Building generative AI for pitch decks with Tome CEO Keith Peiris | E1727

0:00Okay, everybody. Happy Monday. Welcome to this week in startups. Today, we have two great segments for you. First up, I interviewed Tome, CEO and co-founder, Keith Paris. Basically, Tome is text to pitch deck software. Yes, we've been talking about Genre of AI. Imagine if you could just talk to your PowerPoint, if you could talk to Keynote, Google presentations, whatever it is. Then we have another awesome tactical talk from the Founder University podcast. Jen, Ryan, the CMO of Velocity Growth, is going to walk you through how to get started with paid acquisition. You're going to learn how to find and create custom audiences that you can use for ad targeting and then how to create engaging and relevant ads where you can find your competitors ads and how to see what's working for them.

0:41So, hey, listen, you can copy what's working with other folks. So it's going to be an amazing show. Stick with us. This Week in Startups is brought to you by 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 OpenAI 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. Clumio. Immutable. Air-gapped. Automated.

1:26affordable. Clumio provides simplified backup as a service that helps companies streamline data compliance, optimize long-term data retention, and dramatically reduce their AWS backup bills. Start a free trial to uncover compliance gaps, optimize backup spend, and identify unprotected assets. Learn more at clumio.com slash twist. And Mercury, where innovation meets peace of mind. Now more than ever, startups need a safe place to put their cash. Mercury offers a simple way to manage bank risk and protect every dollar with up to$5 million in FDIC insurance and a money market fund. Visit mercury.com to apply in minutes.

2:13Hey, everybody. Welcome back to This Week in Startups. The pace of artificial intelligence is moving faster than any other technological revolution I have seen. in my 30 years in the industry and i'm including in this the explosion of apps in apple's app store the internet itself during the dot-com boom cloud computing it's just moving so fast so we thought instead of trying to just talk about the news we'll bring the founders on who are building the future and just talk about what they're building and how it's going and then maybe even talk about this pace and where they think ai will be in the next six to twelve months because every literal two or three days somebody produces something that the people i know who've been in the industry for i don't know two or three decades goes whoa and uh the whoa factor here from people in the know is significant today on the program is a founder named keith paris he is the ceo and co-founder of tome t-o-m-e i'm pronouncing it correctly keith yes exactly all right so welcome to the pod tell me what is home and then we'll do a demo of course thanks jason excited to be here i would think about tome as an ai powered storytelling tool for ideas and what i mean by that is um you know when you're when you're thinking about describing ideas using work artifacts you know some of them are pitches you want to make some of them are you know products you want to build at your company and some of them or children's stories you know what you want to put together and tell your kids every night and you know tome is a very simple way to tell a wide variety of stories using different modes whether it's text video image prototype and more awesome so uh let's do a demo here i'm assuming if i wanted to make a pitch deck or i want to make a story to tell my my daughters before bedtime today those things involve uh some amount of text prompting something that seems to be uh really in the wheelhouse of this genre of ai movement and then of course creating images on the fly also in the wheelhouse so it sounds to me like you're combining two of those aspects into unique output am i correct yes exactly what are you what is this all built on are you using uh the chat gpt open ai um you know api or is it a another language model you're using for the language side uh so we're using a couple of different ones to to be honest uh we uh are generating our our story outline from gbt4 you know in the sense that it's read millions and you know maybe billions of stories so it's important to be able to tap into that really large model get the structure of a story um but then we use gpt 3.5 for the littler things uh maybe you want to rewrite a paragraph or something and you know we're playing with uh with anthropic for things like i want to rewrite a sentence because it's just really fast and uh you know we're just trying to use the best tool for the the job to create this customer experience and uh this is truly amazing like while you were talking i was like well let me just sign up real quick over here and uh i just said uh i said what do you want to create and i said i want to create a deck for launch fund for our latest venture capital fund and it went out and it made me a series of uh looks like eight slides introduction to our latest venture capital fund what makes our fund different our investment philosophy our portfolio companies our track record conclusion i mean literally if you were going to create a debt for a venture capital fund these would be some of the seminal slides and this blank page experience that people have is paralysis and what you've already done just in the first sentence here is just like hey uh pretty pretty great start so i don't want to steal your thunder but why don't you start a demo here and uh it also by the way added an image to each of these now these are not stock images i'm seeing these are images that were created and it also put a paragraph on each page to kind of get me started so all of this was done with chat gpt4 and then just creating some images sort of and putting it into a presentation format now were you making presentation software before this and then you added ai or was this all at the same time so so we were uh we were sort of building this product for for years and um uh you know i'd say there's a couple of things that we were sort of tapping into i mean henry my co-founder and i were both consumer mobile people so you know we were really frustrated with the fact that your powerpoints don't scale to a vertical touchscreen device um you know we also were we were both working at cloud companies i was i was at instagram he was at facebook and what you know all of our presentations were screenshots of different web services um and we're like oh we got to build something that connects to all of these different data sources and um and then we sort of had this idea of at some point we got to be able to drive this thing using a large ml model um and we we sort of got this idea because we were reading every deck we could find you know on slideshare um from our friends pitching and um there's sort of this like common narrative structure to every story right like every startup pitch deck has like a problem solution opportunity arc um you know if if you uh i love every pixar movie but they all have a sort of similar heroes journey so we're like wait a minute If we could find a foundational model that had read every story on the internet, we could build something that's truly a 10x gain in productivity.

