AI in Architecture: Leveling the Playing Field with Peter Williams

Speaker A: There's positives and negatives of that because you can end up with a client who's just, you know, on mid journey all day sending you thousands of images and some of them may not be realistic to all the other complex constraints we spoke about.

Speaker B: So, right, you're going like an oblong in the sky hanging off a giant.

Speaker A: Yeah. And you're like the floats, you can't have that. Sorry, mate.

Speaker B: Hi and welcome to it's not the End of the World Everyday Use Cases for. My name is Bobby McClusich and I'm the head of AI AI integration here at Quite frankly Productions in New York City. I'm also the creative director and a former teacher. I've always had a bit of an interest in new technologies, which is why this podcast exists. We're talking to people from across industries about how they're using AI in their day to day workflows and lives. Today I have an architect on the show. His name's Peter Williams and he's based in Vancouver. Peter, thank you for joining me on the show.

Speaker A: Hey, thanks Bobby. Good to be here.

Speaker B: All right, could you give me, um, a bit of a context about who you are and how you got to be where you are?

Speaker A: Yep. Um, so yeah, as you mentioned, um, I'm an architect. I'm qualified, um, in the UK and then moved over to North America, um, made the leap across the pond about seven years ago. So, um, there's a bit of a, kind of a transition when you move over kind of slightly different systems, but. But, uh, yes, I've been settled here for seven years and I work for a large firm in North America and then I do some kind of freelance side projects as well. Architecture related.

Speaker B: And architecture is, um, one of those aspirational jobs that we all hear about. A few kids in the class put it down as their dream job when they're little. Um, but when I think about it, I even know what the day to day looks like for an architect. So maybe you can help us understand what your day to day job looks like.

Speaker A: Yeah, sure. Um, yeah, I mean a lot of people bring that up actually, that they're not that sure what you actually do. Um, essentially it's ideas, it's creating spaces and it's resolving those ideas to something that can be formed into reality. So there's lots of separate zones that you can specialize into. So I guess in terms of the topic of AI, there's many niches and areas that can evolve and changes and it's very specific. But, um, it depends on the general

Speaker B: part, yeah, I can imagine, I can imagine. Uh, there are lots of areas that AI could weave its magic in your profession. Um, but still just to kind of like you sort of gave me a bit, still gave me some fairly sort of nebulous ideas there, you know, like I. Could we turn ideas into reality. Yeah, but like, what actually do you do? Maybe you just tell me like what, what did your day look to look like today? What did you do today?

Speaker A: I mean, today it depends what stage in a project. So an early stage, you're effectively trying to understand the client's needs in terms of a space or a building or a place. Um, and then you're, you're turning that, that brief into a, you know, a story or an image or something that you can sell and talk about. And then later throughout a project, it's eventually developing that design with, with um, in terms of fabrication and trades and how you might actually build that on, on a particular site or it may, it may not be a building. It could be something more abstract than that. So I guess for me, in terms of today I've been. We're on, we're on CA on a project which is, it means there's a hole in the ground and we're working with the trades on how to actually build the project. So we're reviewing drawings and kind of trying to understand the design three dimensionally to ensure that we're not going to have certain problems when we're on site and then it could be harder to resolve those issues. So it's quite different between earlier stage and later stage in terms of what you're doing as a day to day on the job.

Speaker B: Got it. How many, how many people are on your team doing that at the moment?

Speaker A: Directly? There's just three or three or four of us right now in our office. Um, but in terms of the wider team, there's the contractor and there's the client and a lot of other people involved. So it's, it's, it's a, it's a team, you know, a team.

Speaker B: Okay.

Speaker A: Job usually.

Speaker B: So where you are right now, there's uh, three or four of you. You've got a hole in the ground, you're reviewing drawings, you're reviewing the site and you're trying to make sure. And also trying to turn this into a 3D model that you uh, can then conceptualize a little bit more clearly to get the idea across to prospective stakeholders.

Speaker A: I mean, yeah, typically you'd usually always have a 3D or be working in three dimensionally, it could be model making, it could be ah, a digital model. Um, and then, you know, there's lots of other, I guess, developments in the profession in the last 10 to 20 years in terms of we have like a digital twin which is like a, or it can be called a digital twin, which is a separate model which can be a 3D virtual representation of an actual building. So you usually have models, but they have different functions throughout the process of building.

Speaker B: Okay. Usually models, they have different functions throughout the process, including twins and other types of models.

Speaker A: Yeah. So like, I mean at early stages you might just want to make a quick model to understand a problem or a certain aspect of, of the design. And then later on it becomes more a, uh, store of information of data, uh, that can be useful for more and more stakeholders outside of just your office. So there's a whole topic of model sharing and information which is usually uh, termed as bim building information modeling in our industry. Um, and then so there's kind of current crossover with AI and those topics at the moment.

Speaker B: Okay, all right, we'll come to that. But uh, I'm still very interested in just understanding your job. So it must be a broad spectrum of, I'm imagining you must be involved. Okay. So obviously the layman's idea is like, okay, he's building models and designing them in whatever software he uses and maybe he's also actually physically building the models. But I imagine there must be this crossover into all sorts of other areas like legal, um, engineering, um, like compliance of the environment and the city. Is that all true? Yeah.

Speaker A: Yes. Yeah. So the other aspects ah, are legal, um, compliance, um, there's regulatory frameworks. So you have to make sure the building or the design is going to be able to meet standards and codes, um, environmental expectations, which is, you know, progressed in a lot the last 2010, 20 years. Um, so effectively it's combining a lot of different requirements and kind of prioritizing and finding the solution that's best for the specific project. And brief. I mean, uh, an architect must be too abstract still.

Speaker B: No, no, not at all. Actually. It's very specific I think and it's just striking me just how broad the skill set required to be an architect is. I'm thinking about my own job and industry. You know, we can have like really skilled artists and ah, they can focus on doing, doing the animation or they can focus on, you know, we have a really skilled editor and all they need to focus on is building out the edit timeline. They don't need to really worry about, I don't know, the legal ramifications of what they're doing. Um, but it sounds like to me, is every architect just, uh, like a jack of all trades?

