk a little bit more about this, for example, write me a short story about Tinder in the style of Charles Dickens and it will generate something that sounds just like Dickens, is all about Tinder and didn’t exist before. And this is what has set the world apart, or what set the world on fire in 2022 because it feels as though AI now is encroaching on something we felt was uniquely human and was never going to be touched by machine intelligence, which is creativity. The ability to create. That is something fundamentally new.

And then that has huge implications across all kinds of domains and all kinds of industries. This is why we should care. And we can talk a little bit more about the advances we’ve seen this year beyond GPT-3, which was launched in 2020 that are going to be put to use inside industry and that are fueling these investments.

Ali: So, you’ve mentioned GPT-3 a few times. Explain what GPT is. It’s part of this company called OpenAI, which is a company that’s now very heavily funded as well. So it’s a startup that was started by the Y Combinator founder. So not founder, former CEO.

Mattin: Exactly. Yeah. So GPT-3 is essentially all about applying a transformer model to language, take language as your sequential data and user transformer model to build models of the patterns and underlying relationships that weave language together. And so essentially what GPT-3 is doing is when you give it a prompt, it’s using incredibly complex statistical analysis essentially to make a decision on what the most likely next word would be, because this is a transform model that has been trained on a huge amount of text, like an appreciable segment of all the writing humans have created ever.

So you take that huge data set, you train a transformer model on that data set and then you can step to that transformer model and give it some language and it will make a set of decisions, a set of considerations about, given its huge dataset training, about what the most likely next set of words would be. And so what you get is a machine that’s able to output text, novel text, something entirely new, that sounds just like, just as though a human would’ve written it, because it’s basing those decisions on its incredible knowledge, if you like, of what humans tend to say in these circumstances-.

Ali: And in a matter of-.

Mattin: … that I hope is a broad but useful summary of what GPT-3 is doing. And it was released in 2020. Previous iterations, GPT-2 obviously was earlier, but it was GPT-3 in 2020 that really first set the world on fire about large language models and generative AI because it was a step change in the ability of this tool to generate language that sounded human, that sounded just like a person might have said it.

Ali: And GPT-3 is, and I don’t know what the right word is, but is it owned by OpenAI technically speaking?

Mattin: Yes, I believe so. And OpenAI take a lot of funding from Microsoft.

Ali: Okay. And-.

Mattin: And Microsoft money is helping this happen.

Ali: Which is very interesting.

And obviously there are other companies. I saw a company called Hugging Face, which I guess a cute name, and a bunch of other startups in this sector. So let’s focus on what happens, some of the developments that happened this year, which is, and again I won’t use the right technical words, this is why you’re the expert, you ask one of the tools that OpenAI released was something called DALL-E, which is you input a text and it would create an original image based on your, in a matter of I guess minutes or something, the chat is what we’re going to come to is in a matter of seconds, but I guess image obviously takes a little more time to create, or a set of images. And then also maybe it was Google who launched this, but text to video as well.

Mattin: Yes, yes. So that’s what we’ve seen across the last couple of years. So OpenAI released the first version of DALL-E, their text to image tool early in 2021. They kind of revised it. So DALL-E two was released April, I think 2022, April this year. Yeah. So this is about training a transformer model on an appreciable section of all the images humans have ever created. Okay. And then marrying that to a language model so that you can write then a request or a prompt like, okay, draw me a picture of a puppy in the style of Picasso. And the image transformer model will essentially throw a completely random set of pixels at a Canvas and then it will go back and forth iterating that. It will go to the text prompt and ask itself, does this look like what I’ve been asked for? No. Okay, revise, does this look like it? No revise, revise. It will iterate back and forth, back and forth a zillion times until it satisfies itself that the image it’s created looks as the prompt would expect it to look.

And then it delivers this image to you. Again, a somewhat technical explanation for what is a magical effect for the end user where you can just write a simple sentence and see an incredible beautiful image generated in a matter of, yeah, yeah, maybe 30 seconds. And stable diffusion. We have to mention another texter image model that was released early in 2022. They’ve just released their second iteration last week and it’s a step change as well. The photo realistic image generation ability now is just astounding. And then yes, as you say, you have Google and others working on text to video. So Google Imagen is a tool now where you can write a short prompt and get a short five, ten second video clip, and they’re working on longer prompts to generate longer video clips.