8:24So we were always building the system in mind, thinking it could be driven by an ML model. But we just weren't sure if it would be in 2022 or in 2025 or in 2030. And then the technology sort of really picked up with GPT-3. And we started playing around with it and realized, actually, this is creating meaningful results already. All right, everybody. Tom Davis is here. He's a senior director at Microsoft for Startup. He's a former founder himself. Tom, Microsoft has a huge, huge 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.

9:11And this is a lot of iron, isn't it? Absolutely. And really thanks to the partnership that we have with OpenAI, 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 and make them available 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.

9:51And 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. We offer$150 ,000 worth of credits so startups can leverage that and really get going on their journey. 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 thisweekendstartups. Thanks so much, Tom. Yeah, fantastic. So give me a demo here. We'll show your screen.

10:30If you're not watching the video, Spotify supports our video on thisweekendstartups. There should be a little video icon. You can click or turn your phone sideways. Or if you're on Apple Podcasts, you can search for thisweekendstartupvideo. There's a different separate feed. We can go to youtube.com slash thisweekend to see the videos as we do them. So yeah, Keith, take it away. Great. so i'll jump right into tome and um you'll notice that some things sort of feel familiar like a slide you know you've got some some slides on the left or pages on the left and uh some things don't feel familiar right wait a minute where's the the 16 by 9 rectangle that's gone um and uh you'll also notice that the way you interact with tome is a little bit different than the way you'd interact with uh you know a slide tool that that came out in the 80s right there are some things that you can directly manipulate there are things you can add but you also have this sort of command system um where you can you know take a document turn it into a presentation you could add a page you could uh you know maybe add a tile like a figma tile if you want to show off your design work and um you know we're going to go into create a presentation um and you know in the old world you'd have to look up a template that someone made years ago try to find something that's roughly matched what you were trying to do um and then you know edit it and modify it but the way tome works is that you can just tell tome what kind of presentation you want to make you know when i had queued up was a fundraising pitch we have a lot of early stage founders building pitch decks and um you can write as much or as little as you want and we'll try to fill it out for you and um you know you can see here you can either read that prompt there yeah yeah for sure it's a fundraising pitch for a company that makes telescopes that takes beautiful images of planets in the night sky um and by the way the aesthetic feels a little superhuman-esque so you have this sort of like command k you kind of uh in the middle of the screen instead of just having to navigate the file new menu like in the microsoft or apple you know google offices kind of thing it's just like here's a all the commands in one location just go ahead and start typing a command and and we'll figure it out with you exactly exactly and we're building more controls where um i'm the sort of person that prefers an outline where i just want the prompt and i'll write my own story or you can you know choose a descriptive story if you want um and then you can choose the aesthetic if you'd like and then examples of the aesthetic there oh great question You might want fantasy, you might want some watercolor, you might want some pop art.

13:10Cyberpunk, my favorite. Exactly. My personal favorite. Yeah. And we're adding more and more controls here, just because a lot of folks are not sure what they can type in. And then you can hit enter. And you can sort of see on the screen, we're first creating the structure. um and then we're filling out each page with you know what what we think should be there um some example contents and uh and an image um you can sort of see in seconds i'll go through these pages first we created this outline you know your pitch you should probably talk about your telescopes your mission your team um and then on each page you know we uh generate an image for you using Dali at the moment and some example text.

14:02And now the really fun thing about Tome is maybe this isn't exactly what you want. This being the R impact page. So we're on the R impact page. You're like, maybe this copy doesn't make sense. Exactly. So you can type in and edit it just like an editor like you would in Google Slides. Or you can sort of ask Tome to help shape this for you. and wait a minute i actually want to uh i want to reduce that you know this is the button i wish i had um at when i was at instagram i've just make this a little shorter um and then it rewrites and then you're like uh so we'll take the four sentences and make it three take three sentences from 12 words each down to nine or something so kind of doing a little hemming way on it just let's make it a little simpler let's make it a little tighter exactly exactly and uh you know maybe this isn't exactly what you want but uh but you just want to see another version of it you can hit try again it's kind of like working with a friend right you're riffing you know um yeah and uh and then you know you can you can add stuff to it too right and like i don't really like this this sort of uh layout so i'm just dragging this handle in the middle making the image bigger or smaller um you know we've sort of designed the system so that you don't have to think about the layout it just sort of works um then you click it into a format that makes sense so you know image aligned right text left image on top text below it basically is a bit more intelligent than just randomly drop it and make it look ugly it's going to snap it into some sort of logic and is that ai doing that or is that just you uh doing some a basic decision tree it's a It's a good question.

