SPEAKER_01: And now what's really cool is that I think everyone is aware of the language model, so it makes the conversation way easier. I think if I was trying to do language model research as a startup three years ago, it would be really hard. But today, because people can see the benefits of using those type of algorithms, I think it's a super exciting opportunity. And I really see that things keep getting better and better and better, so that's even more exciting.

SPEAKER_02: This Week in Startups is brought to you by Crowdbotics. Great ideas can change the world, and Crowdbotics is the fastest way to turn those ideas into code. Get a free scoping session for your next big app idea at crowdbotics.com slash twist. Hampton. Are you a startup founder or CEO seeking expert advice? Join Hampton, the private, highly vetted community for high-growth founders. Get invaluable insights, connections, and support to accelerate your business at joinhampton.com slash twist today. And the Microsoft for Startups Founders Hub helps all founders build a better startup at a lower cost from day one. Startups get up to $150,000 in Azure credits, access to free open AI credits, free dev tools like GitHub, technical advisory, access to mentors and experts, and so much more. There is no funding requirement, and it only takes minutes to join. Sign up today at aka.ms slash thisweekinstartups.

SPEAKER_05: All right, everybody, welcome back to the program. We have another edition of our innovators and builders in AI. Today, I'm really excited about our guest because I love building product.

SPEAKER_07: I love UI, user interface, as we say, in the industry. And our guest today is building Galileo AI. His name is Renaud Bonnard. Welcome to the program, Arnaud.

SPEAKER_11: Yeah, super happy to be here today. Thanks for inviting me on the show.

SPEAKER_05: And you're French, I can hear. Are you in France or are you here in the Bay Area?

SPEAKER_16: I'm in San Francisco. I've been living in the Bay Area for more than five years now. And I'm from Geneva in Switzerland, which is kind of a mix between France and Switzerland. Yes.

SPEAKER_18: Yeah, they do. And where is there more chaos right now on the streets? Paris or San Francisco?

SPEAKER_20: I kid, I kid. Let's not get political here.

SPEAKER_21: It's too early for that. But I mentioned you on the all-in episode 115 when we were talking about your product. I don't know if anybody mentioned it to you, the mention.

SPEAKER_00: Of course, and I'd like to thank you again for the shout-outs. It was really a huge catalyzer for our launch. So really appreciate it. I was actually just listening to it while doing an errand. And I was like, oh, well, that's us.

SPEAKER_11: So it was pretty funny to hear the name of our company while doing something else. And it was a nice surprise. Thank you again.

SPEAKER_21: That's fine. Just leave a little room on the cap table for your bestie, J-Cal. So I always ask my team, my producers here at This Week in Startups, to show Don't Tell. I was wondering if you could demo your product. Now, because some people are listening, we want to sportscast it a bit. So if you could describe what's happening on the screen.

SPEAKER_05: And I'll help you do that a bit if you don't do it perfectly. So no worries there. But if you are not watching right now, a good time to go to youtube.com slash thisweekin and watch this episode there. Or if you're using Spotify, you can just click the video button or turn it sideways. It'll play the video. And you can always search for This Week in Startups, our video feed on iTunes. So we have a separate feed. Podcasting feeds are not perfect. But Arnaud, take us through.

SPEAKER_29: I see we have here a prompt box that says, describe your design. And then the word generate. So walk us through how your product works.

SPEAKER_00: Yeah, so our product is, it's very simple. So we, we allow people to type the design they want to generate. And basically, I'm going to narrate what's on the screen. It says, design me a chat interface. Well, Ed Sheeran and Batman having a conversation about quantum physics, as they usually do. And it will say that it's a back and forth conversation. And basically, what we do is that we use that prompt for our AI. And the AI will generate a layout for you that is relevant to the prompt you wrote. And it takes a bit of time to generate, but then you will get the layout that makes sense. So you can see here, Batman and Ed Sheeran having a conversation about quantum physics. And it's all generated right now in the browser.

SPEAKER_16: So you can see those the content, the images, and everything. And it looks like iMessage, right?

SPEAKER_34: Like an iMessage thread.

SPEAKER_00: Yeah, big, I mean, the most messaging interface, I would say have a similar design, but yeah.

SPEAKER_36: Yeah, yeah, like, or maybe WhatsApp or whatever.

SPEAKER_00: Okay, exactly. And then another one is design me a page to create an event, asking users about event title, description. And I think it's time afterwards. And I think this example is interesting, because you can be very prescriptive about what you want to generate. So before we're a bit on the high level side, where I say, just give me some messaging.

SPEAKER_01: But what if you're a PM that is working on Facebook events, and you really need those fields to be on the UI?

SPEAKER_00: Well, the AI is able to understand that and make sure that those elements are going to be on your screen so that you can have a bit more control when you generate the layout. So you can see here, you will get the title, the description, date on time, and also a way to invite attendees and create the events.

SPEAKER_39: And now, just to be clear, what are we actually making here? Are we making code? Are we making a JPEG? What exactly is being built here?

Chamath Palihapitiya: That's a great question. When we first launched, a lot of people thought we're generating an image, and they're like, okay, I have this image.

SPEAKER_01: Now maybe I can, you know, paste it in a slide or in Notion doc or share it with my team.

SPEAKER_00: But what we're actually generating here is actual structured data, actual code.