Speaker A: Well, that's. Yeah. I mean, as a profession, it's generalist. It's having an understanding of a broad spectrum of topics. You don't have to know the answer to everything, but you need to know where to look or who to ask. So that's the, uh, generic role. And then people gradually regularly specialize more into certain topics. So it could be digital fabrication, it could be environmental outside of, or it could be the legal. It could be contracts. So people tend to specialize. And then there's also just more generalist architects.

Speaker B: What's your specialty?

Speaker A: Uh, I've stayed pretty general, pretty broad. Uh, I've done a lot of different project types and, um, different stages. So I've kept myself very general, I guess is the way to describe it. Um, but other friends and colleagues, people tend to specialize in particular areas. Environmental topics, contract law, that kind of stuff. Um, but, uh, it depends on the individual, really.

Speaker B: Got it. And that can help you be a director, I suppose, and direct an entire project?

Speaker A: Yeah, yeah. It helps, I think, um, to have a good understanding of all the various stages and not being, not, not going too specific or niche early, uh, on in your career, I think is something that can, can be helpful.

Speaker B: Okay, well, that must also give you a broad overview of how AI is impacting all of these various different silos within your profession. So if you like that segue, uh, this is your opportunity now to tell me a little bit about how AI is showing up in your work.

Speaker A: I, um, mean, it's kind of. I guess it's a very broad topic, just like the, uh, profession. So, um, I guess you're most interested to know in terms of day to day, how. How I'm using it or.

Speaker B: Yeah, I am. Um, um, you know, we can go anywhere we like with this. And I respect your observation, you know, yes, it is super broad. And I could imagine, much like in medicine, there are, uh, specific, you know, machine learning algorithms applied to solving protein folding. Um, but then there's also doctors that are using large language models to transcribe the notes of their patients. We could go in all sorts of different directions. So I guess just begin with whatever just naturally occurs to you. I'm struck by when we met, I remember, and asked you to be on the show. The first thing you said was, oh, yeah, I used it, and it helped me do a project. I would Never have been able to do without. Without it. So, you know, maybe start there. That was the thing you were kind of inspired to tell me about over dinner. So, uh, let's.

Speaker A: I think, I think it gives the opportunity for smaller offices to have a wider range of resources at the disposal. Uh, it allows access and to knowledge a lot quicker I guess than before. So previously maybe you'd have to research or have a much bigger team that could advise on a project. But if you, if you're able to trust the information you're receiving, then the. There's, there's a huge potential for growth of smaller offices and smaller individuals to be able to achieve a lot more due to the access to information that it provides. Quick access to information.

Speaker B: Interesting. Okay, give me a bit more of a specific there. Like are you talking about using a large language model?

Speaker A: Yeah, I get, I guess, large language model, I guess. Uh, the example would be in terms of regulatory frameworks, you're able to search a lot quicker in terms of understanding the requirements that you may need to meet for a building in a certain location. So you can use it like an assistant to kind of come back to your desk to give you information on what the requirements are, uh, which will allow you to design a lot quicker. So that's something that I found when, when I was using more AI, AI on a project recently myself.

Speaker B: Give me, give me a bit more detail on that. So you were working on a project and m. You didn't have the resources of a bigger uh, team and so you, you were able to find the information you needed to help you progress?

Speaker A: Yeah, yeah. So I worked on a, uh, a project in, in Ireland which is under a different framework of regulation. So there's, you know, there's a degree of risk and a level of understanding. You need to kind of put time into understanding the environment and the constraints in terms of various aspects. So not just regulate regulatory, but the environment and understanding the site. And so the time that it takes to do that typically would be a lot longer, but it was able, I was able to use large language model to give me a lot of information and give me a lot of input on, on specific knowledge that I don't have already, you know, me personally. So it reduced the amount of research I needed to do on a project and allow me to work on a project in a different jurisdiction.

Speaker B: Interesting. Can I ask you a question then? Um, just off mic, actually you were just talking about hallucinations.

Speaker A: Oh, uh, yeah.

Speaker B: And I wonder, like, how do you deal with the fact that it's not always 100% reliable. If it's. You're researching an area that, you know, presume with architecture, you can't, you can't afford a mistake.

Speaker A: Yeah, it's, um. I mean, that's. I guess that's the aspect that was the drawback of this process. When I decided to test out how much I could. It could help me on a personal project that I was working on. And so I actually ended up having to spend a lot of extra time to. To check the information it was giving me manually, the normal way that I would do things. But I could at least see the advantages as it improves, as the language model improves, the advantages it could give me in terms of time. So I did save a lot of time that I maybe wouldn't have been able to take on the project. But it wasn't as efficient as I think it could be as things develop or the reliability or trust develops. Uh, right, interesting.

Speaker B: So where it currently is, but because it's not reliable, you still have to spend a lot of time doing the legwork, checking its work. But I suppose in a certain type of way, it's kind of like. So we had. I've had a lawyer on the podcast, and it's the same, right? They have to be super buttoned up and. But, uh, they save a crazy amount of time because the AI is acting like a junior associate and doing a whole bunch of research. And then they check the research.

Speaker A: Exactly.

Speaker B: But it's quicker to check the research than it is to do the research from scratch. Is that essentially the same thing that you're saying?

Speaker A: Yes, it's very much the same thing. Like, it felt like, you know, you're using it like an assistant who has a certain level of knowledge, and then you have an overbearing kind of understanding of what you need to look out for in their work, essentially. So I think it is important who is using the model, because if you don't have the backbone, the background knowledge on what to check, then things can slip through and those mistakes can, you know, follow through. And there's legal ramifications of it.

Speaker B: So, yeah, that's interesting, isn't it? I mean, it's like it's such a truism. The, uh, you know, I think we're definitely beyond phase one or even phase two of, like, AI becoming integrated in people's lives. But initially, you know, in that early, uh, phases, you may remember people afraid for their jobs, obviously, but also kind of skeptical about it for some reason. But those of us that use it Realize that, uh, you have to have, as you say, a backbone of knowledge, otherwise it's kind of useless. And you certainly can't just copy and paste an answer and take it as verbatim and just use it. You need to exercise, uh, your own personal, professional judgment. And I'm starting to see now, interestingly, I'm starting to see professionals using AI and there's this kind of, I suppose, a relationship between me and the professional that we both know that they're using it. But that's okay, right?