15:44So this is all a decision tree with heuristics. But the next thing we realized with our growth is that we have millions of these being created each week. So now we're going through the process of labeling all of these these anonymized layouts. And, you know, very soon we're going to be able to to use machine learning to generate the perfect layout, depending on your content. um so that's sort of one piece and then you know when building your telescope company you might want to add a uh prototype to to to show like what's this light so i can uh for example i'll drag this web tile that you know terms based on this tile system into the page and um let me grab let me grab a like a 3d model you know so i just went to another tab i opened up this product called spline a lot of folks use for for 3d work and then i can go back here and just paste this prototype you know and it knows that format and it makes a nice clean embed yeah exactly exactly and we we play with it so they don't flicker and refresh and you know all of the things that you don't like about about embeds um so that's that's what it's like to to construct a story and i'll show you one more thing if uh if i if you're up for it jason yeah yeah please um so one of the most common workflows that that our customers have is you know they're like i've already written what i want to say i've written the story i want to tell um in a document um so can you ingest that document and turn it into a presentation instead of using you know the corpus of content that gpt is trained on um so i could take a book i could take a pdf a report a deal memo a long email a notion page whatever it is describing our project we're gonna we're gonna do our uh you know parents 50th wedding anniversary and we've made a bunch of uh items and have subheads etc i can just upload that and have it go to work exactly exactly makes sense I'll grab a notion doc I have around of this is about nuclear energy.

18:00You know, it's sort of a sort of heavy piece, you know, and you can tell it's pretty unstructured. It's all sort of like narrative prose. And I mean, it's 20 power dense paragraphs. It has no headings except for the title. So you're dealing with something that is not structured. Exactly. We're not highly structured. Yeah, exactly. I'm going to copy that. go uh go back into tome open this command menu on the right and hit document to presentation and um now i'm just going to paste this here uh you know and it sort of keeps the the structure that's there and now i'm going to hit the generate button here and we actually read the whole document understand it break it up into parts and then summarize um you know the contents on each page and uh you know this was this was possible before but so honestly with the huge context window in gbt4 that sort of came out in the past few weeks um we were able to make this really good really fast and so when you look at this you know we we had a bunch of paragraphs without subheads and now it's made uh it looks like an eight or nine tights not eight to nine slide deck but it's really said hey here's the importance of nuclear energy research team and presentation electricity sector and cost implications recommendation for construction technology etc etc yeah let's figure it out what each paragraph is about or so and then did this now you said this is some contextual uh feature of gpt4 can you explain what that is and and what it enables yeah yeah so um i would say with the some of the older foundation models um you had this context window where you could share, you know, something about the document or something about what you're trying to create as well as the prompt.

19:51And the sort of problem with the language models from last year was that you could only maybe put three or four pages in there. So we, you know, we would need to be really creative for your 25 page memo, right? Where we would need to summarize it and chop it up before we could understand the whole document but um you know with uh with gpt4 you know we can read 25 pages at a time um so now we just have like this much deeper understanding of what you're trying to say so we can capture the the essence of it and break it up so here's a question for you um could you take another aesthetic an unnamed aesthetic and then give it to your software and say hey and i just did one i just did like a cyberpunk you know tokyo kind of situation and if i said here's the url of this cyberpunk there it is here's a pinterest board um which is you know people create mood boards on pinterest so here's a mood board it's cyberpunk japanese neon kind of aesthetic you can you can imagine it yourself uh which is how ridley scott sort of imagined los angeles people would be speaking some combination of chinese and spanish i think they called it changlish uh was the what you hear some of the police officers speaking in blade runner so if i gave you that are you at the point where you could say just try this style and here's a url to get inspired by or is that next i would say that that is next um it's exciting because it's possible now right to feel like here's a board here's an image what's the style what's in it and now let's create a prompt to create more and i would say we're months away from that not years well i mean and so if i said write it in the style of michael lewis or malcolm gladwell could you do that yet or is that another kind of concept coming in the future i would say that we're very close with with some characters um which is to say that you know we could rewrite it uh in the tone of elon musk uh because there's just enough content from him that have trained these foundation models exactly style of elon was could get dangerous quick yeah for sure i mean it raises some some interesting questions of you know should we even let people do that right well yeah i mean it'll be exciting i mean you could also just take a i mean i assume you could just take a video a youtube video of sam harris giving a talk and say you know well or christopher hitchens and say you know i want this to be clever like hitch or i want it to be more somber like sam harris or i want it to be more enthusiastic and wacky like jordan peterson it was kind of interesting to apply styles and it'll be super interesting when it's not a drop down of the styles it's you know give us three examples of style and let us go from there here i want to write like stephen king you know i want to write like ernest hemingway and that to me is really interesting same article but you know using different styles.