SPEAKER_01: So the way we see it is that it can interface with design tools that are already part of the designer workflow. But if you can also generate code, so you can imagine that if you want to use this design part of your developer workflow, you can also do that. So it's, it's, it's not a JPEG, it's actual code that gets generated.

SPEAKER_44: Got it.

SPEAKER_00: So another example here is the homepage for eCommerce app, selling cool stuff for techies, like camping gear. So back to the San Francisco theme, we were able to generate an eCommerce app that can sell camping gear.

SPEAKER_21: And I could take these and I guess click a button to export it.

SPEAKER_05: And I don't want to go ahead of the demo here, but into a Figma, or maybe I could just publish an HTML page and you could host it for me just as like a little mock-up page.

SPEAKER_48: Yeah, so for now we are doing an iteration, we are integrating with Figma, but as you can imagine, we are iterating with our users to see what they want to do.

SPEAKER_01: I think in general, in that field, everything is kind of open because Genetive AI is only one year old.

SPEAKER_00: So the way we see it right now is that we focus on two things, speed and quality. So we want to make sure that our designs are high quality and also that it doesn't take too much time to generate.

SPEAKER_01: And that way, the person making the designs can supercharge their workflow. So you can see here on the screen that it's able to generate a eCommerce app. And I used to work at FAIR, the marketplace startup, and I can tell you that collection is something that an eCommerce website will have and popular items as well.

SPEAKER_00: So it's because we are sitting on, you know, shoulders of giants and large language models, we're able to generate very interesting features.

SPEAKER_21: You're saying some of the features here that are put into the design, we're not explicitly asked for in the prompt, it's referring that you might want a feature collection or popular items.

SPEAKER_01: Exactly, because that's something that is very popular in eCommerce. And I think that's the value that AI can provide is that it can kind of expand the creativity of the person using it and get new ideas for features. So you see here, I put the same prompt a second time, but this time the features are very different. It asks about, it shows, for example, the top deals. So maybe if your brand is more about discounting and selling stuff in bulk, you might do better, you might do deals. But if on the left, you're more of a fancy brand and you want to do collections, maybe, you know, a handbag brand, something like that, there'll be more of a feature on the left. So I think that's an interesting part of AI. It's able to really show new ways of designing your product without you prompting about it.

SPEAKER_56: Amazing. It truly is impressive.

SPEAKER_58: Probably the most common challenge I hear from founders is related to building. Either they aren't technical and are searching for a technical co-founder, or they can code, but they're just spread too thin. This is one of the first major obstacles you're going to face, and I know how discouraging it can be. But there is a solution. Do you have a great idea, but you don't have a technical co-founder? Well, Crowdbotics can be your CTO as a service. Boom! Just like that. This means you can focus on building an awesome product and delighting your customers, rather than wasting your time on infrastructure planning, architecture, compliance, and all that boring stuff. Crowdbotics also offers professional scoping to help you flesh out your project at the MVP stage and beyond. So cut out the hassle and get back to building that perfect product for your delighted customers. And when you think Crowdbotics, I want you to think, getting your time back to focus on product. Product drives everything in a startup. So let the folks at Crowdbotics show you how it works. Schedule a free scoping session and get your detailed build plan at crowdbotics.com slash twist. That's crowdbotics.com slash twist.

SPEAKER_21: How long have you been working on this?

Chamath Palihapitiya: So we started the company in September, 2022. It was me and my co-founder, Helen.

SPEAKER_01: And we've been working on this, well, since September, 2022 and run more than a hundred experiments. But at the same time, I've been working in AI for most of my career. I was working on language model and language model research at Google.

SPEAKER_00: And my co-founder was working at Facebook on consumer products. Also led a team at Cruise in design.

SPEAKER_01: So I think it's been brewing for a while, the idea of AI and design for both of us. But then the technology happened, LLM happened, diffusion model happened.

SPEAKER_11: And I think it created the perfect combo for creating that type of product.

SPEAKER_65: It's, yeah, it's pretty amazing how quickly this is moving. And so, uh, do you have users yet?

SPEAKER_21: Are you still in the laboratory building, essentially this, uh, MVP, our customers on the product yet? And then how do you make money?

SPEAKER_01: Yeah, so at the, at the current point, we, we launched in February, the, we did the soft launch and we got a lot of people on the wait list. And right now we're slowly letting people in. So we're taking our time to make sure that we get the right feedback from people

SPEAKER_00: because if we let too many people at the same time, you know, we might get people that say, I want an e-commerce app, other people that say, I want food delivery. And then we don't know how to focus, prioritize the product. So for now we're slowly letting people in.

SPEAKER_01: And in terms of pricing, what I've noticed with generative AI products is they tend to do, um, a lot of, uh, they call it consumption-based pricing, which is you pay per generation, which is kind of grouped as a, uh, a subscription fee. So we see doing something very similar where you would pay per generation and maybe per month, you, you get a certain amount of generations, uh, for free. Yeah.

David Friedberg: Uh, I mean, I mean, a certain generation is included in the price. How do you look at the sort of AI models you're, you're, I'm assuming you're working

SPEAKER_21: with the chat GPT or open AI, uh, stack right now in Azure or something like that.