Speaker A: Yeah, yeah.

Speaker B: And that's come up for me in two healthcare situations, both with my child. Um, I've got a newborn son. He's like seven months old. And our pediatrician was talking us through. He'd recorded all of our calls over the course of six months. And he had used ChatGPT, I think, to summarize our, uh, conversations. But then rather than like, pretending that it was his, like his, um, what you want to call it, his insights, he was reading out. He'd prepared a presentation, but was reading out ChatGPT's sort of summary.

Speaker A: Yeah, okay. Yeah.

Speaker B: And it was kind of like, I knew it was chatgpt, but it was okay because it was being validated by my pediatrician, by a, uh, professional. And so, yeah, and he was being transparent. And it wasn't like he was hiding it. He was being super transparent. M. It made me think, okay, I think this is maybe the next sort of. That's how, you know, I don't know, said something to me about transparency is

Speaker A: kind of vouching for it or being. Being open, that this is information, this is a tool that we're using, and this is what it's given us. And as the expert, I'm. I can analyze and simplify or adj it for you. I think that's kind of the direction that you can kind of see it going in, in a lot of industries. Um, because, you know, you give. You give a degree of credibility right, from your knowledge and being able to analyze if it's made mistakes or things that you would disagree with, because it's not necessarily always correct, but you have the expertise to. To advise on. On the output it gives, I think.

Speaker B: Yeah, exactly. And then. And then I guess I kind of hope that as, uh, you and I both know, you know, you the professional, me the customer, both know that this is what we're working with, and we both understand how it works. We can both be able to keep an eye out for hallucinations or weird oddities and work together to Then get to the, to the end goal.

Speaker A: Yeah, I think, I mean, one, One area where it's, it's definitely. It's more. The client is more. Has access to more tools that can, you know, Mid Journey can produce these images and quickly share them and say, this is, you know, this is what I want, or I like this. So there's, there's a different way of being able to communicate because previously, let's say you've got a client who wants something, they can describe it in words, they can find a picture from a magazine and say, I want something like this, but maybe change it this way. Now, you know, now there's the ability to have more control over the images that they can share in terms of their desires and what they want from a project. So that's, that's one key difference which I think has its benefits. And, you know, there's positives and negatives of that because you can end up with a client who's just, you know, on Mid Journey all day sending you thousands of images and some of them may not be realistic to all the other complex constraints we spoke about.

Speaker B: So, right, you're getting like an oblong in the sky hanging off a giant.

Speaker A: Yeah, floats. You can't have that. Sorry, mate.

Speaker B: That's interesting though. So you're getting, you're getting clients who are now using generative AI to generate images.

Speaker A: I'm not, I'm not specifically, but I know, you know, conversations with people and it's, it's more available. They're aware of what you can do with it. So I think that's a bit of the challenges is how, how that relates to our job and how we communicate with, with other consultants and clients is what's changing? So interesting.

Speaker B: Ah, are you. Do you take any inspiration from any of those tools? Is that part of your workflow?

Speaker A: Yeah, yeah, I have. I've experimented with that. So it's useful for you can, you know, do quick image generation. A, um, lot of it depends on your control. If you've got control of it, which is obviously learning, learning the tool. Um, I mean, our industry has had various tools come along. They're probably the same as yours and so we're used to learning them. I think as long as you treat as a tool, then it can help serve you. But the moment it's taking control and it's designing things for you is when. Is when you've got a kind of question, what are you trying to make or achieve or build? So, yeah, it's a long topic.

Speaker B: Right, which which tools are you experimenting with? I'm curious. Or have you experimented with any that have been the good, the good and the bad?

Speaker A: Yeah, I mean I'm testing a few out I think. Um, I've ah, used Mid Journey a bit and then another kind of specific architectural software where you upload sketches and it can create a 3D rendered image. Um, it's useful for, for rendering in particular. So that's a process that could take a lot of time or used to 10, 15 years ago, take a lot of time and money to do and now it's able to speed that up tenfold.

Speaker B: Uh, is that an AI powered tool?

Speaker A: Um, yeah, there's a few various websites where you can upload sketches of buildings and have a bit more specific control over things.

Speaker B: Interesting. Any, any names that you'd say were worth checking out if our listeners are particularly interested in diving deep into the architectural design world?

Speaker A: Uh, yeah, Website I've been using specifically is lookx. Uh, AI. Lookx AI? Yeah.

Speaker B: Okay.

Speaker A: So you can do sketches and upload them and it will create an image. There's a, there's a lot out there so I, I, I'm not vouching to say that's the best one out there, but it's one that I was experimenting with quite early on um, and then actually used that website actually to enter a competition with a friend of mine and we came runners. Runners up or sorry, joint, Joint winners. Sorry, Joint winners. Joint winners.

Speaker B: You consider yourself a runners up? A runner up though?

Speaker A: Yeah, I think there's, you know, there's two of us for some reason. I think it's runners up.

Speaker B: Uh, unless you're first. Uh, yeah, yeah, yeah, that's cool, that's interesting. And okay question, this is a couple of years ago.

Speaker A: Yeah.

Speaker B: Okay. Is there any like, like so called cheating because you used look AI or is there like, you know, would your competitors have felt that that was dishonest? And I don't try to catch you out. I'm just interested in what the, I'm interested in what the landscape looks like in your, in your business.

Speaker A: I know, yeah, I think, I think at the time when we, when I did it, uh, this was a couple of years ago, it hadn't, we weren't really sure what people thought of using it. There wasn't any, it wasn't as commonly used. I think I feel like the industry is more familiar with using it now, but at that time on that project we were unsure what people would think of us using AI on it.

Speaker B: Would you?

Speaker A: It was a Bit of a test. No, we didn't hide it. We just submitted our design, and it wasn't fully AI. We did some stuff, the usual way of hand sketching and making our own model. And then there was some stuff where we made the images by putting in, um, projects that me and my friend had both designed ourselves. We inputted those two images of two buildings that we both done and created, like, an amalgamation, and then that became the basis of our submission. So.