23:09Are you trying to build the next Uber or Robin Hood? I hope so. Well, here are three important tips for you. First, make sure your data is protected. Obviously, remember, just because your data is in the cloud doesn't mean it's automatically backed up. Your data is your responsibility. You got to make sure it's backed up. Second, don't let backups and compliance requirements distract you. Let your engineers focus on product innovation, not writing scripts for compliance audits. And when it comes to backups, the major players have major limitations like snapshots. If your account is compromised, your backups are compromised.

23:41But Clumio stores your backups on a different server. So if your main account is compromised, your backups are always safe the way it should be. And third, take control of your cloud costs, you'd be surprised how much of your storage costs come from backups, snapshots, versioning and replication, or maybe you're maybe you wouldn't be surprised because you've seen these jaw dropping bills. Well, Clumio can help you with all three of these important points. They provide turnkey data protection that is air-gapped, immutable and cost optimized. Clumio has saved customers over 30 % on backup costs while putting security and compliance on autopilot.

24:16Visit them at clumio.com slash twist to start a free backup or sign up for a demo. That's clumio.com slash twist. How do you think about training data sets? Do you believe you provide the data sets or you empower users to do it where's this all going to wind up in the in the coming weeks months here yeah it's a good question so so i i think um you know we're we're both a consumer well we're we're both a consumer and an enterprise company and i imagine you know for our consumer offering we're going to continue to use these sort of large base models um and and you can get pretty far with them You can sort of get stuff that feels sort of globally compelling.

25:02But the most common ask that we get from our customers that are using Tome for work is, wait a minute, I want to train these models on what's in my Google Drive. You know, I want to train these models on what, you know, what's written, the data in our data warehouse. And I want you to use the facts that we hold to be true. Do you have customers doing that yet? have you have you because it feels this feels very single player mode not enterprise mode yet but have you gotten to the multiplayer multi-department multi-international company mode where you know a google says hey here's our style guide or apple probably be more evocative example but here's the apple style guide here's examples of you know i mean perhaps source canonical like just the the here's the origin texts of steve jobs and how he speaks and how our copy editors write and just try to keep it towards this aesthetic and this type of copy has an enterprise done that yet yeah so we're working with a couple of larger companies but we're in the very early phase of sort of doing this by hand um actually a very common ask they've had is you know when you're in a giant company uh the ceo or the leaders like to be presented to at a certain way and and and that sort of gets an outcome like when i worked at facebook there was definitely a way to write a memo so that mark would read it and then think about tell me about that unpack that what would it what was what was mark's memo style oh it was uh it was definitely sort of terse um factual you know he wanted to see the data up front and was sort of looking for this really pragmatic recommendation and um and it was crazy because uh you know that's sort of tribal knowledge right you only know that if you were high up at meta right and you had spent a bunch of time with mark and the sort of promise with where ai is going wait a minute we can take this thing that only only mark's direct reports know and sort of make it available to anyone at a company who has an idea to share yeah it's fascinating uh you know there is an entire there entire books about netflix i don't know i've had patty mccord on this podcast um and there are two just really great books about netflix's culture as just but one example and they have a culture deck and then they have no rules rules um which um erin mayer did she was also on the program haven't had reed hastings yet but patty mccord had hers and uh then reed hastings did one this no rules rules netflix and the culture of um reinvention so you take those two books you take some interviews with those individuals like having patty or erin on this podcast you start to put that together with the culture deck and man you got a really interesting internal ai base then you take bezos's 20 letters that he's written over time and you can really start to understand some things about business i think um that are particular to the organization that's kind of brilliant and proprietary and people don't want to share that so i think if i was gonna man if i was an investor in your company which seems like i missed the window uh how much is the company worth now you guys raise a lot of money is this ai thing you probably were toiling away telling investors you're doing a presentation software layer and they ignored you they added ai to it and now they must be knocking your door down at south park what's going on with the uh tell me what it's like to add the the ai pixie dust in 2022 and 2023 to your product yeah it's a it's a good question you know we um we were building this for years and uh we sort of were selling people on um building better collaboration and editing software and it's the sort of thing where um you know if you play with tome if you're in it for hours you start to recognize the value of the format um and all of the non-ai parts of it but we know we had that the hardest time selling the average person on, hey, this is a better way of sharing ideas versus using PowerPoint or using Google Slides.