SPEAKER_01: Yeah. So the way I look at it is that we kind of change with the time a lot because every two months there's basically a model that keeps getting better and better. So when we started the project and the product in, in September, we're working on specific type of model, then that someone releases a new model and we kind of bring our data to the new model. I think kind of every two to three months, we need to change the way we do things because the field is moving so fast. I think what's interesting is when I was working in NLP research at Google, there was maybe in 2018, a cool paper every three months that people would talk about and now it's every week. So we need to make sure that we keep the pace and we're able to adapt to all the new models.

SPEAKER_21: And so the training data is whatever they trained it on, or are you making your own training data because there's a, there are specific websites, dribble behance that have incredible collections. And I guess the question that will come up here is, Hey, if you're coming up with a feature set and it's learning, uh, from the open web, what other people's products are, um, how do those designers and UX folks get compensated, uh, if the AI is now coming up with the answer?

SPEAKER_04: So have you given that thought?

SPEAKER_01: Yeah, that's a good question. I think the way we see it is, uh, in terms of the data, we don't use, you know, the raw data from, from dribble and trade a model on it. We created our own process to create that dataset. I, I, I used to build datasets for Google Translate for languages that are kind of unknown or really rare languages. And we had to be really creative on how we create that data. So the, the, the way we, we generate that data for us, it's, it's all in-house. We all do it ourselves. And so we're not very concerned about, well, I wouldn't say concerned, but we don't fall into the, maybe other problematic usage of AI where it uses directly the, the other people's work as input, because that's not what we do.

SPEAKER_21: Yeah, getting, getting images is not your source material. You could build up models of what apps look like based on your own knowledge of design, et cetera. So you can start making these definitions, et cetera, but you could also do a commercial relationship with dribble or Behance, uh, or the artist on there and say, would you like to contribute your designs and then link back to them and they could get traffic back? So if I was using, if I was a designer who had made beautiful, I don't know, uh, e-commerce websites, I might very much want you to feature some of my work in here.

SPEAKER_01: Yeah, no, that makes all sense. And when you think about it, that's, that's how, you know, webs that, that's how templates

SPEAKER_00: work today in a way, right? Well, people sell templates on another platform and they take a cut out of it. I think with generative AI, the, the interesting part is that I like to think that it generates completely novel things. So I think the, the, the hard part to solve is to say how much of the data was used to generate this versus, uh, is it something completely novel? It interpolated from, you know, just previous knowledge.

SPEAKER_21: Yeah. This seems to be like a sort of interesting part of all this, which is, Hey, can we come up with new ideas, uh, and then where do these new ideas and then how does the attribution happen, um, it, and then there, there are gonna be IP questions. I mean, if it pulls, you know, who is, who, nobody owns the standard of a shopping cart. Shopping cart is a generic device on the web, uh, but the execution of shopping carts, there might be some copyright involved in there. So there, you're gonna have to be very thoughtful about that. And then your, your video went viral. Yeah. And so you had over 2 million people watch that. Congratulations. Thank you. And, uh, that's got a, oh, and I see you went to an accelerator previously. Yeah.

SPEAKER_11: Yes. I went to South Park comments.

SPEAKER_85: I I've never heard of that. Tell me about that accelerator. How did you find out about it and what do they do there?

SPEAKER_00: Yeah. So this accelerator basically is, uh, based in Soma. It was created by the former CTO of Dropbox and they have a big focus on, uh, AI. And so that's a, it's a great community where you will find people that, uh, are looking for the next step in their career towards entrepreneurship and you get to, uh, connect with people in the field of AI and they have a program called founder fellowship where they, they fund, uh, just a few startup is very small cohorts. Uh, I think my court had less than 10 people. And so they focus program where they help companies, uh, you know, find product, like find what they want to work on and what they want to build.

SPEAKER_85: Amazing.

SPEAKER_18: When you're the founder of a high growth startup, things are chaotic. Pretty often. Let's be honest. You face a ton of questions. You got problems that you don't always know the answer to, especially if it's

SPEAKER_07: your first startup, but you don't have to do this alone. That's where Hampton comes in. Hampton is a highly vetted private community exclusively for founders and CEOs. What's their mission? Founders mission is to create the most valuable and engaged community for high growth founders. And here are why some members have already called it life-changing. When you join Hampton, their membership team carefully handpicks seven other members to join your core group. It's almost like a mastermind group, right? Okay. So now you've got your core group. This group is like a personal board of directors for you. They're going to provide you with advice, critical feedback. And they're going to aim to help you accelerate your growth.

SPEAKER_18: I mean, that's what we're all in it for anyway, right? Personal growth. We want our companies to grow, revenue to grow, uh, the amount of money we raise to grow, all that good stuff.

SPEAKER_68: So Hampton members run some of the fastest growing startups in the world. And you probably use some of their products like morning brew, blank street, coffee, dribble, and more. The connections, personal accomplishments, and a sense of belonging are what make the Hampton membership. So special community is everything. I, you know, that like here, this being started artistic community, and here's an even smaller, more focused, tighter community for you to take a look at joining. If you're ready to scale your business, join the Hampton community today at joinhampton.com slash twist. I love this idea joinhampton.com slash twist today.

SPEAKER_29: You must be drowning in venture capitalists chasing you since everybody's in AI right now. Tell me about what it's like to run a promising AI startup right now in Silicon Valley.

SPEAKER_00: Um, it's super interesting. Uh, it's interesting because when I used to do language model in 2018, no one cared about it. Uh, I, I, I worked on the, um, you know, I work on Gboard, uh, the Google keyboard on Android. And when I was explaining things to people, I think they were excited because of, of the, the product itself. But then when I went into the technologies, they're like, what do you mean language model? Okay. And now it's, what's really cool is that, uh, I think everyone is aware of a language model. So it makes the conversation way easier. I think if I was trying to do language model research as a startup three years ago, it would be really hard.