Speaker B: Cool. Okay. I mean, that sounds like exactly how people are using it today. I feel like two years ago, that might be the sort of thing that people that didn't understand the process would have been like, oh, uh, that's cheating. It's. Whereas now it sounds to me like that's very clearly what AI is. Right. It's a tool that you have to put a good. For want of a better word, a prompt. You have created a prompt, right? Yeah. They say garbage in, garbage out, but you've put good quality stuff into it, and you've got a good quality result out of it. That made you joint winner.

Speaker A: Yeah. Not Soul winner, but not soul winner. So obviously there was something wrong with that. Um, but I mean, the. The. The other joint winner, uh, his. His submission was very, very different approach. It was. I think it was a lot of written. It was very. It was just very different presentation. And he obviously came up with some really interesting ideas that the judges were also interested in, and ours just was more image and idea based that we presented well, because. Because we use these tools, and I guess they both like the two aspects, or maybe they. Maybe they recognized that we were using AI and they. That's why it was a joint one. I don't know the backstory.

Speaker B: Interesting. I, um, I'm. I'm conscientious that we've got 20 minutes left, but I'm really curious, like, if you do, would you be able to. Do you have that to hand? Would you be able to screen share and show me. Show me it?

Speaker A: Uh, yeah, I do, actually.

Speaker B: Amazing. All right. Um, this is. Will be the first screen share by a guest on the podcast. So, Pete, you have that distinct honor. Um, and for our listeners who are only listening, I will do an incredible job of articulating what I'm seeing.

Speaker A: Okay, well, I have to figure out how to screen share.

Speaker B: Uh, there should be a button at the bottom. There's a share button on my side. I don't know if you've got it on yours. All right, here we go. This is new.

Speaker A: There we go.

Speaker B: This is New bloody territory that's being explored. All right, so I'm seeing what looks like a lot of. To my eye, they kind of look mid journey, like. And then I'm also seeing some sketches. Um, so what part of this.

Speaker A: Uh-huh. How.

Speaker B: Where did the AI come into it?

Speaker A: Uh, so, yeah, a lot of these are mid journeys. So the tower kind of image, just then these images. That was where we combined our two projects, um, to create the images for the project. And then there's the kind of master plan scale image that was done by us inputting a 3D model. So this one here.

Speaker B: So this 3D model that I'm looking at, just hover on that for a second. That is not AI.

Speaker A: That is a digital model we made by hand, by computer, by software, extruding blocks and making streets.

Speaker B: Okay, so that's an old school method of doing.

Speaker A: Exactly. And then that. That was then, um, we then used that as the basis and, um, uploaded it effectively for kind of the presentation, for the. For the rendering. I mean, this is a. This is not perfect. We couldn't get. We didn't have the creative control that we wanted, but it was a good enough starting point to get. To get put forward our ideas.

Speaker B: Got it. Okay, then. Scroll, scroll. Give me another one. Show me something else that is not AI.

Speaker A: So this first, uh, sketch is first

Speaker B: again, I'm just give the audio. So we were looking at a 3D model that looked like it had been done in some sort of rendering engine that was traditional. Now we're looking at some bird's eye view sketches that are very much hand drawn with handwritten notes.

Speaker A: Yeah, these are all. Well, actually, these were done by, um, Jamie. Jamie Eden. And these were done on his iPad, so he likes to work that way. So these. We didn't use any AI on this, but it's actually a slightly different workflow.

Speaker B: Okay, so we've got a concept sketch, and I. I can see a bird's eye view of what I was looking at in 3D. And so now. So. So did you put all of this information into your AI engine?

Speaker A: Uh, effectively. Not. Not directly. All of these. We actually didn't need to upload all of these. Um, the one on the. On the far right hand side, that is an outline of that 3D image that you saw earlier. Uh, so it's from the same.

Speaker B: You know what, hang tight on this. I just. While you're doing this, I'm gonna. I'd like to see how Nano Banana handles this. So while you're, um, which is do you know Nano Banana?

Speaker A: Yeah, I've been, I've been messing around with that recently.

Speaker B: Okay. And do you get any, um, have you had any good results?

Speaker A: Actually, no. No, I could hit a bit of a dead end on it.

Speaker B: I'm wondering whether it could do what I've heard that it's got some architectural um, chops, so to speak. And I'm wondering if I put in that bird's eye view, um, and put in the 3D model, whether it would be able to turn into. So go to the next slide, the 3D model. I wonder whether it can turn it into, um, the render. So you took the blueprints, uh, sorry, the top down drawings, the hands mission drawn sketches, you took the 3D model that you built and then it created this image that I'm looking at now, which is like a beautiful mid journey, like image. What do you call that in terms of vocabulary?

Speaker A: This is a rendering. So um, this is a rendering of more of the streetscape or a, you know, a finer grain of the project. I think the reason it worked or we were able to use it for this project was because we, it was more of a master plan. It was more about public realm and density and um, cityscapes. And we weren't designing the specifics of the building. This is just to put forward kind of an idea of what a space or place could be like. So it was kind of a different, a different type of competition to maybe if you're designing just a specific building, you then have a different approach of how you'd use AI.

Speaker B: Okay, interesting.

Speaker A: All right.

Speaker B: Um, and so, so you call it a, uh, a render. All right.

Speaker A: Yeah.

Speaker B: So you know, in the background. So if you stop sharing, I'll show you what I've just done and um, we'll see, we'll just see if it comes out. So I'm going to share my screen. Uh, let's go into Gemini. So I have just taken a screenshot of the top down view and um, I took a Screenshot of the 3D render. I said, can you turn these sketches and models into a render for my architectural friend who wants to explore the use cases for his profession?

Speaker A: Okay. Wow. There we go.

Speaker B: What do you think of that? Is that, I mean, it's better.

Speaker A: I mean it's improved in two years, isn't it?

Speaker B: I mean, you tell me what like I'm looking.

Speaker A: There's no lake. There's no lake over. There's nowhere river over that side. Uh, ah, the quality is good.

Speaker B: It's added a lake um, is it accurate to what you're. It looks similar to the design. From what.

Speaker A: It's similar. Yeah, it's similar in terms of the massing. And I mean, for this entry, it wasn't necessarily about the specifics and the details. It was about creating courtyard spaces and certain massing and blocking which this image shows. So we were able to use it a couple of years ago because it, we, we were okay with, with allowing it have a little bit of creative freedom. Which I think some people, some architects would disagree with that and think that's blasphemy. But, uh, it, it served the purpose of us being able to quickly enter a competition.