29:08And when we added that AI component, it just became so clear to everyone, investors, customers, students. We have a ton of students using Tom. Wait a minute, this thing is different. And in the past few months, we sort of went from a couple thousand folks using tome to uh you know a little over three million um and and growing and you know as a result uh we've had a lot of investors uh sort of interested and uh you know over the holidays uh as we were growing and we sort of had this this hockey stick moment we were we were lucky enough to be approached by some great folks to to get our series b done amazing 300 million dollar valuation for 40 million dollars raised with just a handful of paying customers it's an amazing time how do you deploy that capital intelligently and get from wild enthusiasm and sky-high valuation to filling in that valuation because you need to have for that valuation to work as a public market company you know you're going to get a 10x multiple not a whatever multiple 3 ,300 whatever you wound up getting here how do you fill that in it's a good question um i mean one of the things that's really exciting about the space is if you truly have a 10x better product uh the internet just finds it you know which is to say that you know we grow a ton from tiktok instagram um twitter people sharing terms on the internet so a lot of our work right now is just deploying that capital for the product to catch up you know just to say that we're uh we're building a real ai team we just made made our first head of ai higher um we're we're tuning models we're sort of moving on to dedicated gpus so we can be blazing fast and reliable and um starting to scratch at the surface now for this this really differentiated enterprise product you know um a couple of years ago uh when i was incubating this at graylock uh uh reed asked me to uh present tome to a bunch of cios you know and at first they were like oh great another productivity tool that we're gonna have to pay for and um you know over our dead bodies uh that we're excited about this but now that the landscape has changed you know i think those buyers and and and those company leaders like wait a minute i can see how i could save hours and hours of my team's time here and they're way more open to an enterprise product now than maybe they ever were so we're just trying to build the product catch up and sort of discover this enterprise motion on top yeah um talent are you able to find talent uh or do you have to go uh make talent at this point because there's so many people who are looking for mlai there just can't possibly be enough people out there who've worked on this stuff uh it's kind of like asking people you know at the beginning of bitcoin you know i need somebody with 10 years in crypto and it's like well crypto has been around for 18 months so generative ai has been around for a couple of years in terms of you know beyond the early 100 200 people working on it so how do you how do you staff something like this in a market like this yeah it's uh it's gotten a lot easier yeah well just to say um you know i think about when we started the company in 2020 and uh there were an abundance of startups uh everyone was sort of connected to their golden handcuffs in the in the public companies and um you know it was hard i think that one of the couple of things that have changed are that everyone's public stock comp in these in these giant companies has come back down to earth and um the other sort of exciting thing is that you know to really be inventing in this space you want to be at a startup you want to be at a place that's willing to to do the imperfect thing that doesn't have managers of managers of uh you know ethics committees uh so we're getting so many folks coming from google and meta being like hey i want to build you know i want to i want to build and it seems like you all are moving at light speed compared to to what i'm used to was it like that your your last couple years there at the uh at the facebook corporation was it you know like mark is saying in his memos now hey we got to get back to pushing product and we got to get rid of managers managing managers managers what were you in the thick of that managers managing managers conundrum that he seems to have fixed now yeah i mean it was sort of happening uh it was they were boiling the frog slowly right which is to say that uh during my early years there uh it was all about creating value and speed and you know i remember this uh this poster that said code wins arguments uh and it was all about you know being action oriented and getting stuff out and then at some point um i think it became more about protecting what we had.

34:16And that's when, you know, there were managers and process and, and all of these safeguards put into place. And that sort of slowed things down and slowed things down to the point where it just became, you know, really hard to create anything new, or sort of move against the grain. Listen, as a founder, ensuring your cash is safe is priority number one. It's been a bit crazy out there. I don't want to put caps locks on right now, but I will if you need me to. The FDIC$250 ,000 limit is just not enough for most businesses. So let me tell you about Mercury. Through its partner banks and sweep networks, Mercury customers can access up to 5 million in FDIC insurance.

35:01That's 20 times the per bank limit. Sounds a lot safer, doesn't it? Well, these sweep networks protect your deposits by spreading them across multiple banks, which limits the risk of any single point of failure. And with Mercury Vault, any funds above the FDIC limit can easily be invested in a money market fund, mostly comprised of US government backed securities. So it's easy to get started with opening an account, you can apply in minutes, and many customers are approved and onboarded in less than two hours. Mercury also offers great resources for founders, including my favorite Mercury Rays.