SPEAKER_01: But today, because people can see the benefits of using, uh, those, those type of algorithm, uh, I think it's, uh, it's a super exciting opportunity. And I really see that things keep getting better and better and better. So that's even more exciting.

SPEAKER_21: It does seem like, uh, yeah, the world understands this now and it's gotten just a lot better.

SPEAKER_93: You know, guessing the next word in Gmail or a Google keyboard was, you know, delightful, but it didn't have a huge impact on people's lives, but certainly, you know, composing an entire email does have a big impact, uh, or, you know, filling out a whole box of code or making a UX. It does seem like you're also on a collision course with a couple of groups of people, uh, somebody like Figma or Canva and, uh, and Adobe, they're all gonna be looking at this and saying, Hey, how do we incorporate these features into our platforms? Uh, whether Canva for people doing more casual designs, uh, or Figma for doing more, you know, rigorous designs, uh, and the same with Adobe's suite of products.

SPEAKER_05: So how, how do you look at being an AI first company, I think where people just want to go to is they, they just want to publish these pages to the web and then eventually build a functional websites. You got people who are building code.

SPEAKER_93: You're building front end. Eventually this is all gonna come together in one product. Is it not?

SPEAKER_01: Yeah, that's, that's, uh, that's what we have been thinking about a lot.

SPEAKER_00: I think when it comes to, uh, in general competition, I think was just at the very early stage of, of that new generative AI revolution. Uh, the way I see it, I think that software is gonna dramatically change. I think the, what we see currently in, in the products that people use every day, it's gonna change, right? Like I use chat GPT every day for, for coding.

SPEAKER_01: I wasn't before. So I think there's gonna be in general AI first company are gonna have that advantage of being able to start from the ground up with AI. And in terms of what we can deliver again, I think it's also, there's a lot of, a lot of interesting opportunities. Like you said, like if we generate the front end, then, you know, the, the world is your oyster. You can do whatever you want. I think our focus for us is really quality and speed. And I think once we get really a threshold of quality that is really high and that people are super excited to use our product for that reason, then we can think about, like you said, uh, other type of opportunities.

SPEAKER_00: But for now, we really focused on, on, on design and generating beautiful UIs.

SPEAKER_21: And specifically web design, app design, because there's a lot of design out there. You got people who are making illustrations, images. Yeah. Yeah.

David Friedberg: Yeah, that seems, and the, and the ability for it to write the code correctly is critical.

SPEAKER_98: And so getting that right is, is super important as well for you. Yeah. Yeah.

SPEAKER_00: I think writing code is, is a, is a very interesting, um, you know, I think it's, it's super interesting to think about. And I think the, the, the tricky part is like, how do you get it inside a developer's workflow? I'm a developer myself. And I think developer is a crowd intelligent, internal customer developer is a crowd that, um, have high standards. So you want to make sure that you cater to those high standards. Yeah, for sure.

SPEAKER_29: Uh, all right, listen, this is incredible. How far you've come in a short period of time. We'll be watching.

SPEAKER_21: Uh, I'm, I'm sure you're hiring right now or looking for beta customers. If somebody wanted to try the product, how do they sign up? And if somebody wanted to join your team, what type of people are you looking for? And how might they, uh, contact you and apply? Yeah.

SPEAKER_00: So for the website, you can go on usegalito.ai and for hiring, we're hiring full stack and front-end engineers right now. So if people want to join the adventures, are excited about larger models. They can reach out to me at, uh, on Twitter with, uh, at Arno AI. Uh, that's my Twitter handle.

SPEAKER_109: Or just send me an email. Arno at usegalito.ai. Fantastic.

SPEAKER_111: And did it have to be, are you, are you an in-person guy? Uh, I know you're.

SPEAKER_00: Yeah, I think for us, we like people to be in person for a few days. Um, because we have a tight calibration between AI and design, and it's super important for people to collaborate and able to iterate on our product. I agree with you.

SPEAKER_21: I think that the teams that who are, who are brave enough to, you know, go back to an office and have a couple of people sit around a table are going to do better. Uh, now I know some people have gotten used to working from home and there's benefits to it, but I, the people who work remote will have one set of benefits.

SPEAKER_114: Might be, you know, they can put more time and have less commute times, but the people who are in person a couple of days a week, they're gonna, I think they're gonna iterate a little faster on their product. No.

SPEAKER_00: I think it's just because it's very visual in general, right? Mm. It's like sharing a design on zoom and then you wanna quickly iterate between the design and the code and everything. It just makes things a bit easier, uh, because of the visual aspect of the product.

SPEAKER_21: You know, I, I saw you had tweeted as well, um, about auto GPT. Oh yeah. A lot of people playing with it and having a lot of fun with it. Um, it is early days. Maybe you could explain to the audience what auto GPT is and your position on it.