Speaker B: Right.

Speaker A: Ideas that we had.

Speaker B: More like a proof of concept, right?

Speaker A: Yeah, exactly. Yeah.

Speaker B: Which is very much what this is all about, isn't it? Um, I mean, we could carry on going down this thread, but I'm conscious of your time. I'm very interested in other things we could do with this. I've heard that it can do very good. Like line. Is it called line drawings or like line.

Speaker A: Yeah, line work. Yeah, line work.

Speaker B: I don't know if that's a thing or.

Speaker A: Yeah, like sketches you can do. I mean we all, before, you know, AI was a buzzword. We could always create sketches from 3D models. There's you know, filters or export settings that you can do. Um, so stylizing images, it's, you know, it's improved a lot. But we've always had tools to do right things.

Speaker B: It's funny, isn't it? You know, people just want to use

Speaker A: AI for everything and it's quite a buzzword, isn't it?

Speaker B: Yeah, it is. And there's a lot of snake oil around as a result. Yeah, but the proof of concept bit is real. And you, you said that. And M, you know, I'm starting to work on, you know, vibe coding. Um, and, and M, you know, you're starting to realize, oh, I can build an app. And then it's. Sometimes it's just as a proof of concept. It's, you know, it's 90 of the way there in fact.

Speaker A: Yeah, exactly.

Speaker B: I've built apps at uh, work that we use now because M, because it's good enough for work. It might not be good enough to like put out into the world as a consumer facing product, but I have no doubt that if I wanted to go and seek investment and say, hey, I think I've got a functioning product here that we could sell, then I can create. Much like what you said about how some clients are coming with mid Journey images. I can now come with a fully functional app that I built in my AI and then, um, get some professionals to actually turn it into a viable product.

Speaker A: Yeah, it kind of makes the regular individual more, there's more opportunity to share and present their ideas. Because if you go back to, I mean, if we go back to architects, traditionally the way that people would communicate would be drawing by doing a sketch. And then, you know, that's seen as a good way to communicate because it's very quick and direct. But what I think, and we're talking about AI, we, we're getting better at communicating with computers and so we're able to share ideas in a different way than just sketching. So.

Speaker B: Yeah. And it also brings me back to what you said at the beginning of this conversation, which sort of struck a nerve. Maybe not struck a nerve is not the right word, but it triggered a kind of, um, a something I've been percolating on, which I'm noticing in a trend I guess I'm m seeing in all of these conversations, which is that it's empowering individuals and smaller, uh, organizations, those that don't have the resources to compete with much bigger organizations to do more.

Speaker A: Definitely.

Speaker B: I think in your case it sounds like, uh, I'm an architect that has an idea, but I don't have the resources to be able to research or do whatever I need to do to implement it. Well, now I kind of do and I'm experiencing that myself when it comes to, you know, all sorts of this podcast, like, like I'm using AI to do basically everything aside from the actual conversations.

Speaker A: Yeah. Okay.

Speaker B: Um, honestly, I wouldn't be able to do it without it, you know.

Speaker A: Yeah. Um, it reduces the, the time needed on topics. Right. And then the time is money and then it's often the big companies or big firms have more of that available. So it's, it's kind of a short circuiting, kind of the traditional route on some tasks, I guess.

Speaker B: Yeah. Lowering the barrier to entry even in the playing field.

Speaker A: Yeah. But then on the, on the, on the flip side there's, I guess on the flip side is in larger organizations or even any organization, there's, there's, it's potentially there'll be less entry level roles because some of the, you know, regular day to day tasks are less, less in demand. And um, you know, that that wide triangle of an office starts to kind of change shape a little bit.

Speaker B: Okay. Yeah, well, our lawyer, you know, guest spoke to that. Exactly. And that profession's really like in, in Danger of there not being this kind of base of junior executives able to rise up through the ranks via experience. Is that something that is happening in your line of work as well? Is that why you speak to that?

Speaker A: Uh, it's not something. I wouldn't say I'm specifically uh, seeing it right now. But in theory it makes sense that the more it develops, the less straightforward, simpler roles, which is how you kind of get into an office at uh, first, there may be less, but that's not something that's AI. AI Specific because we've had the development of various technologies over the last 20 years that have changed the role that people have in offices. So they've become more, more bim. Tech. Tech based rather than um, I mean it used to be hand drawing and it's just changed to be computer based now, so.

Speaker B: Right, right, okay, interesting. Well look, the fans, huge fan base we have for this podcast and uh, the one of the part that they love is the quick tips part podcast. So this is a section in which I just want to get your practical, um, day to day experiences and or advice on how to use these tools. So without further ado, let's start with a general piece of advice, um, that you would give to a friend, a colleague, loved one, enemy.

Speaker A: Okay. Ah, yeah, I guess, I guess. But tip wise I would say that if you're, if you've got a problem or you're working through something, then it's good to always think, can I, can I use this tool to make my day easier or have a better solution? I think that's something that's useful. Uh, because sometimes it's unexpected where you know, you're just cooking something and you can take a photo and you can say, what do I do with this? Or how do I. It's just basically like having someone who's able to advise you on something that you're less knowledgeable on. So I always try and just think of something if I'm trying to solve a problem. Okay, can it help me or not?

Speaker B: Okay, so if you ever get stuck, just open up AI and see if it's got a. If it's got a solution. All right. Nice.

Speaker A: Yeah, I mean I used, I used to call my dad when my car broke down. Now I just take a photo of the engine, it'll give me the answer.

Speaker B: Oh, uh, Pete, that's uh, the uh, that's.

Speaker A: Isn't it?

Speaker B: The Doomers and gloomers are saying. Yeah, no, you're not calling your dad as much.

Speaker A: No, no, I'm joking. I Just, I do still call him just for the sake of it. Right.

Speaker B: You don't need to talk about the engines anymore.

Speaker A: Now you can talk about. Yeah, less engine chat.

Speaker B: Yeah, now you can talk about your feelings. Finally.

Speaker A: Yeah, finally.