35:37This program connects founders with investors to help them raise funds head to mercury.com to join more than a hundred thousand startups mercury is a financial technology company not a bank banking services are provided by choice financial group and evolve bank and trust members fdic amazing uh well listen continued success with this it's really impressive already and the pace is getting faster and faster uh as we wrap here do you have concerns about how fast this is moving uh we were talking about it just on all in yesterday and you know people are creating auto gpts uh to go do essentially like cron jobs to be out there prompting other ais and prompting each other and adding items to task lists and that is for people who are listening you know imagine you asked to chat imagine you asked you know this product hey make me a presentation and then it said and then give me the objections that people would have to funding the startup and then put those objections answer those objections in the deck and then present it to you know venture capitalists like these and then you know answer those objections and it just keeps going on forever you know doing these kind of prompts and making various versions obviously can be used for nefarious purposes too you could be doing exploits and you know trying to have it attack vulnerabilities and phishing or human factors engineering uh lots of different weird things it could start doing so are you worried about this stuff and the pace that's moving that hey move fast and break things might might break too many things or do you think it's a little farcical to get that concerned um i would say that there's a lot to be concerned about um and uh you know there's a wide variety of things i mean one we didn't even get to hallucination right and explain that to the audience um because that's super important yeah for sure i would say the but the simplest way to talk about how these fun these large language models exist is they've read a ton of content and now they're trying to predict the next word or the next sentence um relative to what's been written and um you can imagine when you're just trying to predict the next word or the next sentence or the next concept that sometimes it just makes up something that you know isn't really grounded in in facts um and uh you know we're working on a bunch of clever ways to work through that but for example we've uh we've built this classifier that understands if something's a claim you know so we understand if gpt is saying like the sky is green Well, that's a claim.

38:26Where on the internet does someone credible say that the sky is green? And we're not connected it to a web crawler and being able to say, well, here's an article that says the sky is green. It's sort of up to you if you've decided that that's like a credible fact. um so so that that's sort of a you know one area of things we're working on um i think i think to your point um we're helping people produce content really fast compelling content really fast right and uh the sort of downside of it is that there's going to be a lot of compelling sounding content being created for uh you know maybe ideas that that aren't that compelling or that that are dangerous um right and um i literally did this today it's interesting you bring it up keith i i literally was having this debate on all in and the episode 124 comes out today uh or maybe will have come out when this comes out a couple days earlier but i i was uh wondering like hey people could create websites for phishing then have auto gpt running phishing experiments where you're trying to get people to give their passwords into fake accounts and i am like reticent to even start brainstorming these things because it doesn't seem super complicated to do but um it does seem like a lot of these new language models can be connected through plugins to actually create websites to actually go shoot off emails to actually maybe do an a b test make a hundred different variants of emails and see which ones you know are the most effective at getting people to click then which ones are most effective at getting people to log in then which ones are get people to you know change their password and yeah all of a sudden you you know it might come up with ideas that are just crazy and i just asked it what's the best way you know to trick a receptionist into giving her passwords like oh yeah i can't do that and i was like i'm trying to protect my receptionist from giving away her password what are ways in which people would try to trick her that i can and it gave me like a roadmap of how people come in it's like oh yeah you know people would come in dressed as an it person uh knowing other people's names have paperwork and then uh try to trick them to let her use her thing to change her password because there's been a security violation i'm like thanks for the roadmap chat jpt i never thought of like you know coming in in an outfit like a cia agent might and saying hey your computer's been compromised joe schmoe at the home office asked me to come here and i was like yeah pretty obvious but i mean you can trick these models so easily yeah for sure um and and i think we have a lot of catching up to do you know we're just to say that um we need to think even more critically about what we read uh what's out there on the internet where did this come from should i believe it and you know this was true of even social media right with sort of explosion of content and we need to be more discerning and then certainly with our our ranking and security tools we have a long way to go with wait a minute who wrote this email where did this email come from where did this seo article come from um so uh i think it's going to put a lot of strain on us as people of wait there's there's all of these things that are coming up and we're gonna have to catch up and think about how to how to adjust into the new world so so a lot to talk about yeah all right listen keith uh where can people learn more and what jobs you're hiring for and what is the url of your career page because you secured the bag and now you got to build into this valuation i know you need to get some people some technical people listen to this podcast so maybe we can help you fill in some jobs oh um thank you yeah uh so uh we're tom so it's t-o-m-e dot app uh and you can go to t-o-m-e dot app slash careers and uh we are hiring you know incredible people from every technical background you know we're hiring ml engineers we're hiring full stack engineers we're hiring product designers a lot of designers you like in person well what is your preference these days you know to be honest we prefer in person and how's that go i see me too i i am so lonely that i decided i'm getting an office in san mateo and i'm just gonna hang out there and have entrepreneurs come over and start doing in-person interviews again and just start hanging out because this would have been so much fun if we got to meet and get a cup of coffee before or after yeah yeah for sure and i live in tell me why you like it what do you like about the in-person what's happening in person that's not happening over zoom no offense to zoom great problem yeah yeah i would say a couple of things you know one we're we're a creative company and um so much of what we build is because we get in a room and talk about is this the right thing to build and are people gonna love it and that's so much easier when you're sharing a table, when you can get on a whiteboard, when you can look into each other's eyes.