SPEAKER_00: Yeah. Uh, I know, I know a lot about language model and I think auto GPT is like 10 days old. So take that with a grain of salt. But so I think the auto GPT is a way for a language model to, to do task autonomously. They can decide, you can just send them a high level task saying, you know, book me an appointment to the hairdresser. For example, it's gonna be able to interact with an environments and, uh, perform actions basically. So it's kind of like a personal assistant, but as, as an AI, I think it's interesting. Uh, my, I know I wrote a tweet about it and I think my take was that what happens when it fails. And of course we can think about very dramatic failures, but just simple failures, right? Well, there's a time change and it books the wrong appointments. All it's just goes on the wrong website. So it's just, it does something wrong.

SPEAKER_01: So my take was we should figure out a way to maybe that auto GPT to do something. And then the person can just approve that it was the right thing. Maybe the same way that if you have to send an important email to someone, you would just double check and then send it. So maybe that's how those models will go.

SPEAKER_21: I, I am playing with it myself. I got the plugins. I connected Zapier on chat GPT four. And then I sent a hello world email the other day and I had the same exact premonition as you. And I'm, I'm playing with kayak and it's like, do you want to book this? And I was like, Hmm, yeah, this can go really in the wrong direction. And it's sort of like the early days of autopilot. So on my first, my second Tesla, uh, it had autopilot, uh, keep you in the lane adaptive cruise control. And if you could do a lane change, but it confirmed it with you. So it asked you to actually move the stick, uh, to make the change and to look in the mirror.

SPEAKER_05: Now on my model, why I have the latest update of full self-driving and it's like, how aggressive do you want us to be in changing lanes? Just in terms of what kind of riding style do you want? And it is bulletproof as bulletproof could be. It's a thousand percent better than any human could ever be. And it actually shows you the side view camera. It shows you, you know, uh, the simulation. And so this idea of just monitoring the AI, I think is a very, uh, worthwhile guardrail, uh, to put in place and a thoughtful thing to put in place.

SPEAKER_21: So I think you're exactly correct.

SPEAKER_07: And, but the auto GPTs are going to be super powerful. Yeah.

SPEAKER_00: And I thought, I think sometimes also language is not the best way to express certain things. So imagine I'm really, I'm looking, I'm going on an e-commerce website and I am going to say, I want to buy a jacket between 85 and $90. Um, it needs to be denim.

SPEAKER_01: I think, you know, typing that in natural language is not the best interface versus just clicking on, on buttons. I think the same way with auto GPT. Well, if you only describe a task, maybe sometimes there's a limitation with, with language. Maybe you want to have different type of an interfaces to, to talk with it. Yeah.

SPEAKER_39: All right. Listen, this has been an amazing discussion. I wish you, uh, well, any, any, uh, good French food recommendations, uh, in San Francisco Bay area.

David Friedberg: Yes. Cause I, I, you know, like a good steak, frites, mules, frites, uh, escargot. You got a, you got a favorite place.

SPEAKER_00: Uh, I think Chez Mamon is pretty nice in general. They do good, uh, steak and fries. Uh, so that's, that's pretty nice. Chez Mamon? Yeah.

SPEAKER_13: Chez Mamon. Ah, I gotta check that out. Yeah.

SPEAKER_11: What's wrong in San Francisco?

SPEAKER_21: Um, you know, I have a, it's, it's funny. You mentioned this. Um, I really love Japanese food. Okay.

SPEAKER_85: I just got back from Japan. I, you know, I couldn't go for a couple of years.

SPEAKER_21: They had it on lockdown and there is a, um, a specific dish, um, called, uh, katsu, you know, like it's, uh, yeah. Yeah. Yeah.

SPEAKER_93: Yeah. And so not everybody's into it. It's basically like a fried cutlet, right? And there is a place called cafe, Okinawa, uh, like the island.

SPEAKER_21: Uh, and, and it is, uh, on townsend street. Uh, so it's, it's near you if you're in Soma. Uh, and, uh, they make a sando, a sando is, uh, short for sandwich. Yeah. And they use thick white bread.

SPEAKER_93: So when I was in Japan, I got, uh, pork, uh, tonkatsu sandwich, but then, you know, and they put some, you know, nice, uh, cabbage with it or something like some shaved vegetables, but you can dip it into curry sauce. So when you have katsu and curry, so they have this sandwich. Oh my God. I'll take you for it at some point when I get on your cap table and I, I get a little slice of that, put a little investment in there. Slide a little 250 K for my fund into Galileo. You and I can have a celebratory. Tonkatsu sandwich. Um, but they have the katsu sandwich over there and it's mind blowing. And, you know, this is one of the great things about San Francisco.

SPEAKER_21: People can talk about the problems, but the absolute creativity and the food scene is just in the, even in the Bay area, you go to Berkeley or you go to Napa or you come down to San Mateo area where we have a lot of great Asian food. And so every time I go to the city now, I find myself going to cafe Okinawa, um, and, uh, getting their chicken katsu. They also, there's another weird dish that they do in Japan where they make fruit sandwiches. Oh, wow. Again, with this white bread. So, you know, like wonder bread is like a really weird bread here in America. Yeah. They have a better version of it in Japan. They make their, like their own white bread. And then this concept of a fruit sandwich I had when I was in Japan for the first time, they put like a whipped cream, but not too sweet. Yeah.

SPEAKER_93: And then they put strawberries or I had one with mandarin oranges. And so they have a strawberry one there. So you literally imagine strawberries with a very, like not too sweet whipped cream on white bread. Oh my Lord.

SPEAKER_07: You would think it's like a weird thing. It tastes delicious. Uh, sounds really good. Yeah. There's my San Francisco.