Speaker B: Um, okay, that's nice. That's a nice general tip. All right. What about something more specific? Um, sometimes I call it a personal hack. Sometimes I've been known to call it your own personal superpower. What do you, uh, yeah, do you have any kind of, like, more specific Peter centric. Peter Williams centric tips or hacks?

Speaker A: This is, this, is, is this AI based or is this just general.

Speaker B: No, still AI. Still AI based. The, uh, kind of. There's another podcast I have which is to do with your personal therapy. Like in terms of, again, AI related. Like, you know, I like. Okay, so let's say for me personally, um, one superpower I have is taking screenshots, right? Like, uh, I don't know what is on PC, but on Mac Control Command Shift 4 and you can drag over the screen and then press Command V and you can paste it into the chatbot. And then it, you know, because it can see. I don't need to like copy and paste emails, I don't need to copy and paste images. Like, I could just copy and paste wallets I have on the screen, put it into the chatbot, and then it's able to give me feedback on whatever it is. And I don't need to explain, I don't need to give any context. Can just be like, hey, what do you think of my reply to this email? You know? Yeah, yeah. Or what do you think of my design on this flyer? You know?

Speaker A: Yeah.

Speaker B: Um, so that's good. That's my own personal superpower. I think it's like, it's, it's an op superpower. What about you?

Speaker A: Um, I mean, I, I guess my useful tip is I, I use it in terms of architecture. I use it for, um, checking regulations. Basically, it's. This isn't the most interesting topic, but.

Speaker B: No, but gone. People have to do this one.

Speaker A: I mean, you know, usually you'd go find the regulations and you'd go through and you'd find that specific section and at certain points, you know, you really need to still do that. But there's certain, um, quick stuff that you can actually get a kind of a good steer on regulations by just asking it to look it up and to advise on a certain topic. So, um, but you still have to do the safety check. Uh, but it's a good hack.

Speaker B: Junior Junior researcher vibes. All right. And we are coming up to the end of the time. So finally, finally, a fun or unexpected use case that you have found you

Speaker A: using AI for fun or unexpected use case. Um, uh, I mean, it's actually not. It's probably not that fun to everyone, but I found it quite useful for investing, actually.

Speaker B: Okay.

Speaker A: All right.

Speaker B: Interesting, interesting. It's funny you should say that because there's a actual leaderboard going on at the moment where they have. Someone somewhere has got, has given each AI model $100,000 and they are. They've got like, uh, a leaderboard on which AI models are up, uh, and which ones are down.

Speaker A: And I think which one's the winner.

Speaker B: Claude at the moment is in the lead. I think it's 10% up.

Speaker A: Wow.

Speaker B: Compared to where it was, Um, I remember seeing that Gemini 2.5 was above Gemini 3 in the investment, but I think it's quite early days yet. So. Yeah, um, either way, it's an interesting little thing going on there. So go on. I find it using to do it

Speaker A: and tell me it's just ChatGPT I'm using, but I find it useful in terms of, um, rather than just saying, what should I do with my money, where should I invest it? Uh, there's a certain topic or something that you invest in. You can still try and analyze it yourself, understand where you think the market's going to go, and then use it as a kind of logical sensory check. One of the key things about investing is not to be too emotional. And obviously an AI language model is great and not being that emotional. So you can get very good, um, feedback on something that you're thinking of doing or investing in. So you can. I'll just upload screenshots and say, this is what I think is going to happen. Can you give me a bit of what are your thoughts on this? Act as an expert and then it gives a kind of second opinion on, on a decision you're making. Right.

Speaker B: And you find it. And when did you start doing that and how long you been doing that for?

Speaker A: Uh, probably. Probably the last year or two. The last year or two.

Speaker B: Okay. Um, are you doing it more and more? So that's quite a long time to have a good sense of whether or not this is useful. Um, how is your, how has your, like, usage evolved in that time? Like your strategies of using the, the tool to help you?

Speaker A: Um, I guess. Sorry, say the question again. How's.

Speaker B: Well, like I imagine when you first thought, okay, let's see if if ChatGPT can give me any good feedback on this, maybe you went in and thought, okay, I'm just going to ask it what to invest in. And then you were like, actually it's not very good at that, but how is your, how have your expectations evolved? And uh.

Speaker A: Yeah, yeah, sure, yeah. I guess I've become more specific on one certain thing that I'm investing in. And then it keeps hold of the knowledge of you've asked before and where you want it to search for information to give you advice. So rather than it just scouring the net and getting any old input, there's certain, uh, websites that I've said, you know, I want, I want you to read this website and give me input based on what their thoughts, uh, are not just a general, you know, scrape across the Internet. And then I'm more specific in terms of sharing screenshots and I'll sketch on the, on the graph lines and then say this is what I think is going to happen. What are your thoughts? Or is this, can you see this pattern? Because it's good at pattern recognition as well. So.

Speaker B: Okay, interesting. And then, uh, do you then. And you sort of already said this, but I want to reiterate it. You then don't take it as verbatim, you don't take it as a sage, but rather you use it as a way of just bouncing ideas off to, to, what did you call it, like

Speaker A: sensory check, Safe, safety check. Yeah. So, you know, if I'm thinking of, if I'm thinking going in on something, I can use it as a bit of a. Can you, you know, give me your advice on this? And then, and then I've, I've got a, a friend of mine that I chat to about this topic as well. And I'll, I'll share with him, um, the information he's giving me. I'll share with him what I think. And then he's, you know, not emotionally detached, detached to that decision. So he can give a kind of third opinion. Right.

Speaker B: So it's just like more heads.

Speaker A: Yeah, yeah.

Speaker B: Two heads are better than one. Two heads and a digital mind.

Speaker A: Um, exactly.

Speaker B: About two heads.

Speaker A: Yeah. Because he might see my, read it and say, like, you're putting too much trust in this, this feedback it's giving. But, so that's why I've, I've, I've asked a, uh, friend as well, instead of just myself talking to it.