43:19And, you know, maybe when, after you've had a big intellectual debate or fight, you can, you can grab a meal after and sort of talk it over. Um, and we found that, you know, we, we have a couple of great remote employees, but it's just so easy to get mad at the person who you're only seeing for zoom or to misunderstand what they're really trying to say. and we figured you know we're a 40 person company versus teams of like 10 000 working on this at google and microsoft we just have to take every advantage we can we can get yeah i think in person is going to make a major difference uh for the startups that can get that going now you lose the fact that you might have somebody who lives uh you know in lake tahoe that refuses to come an office and they're just a 10x engineer but there's something weird happening let's end on this one are you finding that the prevalence of 10x engineers is increasing because of uh you know chat gpt4 and other tools you know copilot github copilot whatever to make people a little bit more bionic uh this is something i'm starting to hear from my portfolio companies is like you know developers are getting faster and it's not just the 10xers We got people who were rank and file engineers who are now starting to trend towards 10x.

44:40Maybe they're at 3x, you know? Yeah, I would say I'm noticing this in a couple of ways. One, it's really easy to ramp up on a new language or new style of doing things. You're using Replit, you're using GitHub Copilot, you're reaching out to people in communities. So that's certainly true. Another part of that is all of the technology we're using is brand new. so it's really your ability to ramp up and keep you know keep your ear on the ground that matters versus you know having been a classically trained research scientist from from 10 years ago uh so that's sort of shifting things and certainly if you're resourceful and you're connected to a lot of people you'll do better uh and maybe the last thing i would say is that when i think about a 10x employee at home it's not just the amount of code they write it's their ability to work with everyone around them and um you're sort of leaving a huge advantage on the table by being a remote only employee when it comes to sort of collaborative productivity uh so i'd say those are sort of the things that i've noticed are changing in the past few months and you were working at meta when they went remote during covid and you worked there before uh it was remote you experienced both of those things yeah and it it changed didn't it yeah it changed so much it changed so much I think when you're an early stage startup, and I shouldn't speak in generality, when you're our sort of early stage startup, it's not like everyone is clear swim lanes and clear milestones and you can just shut off slack and work for two weeks and come back.

46:18we're sort of figuring it out as we go along we're getting so many new users or the technology's changing we're getting new feedback um so so i i think sort of giving up on that high bandwidth communication uh just felt like the wrong trade-off for us yeah i agree all right listen everybody go try this product mind-blowing number two go work for the company and uh number three yeah maybe you got to think about getting your talented people in a room and start working together a couple days a week at a minimum. All right, everybody, we'll see you next time on This Week in Startups. Thanks for coming on, Keith.

46:50Appreciate you. Thanks, Jason. Okay, thanks so much, Keith, for joining the show. Next up, we have another awesome talk from Founder University for you. This is the talk on how to get started with paid acquisition. It's a great episode. And what you're going to want to do is just do a search in your podcast player or YouTube and search for Founder University so you can find the rest of these tactical talks we've done. So next up, paid acquisition. Listen and learn. Hi, my name is Jen and I'm the co-founder and CMO of Velocity Growth. I specialize in paid advertising and have helped companies across Europe and the US grow their business online.

47:29Today, I really want to give you the high-level strategy thinking you need to position your thoughts towards in order to really grow your business through paid ads. A huge portion of this is actually just making sure you know your customer inside out. But working off the assumption that you've already taken the time to really listen to your customer, here's where to start. So this is my favorite phrase that I like to kick off every presentation on advertising with. People have the misconception that advertising is intrusive. But have you ever hated your favorite brand for showing you an ad for your favorite product?

48:06Probably not. Have you ever hated the ad for that game you'll never play? Absolutely. Why? Because relevancy matters. People don't hate adverts. They hate irrelevant interruptions. So what is the trick to resonating with your audience and actually growing your business in a cost-effective way through marketing? That's what we're going to dig into. There's multiple components to relevancy. The core of advertising is to get in front of the right people at the right place at the right time with the right message. But at different stages of your marketing, you don't need to get all of these right.

48:44But these are the key points you have to touch on in order to reach the conversion point. But you might be bringing in awareness. And when you're looking at growing the awareness of your brand, you're really focused on two things, hitting the right people with the right message. You're in the planting a seed phase, your focus is finding your people and hitting them with a message that helps them remember you. And then as you move them down the funnel, you want to start thinking about hitting them at the right place, where they encounter the problem that your solution fits, for example, or in the right moment of intent, which we'll talk about in a second.

49:27When you hit people at the right place, they start considering you. They're starting to actively think about you. It might not be the right time to convert, but they're on the right track. And that moves all the way down into a customer converting, which is hitting someone at the right place, time, and with the right message. Hitting the nail on the head across the board. This is where you find people. Yes, that might be an impulse buy on social media because you happen to catch them where they had a moment of time to actually do a transaction, the impulse purchased, or they might actually have seeked out your solution.