SPEAKER_00: I feel like lately had a lot of amazing AI events. Uh, I've noticed that, uh, there's been a lot of excitement around AI and I think every week you meet a lot of interesting people in that space. So, so it's been kind of a, I think kind of a rebirth of the city on that, uh, in that space as well.

SPEAKER_19: I hope so.

SPEAKER_38: I was there Monday and I had my accelerator. I had seven companies and we were at Fenwick and West office.

SPEAKER_21: And then I went over to Wilson Sincini's office on the Embarcadero. And we had our founder university, which is like people just making MVPs that haven't even incorporated yet. And I had about a hundred founders there and people, the energy level was really high, but walking between the two places, one in the five die and one in the Embarcadero.

SPEAKER_05: I mean, there was nobody there. And I just remember pre pandemic, the amount of traffic in the city to get out of the city took 45 minutes just to get from like the five die or Embarcadero out of the city, half an hour minimum. And, uh, you could just, I drove, got out of the city in six or seven minutes. I mean, it was like, it was like, it's the middle of the night. So it's, I think that sometimes that's the creative destruction that occurs where young people come into the city. Or like storefronts are cheap. And somebody opens up like a little passion restaurant, like cafe Okinawa.

SPEAKER_18: They also have a egg salad Sando that I'm going to try. So I'm rooting for San Francisco, uh, and I'm hoping some of the really great efforts, like gross San Francisco and, uh, stuff Gary Tan's doing and David Sacks are doing.

SPEAKER_114: Uh, but listen, I, I hope to meet you soon and, uh, hopefully at one of these AI events in San Francisco or the wider Bay area. And maybe we'll go to cafe Okinawa. Sounds good.

SPEAKER_12: Yeah. Thanks for everything. And thanks again for all the shout outs and inviting me to the podcast. Of course.

SPEAKER_152: Thank you. All right. Talk to you soon. Cheers. Tom is a senior director at Microsoft for startups. He's a former founder himself. Tom, Microsoft has a huge, huge, uh, cloud computing, supercomputer, Azure. Everybody knows about that, but you've started to run giant LLMs. Maybe talk a little bit about what it takes to run a giant LLM in the cloud. And this is a lot of iron, isn't it?

SPEAKER_154: Absolutely. And really thanks to the partnership that we have with open AI. For the last few years, we've been spending a lot of time really building an AI focused supercomputer. We went back and looked at everything, the GPU configurations, the networking to really get it right. To be able to run these large language models as, uh, and make them available, uh, for startups through the Azure service that we now have that you can get through founders hub. And so, yeah, it takes a lot of compute power, but we've managed to optimize that. We've spent a lot of time working with startups to help them optimize things as well over time. And as more and more innovations come into this, we will make them available to startups, not just open AI, but anyone just starting up a startup.

SPEAKER_155: We offer $150,000 worth of credits to, so startups can leverage that and really get going on their journey.

SPEAKER_152: The founders hub that Microsoft provides offers $150,000 in Azure cloud credits. All the development tools like GitHub and. Teams office, all that great stuff. You get all that for free five minutes to sign up six figures in benefits, AKA.ms slash this week in startups. Thanks so much, Tom.

SPEAKER_161: Hey there. My name is Scott and I'm a partner at a branding studio lunar. And today we're going to talk about how to design love products. Let's dive into it. So first, are you ready for a scary truth? And this is that 88% of online consumers are less likely to return to a site after just one bad experience. So what does that mean for us looking at the positive side of design? Is that the ultimate goal of us to design well is to solve someone's problem in an enjoyable way. So just creating good, beautiful designs by themselves is just art. However, it should solve a business problem. And so let's dive back into why that matters a little more. First is that 73% of mobile app users turn after just 90 days on average. And this is typically because it doesn't solve their problem for the long term. It was nice at first, but then it died off. Next is that first impressions are 94% design related. So when someone comes to your site, you have about seven seconds for them to see if what they're looking for is right. And if it looks like it can be trusted. And then if they don't find what they're looking for, they'll bounce. And finally, 38% of users will stop engaging if the website's content or layout is unattractive. Again, spoiler alert with this is that people do judge a book by its cover. So looking at how do we design products, you probably have seen this quote before from Reid Hoffman, which is if you're not embarrassed by the first version of your product that you've launched too late. And that means that if you are building this beautiful design and you never launch, then that's probably not as important as just getting out the door and get real feedback from users. So with the goal that we've seen that design matters a ton, but also launching matters more, how do we do both? The way that we do that is what everybody's familiar with, which is starting with the MVP. So here, step one, we design a smaller focused piece. And this is something we can test and make sure that it's solving a real problem in an enjoyable way. And then from there, as we grow, then we can scale and improve as we go on. And that's where we can build out, hire more people and then make the design like we actually want it to be. So I'm going to cover a three part framework for what I use for designing great user experiences for successful designs. And I've done this at large banks and I've done this at startups and the same process works for both. So step one is I always start with text based requirements. And this is simply just what you see in the little thumbnail. If you can see that, which is just text saying, when I go to log in, here's what I expect should happen. When I go to upload a photo, here's what I expect should happen. And I usually stick to something like notion or Google docs for that. Next is that I go into rough wireframes, and this could be a mix of a couple of different layouts. This is something like sticky notes, whiteboards, or honestly, just using a napkin sketch does the job. And then finally, we go into making polished designs. My favorite is Figma, but you can use a bunch of other tools to essentially create the pixel perfect designs that will be handed off to your development team. So let's dive into each. The first one, text based requirements. And again, this is how I start every single product. And I always do this with clients as well. So when we talk about that they want to release a new feature, I'll start with just text. And the test that I use for this too, which I'll cover a little later is, can you explain this over the phone to someone? So if you say, Hey, at this screen, it needs to do this. And again, it's, it's more focused on the function side. So how I do that is I start with a summary statement, which is specific group of people have X problem. And it would be, and they would be willing to pay Y dollars to solve it by, and then whatever your solution is. Again, this is a business problem that we're solving through design. The second part is I'd focus on the problem. I really want to make sure that I'm focused on solving a real problem, not creating a solution in search of a problem. And then finally, with this is I would focus on the personas. So who are these people that we're solving their problems for? And then next I'd move on to what is the home run for them? So they, they have a problem, but what would it actually be for them to have a solution that they want to tell somebody about? And then I go into the requirements. And this is what I mentioned earlier is that I go specifically down step-by-step up. Here's what I expect at each stage. And with this, I'd speak to the, the function verse. This is how it should look. So if I say, Hey, they should log in and then continue to the next step. I didn't specify that the buttons should be look like this or the color should be this. None of that really matters at this stage. It's simply focused on function. And I typically use notion for that, but you can use Google docs, word, anything with just text.