Fast forwarding, if you listen to, for example, Emad Mostaque, the founder of Stable Diffusion, he’s talking about a world ultimately of text to anything, write a text prompt, tell the tool what you want and it will generate you a PowerPoint presentation. He’s working with the music industry he says on text to music. So he wants to train a transformer model on all the world’s music and then you’ll be able to type, compose me a sad ballad in the style of 1920s jazz, whatever. And out it will pop. Just a revolutionary tool for the amplification of human creativity put in the hands of billions of human beings. That is the end game that we’re talking about, that’s got everyone so excited.

Ali: And all of these tools for now are obviously experimental. What you have stressed a few times is this technology is changing so fast that what we thought was possible last year has completely changed this year. And even that has changed multiple times this year.

Mattin: Yeah. Yeah. Look at DALL-E two released this year and everyone’s like, oh my god, text to image, this is incredible. Look at the quality of these images. Months later you have Google Imagen text to video, they’re working on text to long video, text to movies essentially. The pace of iteration and change. And then, you touched on this, we had ChatGPT just last week, a new version of GPT-3. So the pace of innovation across this year has just been absolutely astounding.

Ali: And let’s chat about ChatGPT, which was a tool that OpenAI released for anyone to use. I think it’s chat.openai.com maybe. I think that’s the site. And GPT-3 I guess is the newest engine that they’ve used, apparently GPT-4 is coming as well soon from what I read. Where I think the power of text to text images and video, I think it was a little clunky for people to understand because there was I think more steps, but text was so simple and it was released on last Wednesday and by Monday the CEO said that a million users, but also just it took over all of Twitter. Even Elon Musk drama that was playing out was sidelined to an extent. And I guess turns out Elon Musk is an investor in that company as well.

Mattin: I think, yes, that’s right. Yeah, I’m pretty sure that’s right.

Ali: Yeah, I think he’s an investor in OpenAI as well. And it took over Twitter, also took over LinkedIn where people were trying to figure out, obviously this is LinkedIn, so everybody’s trying to figure out what is the business use case and how will it disrupt this, and how will it disrupt that? And so this, we’ve known chatbots for a while. This is, as you said, a whole different level of chatbot that has… And chatbots were used and are still used, I guess there was a lot of hype a few years ago, but are primarily now used for customer service, I would say, for simple queries and then it goes to human if it becomes more complex. So, if you do WhatsApp within airline, this is what you would probably get. But the chatbot of ChatGPT is a whole different level. It’s basically you ask them anything however short, or however detailed and the tool spits out in a matter of seconds, as you said, in a confident, coherent way, even if it’s not a hundred percent right all the time.

So what’s your sense of the tool itself, and what are the pitfalls as you’re seeing it so far?

Mattin: Yeah, a couple of things happened last week. So OpenAI released a new iteration, if you like, of GPT-3 that is even stronger and even more robust than the previous model. So its text-davinci-003 is just a more robust model. It’s better at generating long form content, it produces a higher quality of language and across long form it remains more coherent. It remembers what it said better and remains accurate across long form. So they released that, that was hugely exciting. Everyone got excited. Then days later they released ChatGPT. And ChatGPT is really, oh, I say just, it’s no small thing, but it’s GPT-3 optimized for dialogue, for chat interaction. So you can have more of a conversation with the tool now. And this is all about the big story since they released GPT-3 in 2020 has been about making it stronger and more robust and making it easier to use.

And ChatGPT is just a huge advance along that journey of ease of use and an interface that really makes people feel comfortable because you can now have a dialogue with this thing and just as you say, you can ask a question or put in a request or give an instruction and it will respond. And that request might be, write me the code in Python for a very simple website that looks like this and does this, and it will literally spit out that code, or you could ask it, plan me a really fun itinerary around New York in a couple of weeks that’s for someone who’s interested in history and modern art. And it will spit out something very coherent and robust that is brand new, that didn’t exist in the world before, to answer that request.

We’re going to see… It’s so funny. Yeah, you very rightly touch on chatbots and there was a huge load of hype around them a few years ago. And it’s so funny when you watch trends as I do and you see hype grow around and innovation like that and then die away and people become very skeptical and say, oh, chatbots are a waste of time. That was all hype. We came to the idea a little early, the technology wasn’t really ready yet, but the fundamental idea of some virtual entity that you can talk to very naturally and that understands you and provides relevant responses, that is clearly a powerful idea, that is clearly a valuable idea. Now we are seeing with generative AI, with GPT-3 and so on the emergence of a technology that can sustain that idea. And yeah, the implications across industry for customer service, for the way people will expect to have interactions with organizations and with brands, those implications are transformative.