50:09But at this point, this is all you're looking for. The right people, right place, right time, and the right message. But before we can start picking a platform to advertise on, there's a big question you need to ask yourself first. What is your problem solution fit scenario? If you have a solution to an existing problem, people are going to search for you, right? That's where you can start thinking about search platforms like Google Ads. If you have a new solution that nobody has heard of, people are not going to search for your solution, right? They're not aware it exists. They don't even know to consider it.

50:48That's where you need to start thinking about social platforms more. And if you're planning on making people aware of a problem they don't know they have, that is going to be an expensive start to paid advertising. And you have to factor that in if you're making people aware or creating a new problem that you have a solution to slot in for. But your problem solution fit scenario really helps you decide how to deliver your content across different platforms. Once you know where you're at with your problem solution fit scenario, it's time to start thinking about your budget and where your customers are and what stage of the funnel they're at.

51:28Different platforms have different levels of intent and cost as a result of that. Let's say you're a known solution for a known problem and you need to start getting customers fast. Google Ads makes most sense, right? What if you're a new solution to a problem that nobody has heard of? Building out on social makes more sense. This matrix is a generic one I have put together, but depending on your audience, all of this might be different. If you had a crypto startup, Twitter might be a much more intentional an audience than Snapchat, for example. But you get the idea. This part of the puzzle is all about getting you thinking about the right place and time.

52:09Once you've selected a platform, it's worth noting that privacy changes have made a significant impact to the data that you can use on those platforms. Most prevalent to that is meta. So we have to get a little bit more strategic when it comes to being able to find the right people at the right time, in the right place, with the right message. For those who don't know, advertisers use a cookie tracking system to monitor your online activity and then target your adverts accordingly. But Apple essentially stopped that from happening by asking users if they would like to opt out of that tracking.

52:45This means your messaging and creative is more important than ever before. But also, I would encourage you to look at server-based tracking rather than cookie-based where possible and leverage what's called UTM parameters where you can so that that can support your reporting. Now that you know how to think about your paid ads and where you can dig in in terms of a platform to grow your business, here are some of my favorite little tips that you can look into. Even if you haven't started advertising, these are good thought exercises so that you can think about just how much you can leverage the different platforms.

53:22Firstly, we know that creatives are more important than ever. So you need to understand what types of ads and creatives are performing well. I would recommend checking out the Facebook ads library and TikTok ads library so that you can see what's happening in your industry and with your competitors in terms of their creatives. You can search for companies, search across your industry and see what are the high performing creatives in that space. TikTok is a great resource right now because high performing creatives in TikTok are difficult to achieve. So if you come across them there, you know that potentially that's going to work very well across Instagram as well, for example.

54:04Google, as of the last few weeks have also launched their ad transparency network. So you can force a view of another company's ads to come up. And how you do this is maybe search for the company name or a keyword that would force an ad to come up. The key to this is forcing an ad to come up. Unfortunately, there's no site available to just view these ads. So once you have your ad up for the company's ads you wish to see, you can click the little downward arrow next to the ad and click see all ads. Now you can see all the ads they have running on Google Ads too. And you can look at that across region as well, display, video, see whatever they're running.

54:44Too often we think about competitiveness in the ad space. But a great additional tip to know is that you can also share your audiences with other people on Meta and vice versa. So this is a great opportunity to collaborate with businesses that have a similar audience, but not competing offers. Take, for example, a chef and a fitness instructor. One might be selling cookbooks, the other fitness classes. Their audience may well very much overlap and they could collaborate on some of their audiences and data so that they can share. As mentioned, the privacy changes are making a big impact, but there's actually a ton of data you can leverage that has not been impacted.

55:29So you can disrupt the data a little bit and manipulate it to work for you. An example of this is on-platform data. The screenshot here shows the difference between your sources of data and meta sources as an example. Essentially, only things that happen away from meta, such as on your website and add to cart is a good example. Maybe someone purchased and you don't want them to see ads anymore. That data is what's impacted. Actions that happen off of meta. But you can still leverage the data that is available on the platform like audiences of people who watch the video, engage with your page, Instagram followers, all of the actions that happen on the platform.

56:16A great tip here, create a long form video, roll it out, and then retarget to people who watch 75 % or more of the video and use that as your leading indicator for intent for retargeting. If someone goes to your website, goes to a product page, then adds to cart, and you set up a remarketing audience of everyone who added to cart but didn't purchase, that data is what's going to be heavily disrupted. That's what's made remarketing audiences very difficult to continue using. even though they still perform well they're not the perfect science that they once were. Lastly Google is often overlooked as an opportunity to target audiences but you can create audiences there too.