SPEAKER_160: And to illustrate why this matters, if we take a simple login screen like this, so this is a, say, create your account. Might look pretty simple. We've all seen this, but if you're going to ride out each step of how this functions, you might pull out some things like this.

SPEAKER_161: Name. Do you need the first and last? Do you need both? Is it a required field or is it optional? What about email? What happens if they misspell it? They misspell Gmail. And how do you verify that? What message do you put under it? Is it really static, like incorrect information? Or do you say, Hey, try again. What about, are there password requirements? Do you require 27 characters? Do you require a special character or are you going to let them put in whatever they want? What's a sign in text for the button? Should that be, get started, sign in, let's go. What about social logins? What happens after they sign in? That's the next step. That's really important. And what about a confirmation email? There's a lot with this too. Once you sign up, do you receive one? How, what software, what provider is it through? What's the text? Who is it from? All of those things need to be defined. Next step two. And there's two parts to this one. I start with journey mapping. So I, I outlined the experience from their shoes. And the first step with this is to make a realistic persona. So as you'll see on the side with the visuals, I use essential digital sticky notes. And, uh, I use fig jam or envision has a great tool called free hand. And this is where I essentially make digital sticky notes. And I say person a let's call him Sam. Sam wants to try to, uh, go downtown from where he's at at his apartment. So he pulls out his phone and he opens an app happens to be the Uber app. And so he puts in the destination. And so if I were to build out the Uber app, I would just simply say what he's doing almost as if I were to watch a movie of this scene and everything is done by stickies. And that's where we map again. This is a, they do this, then they do that. And then they also, they may be thinking here, here's what they're frustrated by. Here's how I can make a home run for them. Some software I use again is envision app or free hand, a fig jam Miro. You could use simply just a white board with paper sticky notes. And then next, the second part of step two is the wire frames. And in short, this is simply just boxes and lines that communicate what all those steps are. And as you'll see from these images, again, these are real ones that I've used and they are ugly. And that's by design. So here make ugly, quick sketches. You should not at this stage, from my perspective, focus on making really polished wire frames. At this point, they're just the focus on what is the experience going to be like. And I've found I get better feedback on the experience. If I make the wire frames ugly, because as soon as I start making them look good, people are going to focus on what the button styles don't look like that. What color is going to be here? What mock-up will be in this place? I don't care about that at this point. I just need to focus on, does it solve the problem in an enjoyable way? Focus on this again. I use grayscale only. I don't even use any color when I do it. Or I should say, unless you see here, there's colors, but it's more to illustrate that these are different objects. But that's not anything to do with aesthetics. I can use Envision app or Freehand, FigJam, Miro, use a whiteboard or paper, napkin, any of those work. Next is jumping into the polished design. This is what you came for. So make development ready designs. And something that designers have to learn too, is that we as designers want to make everything pixel perfect. We work on our eight pixel grid and everything is just beautiful. Well, translating that to the internet where developers have to account for desktop, tablet, mobile, constantly changing screen sizes that we have. It's never going to be exactly pixel perfect, but developers are getting better at doing it that way. So my first step is I will build around five primary screens. So let's say again, if it's Uber is your app, I'd say probably one of the main screens is me putting in my address to where I'm going to go. And then the next, next screen might be what happens if I'm actively driving. And then maybe the other key screen is what happens when I'm done with it. And so that's about three screens. And if I were to do that, I would have some sort of map layout. I'd have inputs to be able to put in for my current address, my destination address. And then what happens when it's actively going. And what that does is I start building these key screens and it gives me buttons, inputs, map, some sort of icon for a vehicle status. And I kind of create a rough hodgepodge of all these styles that I like. And that will show me all the styles that I will need. And then from there, I'll pull the rest of the components from ideally a really well defined UI library. So the UI library is essentially a set of buttons and colors and fonts and other components like inputs and dropdowns and every other thing you can think about for software design. Here you want to just pull from it ideally and keep you reusing the same components as opposed to every single screen. You do not want to create a new button style every single time you go and have a primary, secondary and maybe tertiary options there. And then finally is that I would make sure that I have all those screens built out for all of the or all the layouts built up for all the screens that I'll need. So if I have a standard app, you probably have to take in consideration desktop, tablet and mobile screens. And with this, make sure that it's really clear about here's a desktop version. Here's a tablet. Here's the mobile. Because when developers go to build it, they're going to ask those questions about, well, how does this scale down? Is the picture to the left still or in the text to the right? Or does it shift when it's on mobile? Design that all out. Make your developers happy to give you a quick picture of TimeWell, which is an app that I've designed. These are the screens that I've used. I've cleaned this up since, but you can see here all the variations of the different screens I've had to use. So like I mentioned before, if you're creating an account, what happens if you put in the wrong email? I have that out. Or what happens if you're uploading a photo? What happens if it's in progress uploading? What happens if there's an error? I have to think about all of those different states to design and see desktop, tablet, mobile, all these breakpoints. And the simple part of this is that every single interaction has to be designed or at least talked about. So when we compare now what you're seeing here, which is the design side to the text side of here's what should happen function wise. And then you have the same thing map for here's how those same functions look. You have a really great way to design and develop a really great product. If I can give you one tip is as you are starting to build out your platform, just start with a design system or UI library. These are pre-built buttons, colors, fonts, dropdowns, navigation menus. It's all the pieces that you'll need to start with. And they're usually like the ones I have listed here are around a hundred bucks. And it is probably the best money you could spend when you're starting to build out. Don't waste time building buttons from scratch when you could use great ones. The reason for this is that people are going to judge your app just by glancing at it in a second. And if it doesn't look good, they're not going to trust it. So I'd say first version, just go with somebody else's pre-built framework. It's kind of like using somebody else's Lego blocks just to start there. Finally, we're going to go through a speed round of resources here and then some do's and don'ts for design. So if you are looking to build out your designs and to make sure that they look good, I have a bunch of resources here. I would go ahead and check those out. So let's jump right in to our speed round. So UI UX, which also means user interface and user experience best practices. So what it looks like when it, how it functions. These are polls from refactoring UI, which I'd recommend looking at their resources as well. First one is start with what feels like too much white space. As you'll see on the side, I find this a lot with developers is that they build everything really packed together. You want to make sure that it breathes and that your eye can scan. People don't read, they scan. And so you want to give them the ability to see sections really quickly. And then it also just adds more air to your screens. Next is the Goldilocks spacing.