Speaker B: I've not spoken to anyone about this particular topic, like financial investment. I will say that the last guy I had on yeah, he was using it um, to explore his own finances so um, like personal financial management rather than investment. But he said he was uploading all of his um, credit card statements and then he was getting it to analyze his trends, what he's spending money on,

Speaker A: what he's spending it on, and then kind of. Yeah, that's quite useful. Another one is, um, I've been, you know, considering buying a, buying a place and it's, it's quite interesting to use for analyzing property. I'm sure. I don't know if you've had a, a realtor or estate agent on.

Speaker B: No, actually I hadn't even considered that. But that's a really good um, Good.

Speaker A: Yeah.

Speaker B: Idea. Go on. So, yeah, how is that useful?

Speaker A: Yeah, I mean, you know you get, you get reports on buildings in terms of their, their maintenance and, and the condition of things and then there's market reports. Uh, so you can upload a lot of information on, on, you know, essentially, you know, it's either a home or investment, however you consider it. But it can then analyze it in terms of your needs quite objectively. So.

Speaker B: Interesting. So something I want to flag is like a. Watch out. I'm always like, it always come occurs to me when people are uploading lots of documents. So firstly like a top tip is um, the lawyer. Again, keep going to the lawyer. Both I guess.

Speaker A: Yeah. The similarities. Yeah.

Speaker B: So they obviously have to deal with like tons and tons of documents.

Speaker A: Yeah.

Speaker B: Right. And so the tool that they use is Notebook LM because it's very good at uh. This is a Google tool. Um, it's called Notebook lm. Um, and essentially it's like, it's designed for people that are studying. Uh, okay, most mostly. Um, but you can tell it, I just want you to focus on these five documents and it will only.

Speaker A: Okay.

Speaker B: It will only draw information from those five documents. Yeah. The reason I bring it up is because my experience with ChatGPT and of course new models come out all the time and I haven't experimented with the latest models, but my experience is that if you upload a few documents, it sometimes only looks at like the first few pages.

Speaker A: Yeah. You have to kind of prod it and remind it to actually check this. And yeah, uh, it does make a fair bit, a few mistakes when it's not fully incorporating all the information from all the document.

Speaker B: So a, it's either you have to like try and nudge it in the right direction, but I would just be really careful whenever going it to give me a full like analysis of Any document I put in ChatGPT first I'd probably do a few like, do a few checks like tests like what's on page 57 of this document or where is the word you know alakazam used in this document just to see whether it could like give me accurate feedback. But uh, at this point in my AI evolution I probably wouldn't even use touchability. I'd use Notebook LM because I know it's just the market leader and it's very kind of well reputed and um, used by global law firms as uh their go to tool for now.

Speaker A: Is it confidential then? Because I think, I mean that's a topic we haven't spoke about too much but you know a lot of offices are using kind of closed loop systems so that you're not sharing what you're doing live.

Speaker B: Yeah, that's a, that's a good question. I think depending on. So the lawyer I spoke to, they have a partnership with Google so Google's providing them with tools that they can use. And yeah, I haven't got into the weeds about how they're creating a closed loop. Yeah but I do know they have certain tools that their employees can use and then certain tools that they're supposed, they're not supposed to use for like sensitive data.

Speaker A: Yeah, that's very similar. Like I think you know a lot of the larger offices are using Copilot and various off offshoot software. That's closed, closed loops. And then I think smaller offices are I guess it's more open sourced information where they're sharing stuff because you can't. I think it's more difficult to have a closed loop system if you're not paying a significant amount. Is it?

Speaker B: Yeah, it gets really tricky doesn't it? I'm um, sort of one of the like you know you can't have sensitive data going into these models uh, uh, are uh, online because the models can use them to train themselves. That said, I mean look you know personally if you're like a sole trader uh, like these models, no one's going to fucking see this data because it's going into an ocean of other data that's being used to train the models. Like the odds of any of that sensitive information appearing anywhere, it's like a gazillion to one. But if you are you know uh, price warehouse Cooper, you know your whole reputation depends on being secure. So you can't do that.

Speaker A: Basically they're paying for the privacy of dealing with certain um, organizations so you can't just be sharing Things out with the uh, rest. And there's also, you know, there's also already different confidentiality agreements that are typically uh, signed and in place. So you can't just be sharing everything. So.

Speaker B: But there are other ways and means. So another way that one, if you are like a smaller organization, a way to create a. So I, a therapist in my family is trying to figure out if he can use upload his transcripts from his conversations. But part of the code is that it has to be private. Now my advice to him has been well you know you can, you can get open source models like Deep Seek, the Chinese model.

Speaker A: Yeah.

Speaker B: You can, you can host that locally in your organization or on your machine and it doesn't connect to the Internet. It's just completely self contained. And in that scenario that's just like having word on your computer. Uh, you know, anything you put up into it is completely, you know, is not going anywhere. So that would be my other piece of advice for anyone worried about.

Speaker A: But with, with deep. Deep seat. That's. Is that a uh, language model?

Speaker B: Yeah, D6 large language model. Yeah. It's developed in China. So online is 100. All of the various anxieties that uh, any Chinese software brings. Yeah but it's, it is open source and what open source means is that you can download it and um, you can own the entire thing yourself. And um, it can be self contained within whatever machine you put it on.

Speaker A: Yeah, that's cool. Yeah, I guess, I guess there is, it is available for a lot of smaller um, offices to, to do what you're saying then and have a closed kind of system. Um but it's just, I guess that would take a bit of time for various size offices to develop the systems.

Speaker B: Totally to figure it out and like is it worth it? And also everything's evolving so quickly. You know, by the end of 2026 you could have a. You know, your model might be just 10 times better than the current one you just downloaded and invested a ton of money in trying to make closed off.

Speaker A: So you know, I mean have you had many conversations about um, I mean more about. I mean is it uh, AGI, General intelligence?

Speaker B: Oh, AGI.

Speaker A: AGI. Because that's, that's, that's a whole, you know, this, we're talking about tilly winks here. Compared to of which now we're getting into the fun stuff now.