I’ve talked for a long time about a trend that I call virtual companions, the rise of AI fueled entities, conversational entities that become a counselor and personal assistant and even a friend to the user. This AI you have in your pocket that knows you better than anyone, that is your guide to the world, that knows your most intimate secrets, that knows everything about you. I think that, that is potentially the next big life-changing revolutionary technology for billions of people, that does, that changes life the way the phone did. We haven’t had something that changed like the way the phone did since the phone. I think potentially that is the next big thing and I’ve been thinking about that for years. And part of why I’m so excited about generative AI is because I think we are verging on the ability to create that kind of virtual companion now. And that is hugely powerful. And every brand, travel industry or not, needs to think about that world and how they live in it.

Ali: And I guess speaking of phone, it’s probably going to be coming to your phone much faster than a general online… The companion part will very much be part of the phone, I guess Siri was a very, very crude version of it.

Mattin: Yeah, exactly, exactly. And when I used to talk about these virtual companions, I’d say right now we have Siri. We already live in a world of AI fueled conversational assistance. You have Siri, you have Alexa, but it’s really like Siri, where’s the nearest bank? Or Alexa order some washing powder. It’s very functional, they’re not very good at understanding you. But even in that relationship you can glimpse the beginnings of something new. Skeptical audiences I would say, or skeptical clients, I would say, look, have you ever asked Siri a question with any emotional content at all? If you have, you are already starting to admit to yourself there can be a deeper relationship with these AI agents, with these virtual entities. Now we’re seeing language models that can talk to you, and it is impossible to tell that you are not talking to a real person and they can learn all about you, they can learn your preferences, they can respond to you on an apparently emotional level. They can simulate all of that incredibly convincingly.

So yes, exactly as you say, that is coming to people’s phones. I think that will become the primary interface via which we relate to our phones. This companion will live in your phone. And look, really, if you ask me where I think it’s heading, I think that in the end will become the point of the phone. The phone itself will just be this uninteresting little vehicle that provides you the relationship with the real entity, which is this AI companion, this person it feels as though that lives in your phone, that responds to you. That is where I think this is heading.

Ali: Fascinating. And we can mind boggles in terms of implications. Let’s talk about content creation of which one of the immediate implications of the tools that are there currently is “content creation.” You can take it for what it is, whether it’s for the travel industry and for agencies that work for them, I’m guessing there’s a lot of action on the agency side, ad agencies, creative agencies, marketing agencies on trying to figure out how to use this to be more efficient with their content creation, be more efficient with their images and audio and video, whatever else is to come there. So from that perspective, I guess generally for marketing, what do you see as the implications for marketing and I guess travel marketing as a result of it?

Mattin: Yeah, there are huge implications, and this is already very much in train. You have marketing agencies, you have platforms now, and you have marketing agencies experimenting with these platforms that are all about the generation of marketing copy and marketing campaign ideas and now marketing campaign assets, images via generative AI. So if you look at, for example, Jasper, that’s one of the most, probably the most popular, the best known platform right now that is explicitly about generative AI for marketing campaign copy, or you look at a startup called Copy.ai, that’s what they’re all about. And you have big brands using them already and you can step to these tools and say, generate me marketing copy for a campaign that’s all about travel to Israel in the summer for people that are interested in history and modern art. Again, I’ll choose the same traveler and it will generate compelling marketing copy in seconds, novel copy that didn’t exist before.

Now look, everyone’s going to have that technology, that technology will be commoditized. It doesn’t have to mean the end, it doesn’t mean, and it doesn’t have to mean the end of human creativity, but it means human creativity and the creativity of the marketer massively amplified by this creative companion that’s always there, that never gets tired, that spits out this relentless stream of new ideas and this relentless stream of copy at your instruction. And it turns the creative, in this case, perhaps the marketing creative into a conductor of an orchestra. Someone who is conducting and has oversight, a creative director who has oversight of this suite now of creativity that they can shuffle about and play with at will. That’s where I think creativity is going, and marketing creativity is going. Yeah. And so for the travel industry, you’ll have the ability to generate in seconds Twitter campaigns, marketing campaigns, television ads. You’ll be able to generate those in seconds.