SPEAKER_160: So you might have too tight. You might have too big and both feel really cheap. But if you blend it somewhere in the middle, it'll feel just about right.

SPEAKER_161: Again, using an eight pixel grid is really great. Hierarchy. You'll see here that as you squint your eyes, a lot of times you can find what is the most important thing to look at. So as I scan the left one, all the fonts are kind of the same. There's not a lot of weight. But as I look at the right one, they've highlighted and pulled and made some fonts thinner and some wider. So the weights are thicker or thinner. And then you'll find that some text is larger and smaller. Again, you want to help them scan for the most important things. Here, simplify when possible. On the left side, this is really common to see if something like name and then the actual name. Job title and their actual job title. When in reality, we know what a phone number is. We don't need to specify what phone and the actual number. So on the right side, you'll see just put in the value. So her name, her position, her email, her phone number. Next is prioritize values. So here, if you look at the left screen, chances are you'll see heart rate first. That's not what's actually most important. The most important thing is in context of this app is probably 82 beats per minute. Here's a clear primary action. So as you look on the left, chances are you'll see the red button first. If you want to position options to somebody, you do not want to put delete first. Most of the time you want to put what is a positive way to interact with your products. And so here you do want to do what's on the right, which is you'll see first is publish and then edit and then delete. And these you'll also see focus for the next one, which are to set up the hierarchy of styles correctly. So you'll have a primary button, a secondary button and a tertiary button. And I would keep to one of these styles. So these three rows are different styles that I would use for an app. So I wouldn't intermix these. So I'd keep a blue button. That's solid. I would have an outlined one and then maybe a link or it could have some sort of bare option. And you'll see those down below. A couple more. Don't stretch your content too wide. If you're like me and you've accidentally booked a movie theater ticket too late, you might have to sit up front and you can't see without turning your head side to side. That's essentially what we're doing here. When we put our text or our content too wide or even goes completely off the screen. So keep it to a reasonable amount. Similarly is with text. You can keep paragraphs between 45 to 75 characters. And with text, more and more people, again, don't read. So keep it to where it's really skimmable, not too thin, but in that sweet spot. And last one here is the group, your text and your images in context. So as you'll see on the left side, there's the title. It says the one size fits all platform. And then the need is simple. There's the subtext. It goes as wide as the next content below. When in reality, this is grouped above to the other text. And so make sure that all of your content groups together. So with that, again, I'm a partner at a studio called Lunar. And if you're at a certain spot where you are looking to raise a round or go to the next stage, we'd love to have a conversation. So your customers do judge your book by its cover. And we want to help you uncover the truths about that. How I can be more successful.

SPEAKER_160: Also, I am a founder. So I am in Jason's portfolio at launch. I did that through TimeWell.

SPEAKER_161: TimeWell is a platform where you can save family memories with your voice. I'd love for you to check it out. With that, thank you so much for your time. Please keep me up to date with your designs. Thanks so much. Thanks so much.