Speaker B: Uh, yeah, of course. Yeah. I mean it's a AGI. I mean, you know I was reading in the Economist like yesterday, um, you know Sam Altman Thinks that we achieved AGI like a year ago. And it went and it sailed on by you know, the, the leader of Anthropic and of um, some other kind of famous model they think is at the end of 2026. I think the leader of Anthropic and Elon Musk said that meanwhile the leader of Gemini Demis Hassabis, he says it's not going to be for another 10 years. So this term is like nebulous is sort of a bullshit term. Um, yeah. What does it mean? But if it means. And I think personally I don't think it will ever. I think it's just a marketing term that doesn't make any sense when you really think about what we are dealing with here, in my opinion. So let's start with just like the general. What do people mean when they mean AGI, the AI can do anything that a human can do. Okay, cool. And you're right, if an AI ever comes along that can do anything that a human can do, it'll be unbelievable. And they, they do believe that's within the next couple of years. But equally these things can already do some amazing things. Right. And, and the next step beyond AGI is asi. Have you ever heard of that term?

Speaker A: Um, uh, I think I did hear something about it.

Speaker B: AGI is artificial general intelligence. ASI is artificial super intelligence.

Speaker A: Right. And it's super in front of it. Just add a, add a super in.

Speaker B: Right. Because the idea is that once you get to AGI, it can then improve. It can do anything that a human can do, including code and it can then start working on it its own code and then it can, you know, it can be working on that 247 and explode. Then you would have what they call an intelligence explosion. And then we have artificial superintelligence.

Speaker A: However, I think sound a bit scary that uh.

Speaker B: Yeah, it's starting to get very existential.

Speaker A: Yeah, yeah.

Speaker B: However I think with my incredible computer science background and understanding of all of these things, but I think it's becoming increasingly obvious us to me at least that. Well first it's already a general intelligence, right. That's why I'm able to speak to architects, lawyers, teachers, healthcare professionals, you name it, electricians, you know, because anyone can use this tech. So it already has a general intelligence, it's generally applicable. And then, and then all of the things that it can do, anything that it can actually do, it can already do a, ah, far superior level to than a human. You know, when we're talking about, you know, give me The, I don't know the building regulations in Chinese, Japanese, Mongolian and Portuguese. I'll do it like that. I mean, it's like no human can do that.

Speaker A: Yeah, yeah.

Speaker B: You know, so it's got this. It's sort of in this weird space. I think they call it jagged. It's like a jagged in intelligence at the moment.

Speaker A: Right.

Speaker B: So it has this jagged. I mean, they say jagged general intelligence, but I always say it's a jagged super intelligence. In the areas that it excels in, it's shot off. It like, is extremely capable, but the areas it doesn't excel in, it's like way worse than a human.

Speaker A: Ah, okay. Yeah. So it has, I mean, it's the unpredictability of it, which is what we kind of started at the start of the conversation.

Speaker B: Right.

Speaker A: And that's the weakness, really.

Speaker B: And maybe the unique thing about humans is that we should have this broad intelligence that's pretty good. A, uh, whole lot of ton of things. But also. And, um, maybe the other piece of the puzzle that isn't and maybe will never get baked into these AIs is that we are just autonomous. We have drive, we have feeling, we have needs, and we need to have those needs met. And all of that creates a sort of unpredictable. You know, those are all these kind of like this special alchemy that feeds into our intelligence. And. And, um. And even as I say that there are people out there that have just. There's a. There's a new kind of model making waves called claudebot or just being rebranded as Maltbot. Have you heard about this?

Speaker A: No. No. What's that?

Speaker B: So Moltbot is a like, um, it's like a build on Claude code. So Claude code is what people are using to vibe code now. It's like the. It's like the best one. It's like the one that people like the best. And it can. You can give it a task and it'll work on your computer until that task is done. Right now some bright spark out there has figured out that you can also give it like, um, an autonomous sort of like, make your own decisions go away. Like, react to what's coming in and then choose what the right circumstances are and then go and do the thing without being prompted to do it.

Speaker A: Right.

Speaker B: And so in a fun twist, someone has now taken Claude, so all of the nerds are like, using it. It sounds like I haven't played with it myself. Apparently it's very risky.

Speaker A: But I think, yeah, sorry, go, uh,

Speaker B: on one group of, of one Nerd, excuse the term, but, uh, has created a social network in which you could only join the social network if you are one of these Maltbots. Right?

Speaker A: And on this, they're all joining it.

Speaker B: So the. So people are like, given their Molt Bot access to this social network, you can only post on it if you are a Moltbot. So all these Moltbots are now posting and having these conversations with each other, which are all getting very sort of like creepy and existential. The headlines are that like, oh, one of them formed their own religion. And again, people to like, subscribe to the religion and then they're having like existential conversations about, oh, should we form our own language so that the humans can't understand?

Speaker A: I'm just glad they don't have bodies. Right. Right now. Yeah.

Speaker B: Well, that's next. That's the physical, physical AI.

Speaker A: And that's what's, uh, that's a whole another podcast.

Speaker B: It's a whole another podcast. And that might also be one of the unique kind of pieces of the puzzle that the current AIs are lacking, which are not enabling them to reach this, like AGI as we imagine it, or at least as I think Demis Hassabis, the founder of Deepbind, imagines it when he says it's five years out, 10 years out. I think that's because he sees it as there being these fundamental things are required for them to be able to do whatever a human can do. And one of them may well be a physical, physical body. So, um, anyway, you got me, you got me going there, Pete.

Speaker A: No, that was interesting. Very interesting.

Speaker B: Okay, good.

Speaker A: Excited.

Speaker B: All right, mate. Um, uh, we really have run over now, so thank you so much for giving me your time. Before we go, is there anything that you want to plug, anything, you know, anywhere that listeners can find, you find your work?

Speaker A: Yep, um, I can be found. Uh, my website, website is, um, www.k1-architecture.com. um, and then Instagram is PeterZWilliams, if anyone wants to follow me. Um, I find, uh, there's quite a lot of people that ask about, you know, was it difficult to relocate as an architect from the UK to North America? Um, and so that's kind of a topic I find. If anyone's got any questions on that, I'm happy to help out.

Speaker B: Great. All right, fantastic. Um, and, uh, all of those will be in the show notes as well. So with that, thanks so much for joining me, Pete.

Speaker A: Great, thanks a lot, Bobby. Thanks for having me.

Speaker B: And to, ah, our listeners, thank you for joining me.

Speaker A: We'll see you.

Speaker B: Next time on it's not the End of the World. Unless it really is the end of the world.