It will be about the people who can whisper to these models most effectively, who can prompt these models most effectively and then iterate the outputs most effectively to come to the best outcome, because that is a huge skill in itself. And actually if you look online, we are already seeing the growth of prompt marketplaces. So people are selling their AI prompts to others like, oh, I’ve devised a prompt to GPT-3 or to DALL-E two or to Stable Diffusion that produces great results. And if you give me $50, I’ll give you the prompt. So prompt engineering as it’s being called, is going to be a creative skill all of its own.

Ali: Fascinating. Yeah, I had not even thought of that part. And as ChatGPT was released, I did a post early this week after playing with it over the weekend on what are the implications for the travel industry. And was instead of me guessing it, I just asked the tool itself and it came back with some coherent answers. And some of them were repetitive because I was asking, what is it for hotel and airline? And so, one of the things that it said consistently was itinerary generation, which I think you mentioned as an example as well, which is a huge part of the travel industry. This is what tour operators and travel agencies live off. And I’m guessing there’s already experimentation happening with at least a forward looking travel agencies or people, individual people within these agencies on saying, oh, how good can they be? Because a lot of, if you go to any of their sites, that’s half of their site and that’s how they sell these tours and to people.

And then the other part that I was fascinated by, and this is what you said about design and general creativity in general, something about hotels I asked, and it said about hotel design of how you can come up with new room concepts using and aircraft design. You could come up with airport design as well, which was just fascinating to me that, and I’m guessing some version of these already happening in specialist areas, we’re using different tools, but it just seems fascinating to me that this would become just a mass use tool.

Mattin: Yeah, absolutely. In my head, because it’s so mind spinning, when I think through the implications for the travel industry, I think about the implications for the industry operationally and how it works. And there you have huge capacity, as you said, for the generation of itineraries and the generation of flight schedules and timetables and all kinds of automated information flows. And you have, exactly as you say, the ability to do generative design now, generate me a hotel that will appeal to X, Y, Z people, that is placed in X, Y, Z city. And you’ll be able to generate that design and iterate it and change it and generate other versions. And that will revolutionize design across industries, including the travel industry.

And then you have ways to serve the customer. And, of course, itineraries and schedules are really powerful there. But then you also come to some of the virtual companion thinking that we were talking about before where imagine the evolution of this incredible traveling companion that is a mixture of personal assistant, historical tour guide, counselor, someone who knows you better than anyone, who knows what you are interested in, who knows what you’ll love, and is just always there for you and takes you on this tour through whatever city or whatever village or whatever country you are in. That is the kind of implication that travel brands can think through in the coming year because that world is coming to us.

Ali: Yeah. And they’re probably a hundred users that we can’t even think of today that will-.

Mattin: Yeah. For sure.

Ali: … completely come up and probably startups that specialize in the travel sector, particularly on the B2B side, to help some of these larger companies, as you said, operationally become a lot more efficient or cheaper, whether I’m just making it up but operate an airport better or operate hotels better in general. The one question I had in terms of the, we’ve talked positive everything, the possibilities. From your perspective what are the… Every tool, whether you see Twitter, social media, is so divisive these days. People talk about net, has it been better for the world or worse for the world? Even somebody as tech forward as Elon Musk has talked about the pitfalls of AI publicly and he’s scared about it. In terms of generative AI, the world that we are talking about today, what are your sense of what the pitfalls could be? Disinformation obviously becomes so much more because you could create so much more.

Mattin: Yeah, that’s one big one. We’re talking about a tool, as I said, that hugely amplifies human creativity that will allow us to output much more content, hugely greater scale and much faster than we could before. And very simply, that means the bad people out there who want to generate bad toxic content will be empowered by these tools too. And there are things we can do and OpenAI and others, and Stable Diffusion are putting guardrails around these tools that prevent the creation of obviously toxic content. But those guardrails are imperfect. So there’s bad actors and the bad things they want to do. That’s a clear obvious problem. A more subtle nuanced problem, though one that rightly still gets talked about a lot is look, these tools are trained, as I said, on huge datasets, huge amounts of text, or huge datasets of images. And that means they import the biases that we humans historically have instantiated in that data.

Sadly, we humans are biased, prejudiced creatures and historically we always have been. So those biases are present across our culture and in subtle ways that we perhaps don’t, might not even always notice, generative AI will reproduce those biases. So there’s the danger that we get locked in a bubble of our own prejudices and biases and subtle points of view in ways that are not productive and not healthy and not fair, not just or equitable. That is a clear… And this is being talked about a whole lot. I can’t remember which tool it was, perhaps it’s ChatGPT. I have to be careful about that. Someone wrote something along the lines of a prompt of write me some code that tells me what kind of person makes a good coder? And the AI spat out, well, if they are white, if they are male, then answer is yes. There’s a clear example of it reproducing a bias.

Ali: Bias.

Mattin: Yeah, I am always keen to point out, because I think we can become confused about this problem. Some people talk about this problem as though these tools are imposing biases and prejudices on us. For sure they are. They may amplify them or they will provide new expressions of them, but we mustn’t forget that they’re only reflecting back to us the prejudices that we have expressed historically in the past, they’re really a mirror in that way of ourselves. So we shouldn’t let ourselves off the hook. It’s not these machines that are inventing these prejudices and biases. We invented them. That is nevertheless a problem that we need to address. And it’s not going to be easy, it’s not going to be simple. But that’s a problem that you and I and everyone else pretty much is extremely familiar with. We’ve had 30 years of this online space and we’re still trying to catch up to, we’re still trying to regulate social media, let alone this, it’s very difficult.

Ali: This is fascinating. Let me ask you one quick last question. So there’s been talk back and forth and people have already debunked it in the last few days about how this hurts Google from a search perspective, at least this ChatGPT tool, not the other tools that we talked about, but the ChatGPT, and what’s your sense of what this does to how people search?

Mattin: Yeah, it’s a really interesting question and it’s a complex question. And I think it clearly has huge implications for search. I don’t think they’re as simple as some of the first thoughts we saw this week as, oh ChatGPT destroys Google. This is the new Google. No, because if you sit and think through what a person is trying to achieve when they search, there’s a range of different things. And very often when a person searches, what they are looking for is a fixed defined object output created by another person. They are looking, for example, for an article that X wrote, their favorite journalist wrote seven years ago about the new BMW X3 or whatever. They’re looking for that. They don’t want some novel output created by a machine that is useful, they want exactly that.

So the Google version of search produces that, and Google, yeah, leverages AI, other kinds of AI to be incredibly good at going and finding that one particular thing for you and bringing it back to you. It’s incredibly good at that. That’s what its business is built on obviously. Other times though, you are searching for, okay, I need a patch for this code I’ve written, I know there must be some solution to this. I just need to find it somewhere. It’s not a defined object I’m looking for, it’s an idea, it’s a solution, it’s something less tangible. Then ChatGPT is a hugely powerful solution because it can generate that solution for you. It can just present you with that solution and that’s the job done.

So I think we need to think carefully about the different types of, I guess I’m using the language, of jobs to be done, the different type of jobs to be done when people are searching, and what does what. And there’s also work to be done around, okay, if you are using ChatGPT as a kind of search, you then do start to become interested in where this information is drawn from. So how is it going to prevent, present, sorry, its sources to you? Even if it’s presenting you with what looks like a compelling answer, how does it present you with context that persuade you, yes, this is the right answer, I can trust this? Okay. So there’s all kinds of questions to be answered, but there’s no doubt that implications for search are huge. Google know that. They’ll be looking at this, they’re not just looking at it closely, they’ve built Lambda, they’ve built their own large language model. They’re all over this. They cannot ignore this.

Ali: Yeah. For all we know, even though that Google has already talked to OpenAI about acquiring them, but then regulatory issues probably prevent this kind of deal from happening. And I’m guessing the founders of AI know they have something big on their hands, so why should they sell? But all of it is hugely fascinating. Well, thank you, David. It’s so early that it probably makes sense for us to get you back again when things are more concrete or at least, not more concrete, we have more examples of use cases by next year probably sometime as well, just because so many things are changing as well.

Mattin: I would love that. Yeah. Thank you for having me on. And the space is evolving so fast that if we wait six months, there’ll be a whole new set of innovations and ideas and problems to discuss.

Ali: Well, for those of you who don’t read your newsletter, can you plug your newsletter on the site again, please?

Mattin: I certainly can. Thank you. It’s called New World Same Humans, and you can find it at newworldsamehumans.com and see what other readers say about it there too.

Ali: Okay. All right. I’m going to put a link in the text accompanying this podcast as well. All right.