All right. Good afternoon, everyone. Such an honor to be here. We've got a great panel. Our partner in crime, Shane, is going to come find us here any minute. There he is. I see him walking up rock star on stage. We've got an incredible, incredible group here. My name is Sameer Dholakia. I'm a venture capitalist focused on growth over at Bessemer venture partners. I've been in software now for 28 years and just recently made the switch over, been a CEO of a couple different companies, but just made the switch over to venture. And having been in this industry now for 28 years, I'll tell you without question in my mind that we are in one of those moments where we see a tactical leap forward that is profound, that will change the way that our software industry certainly works and but more importantly, changes the lives of the billions of humans that are online on this planet. And I think it's a really exciting moment for us all to be here, to get to partake in this. And I'm very honored to be up here on stage with these four who are on the front lines of helping shape it at their incredible companies. So I'm just going to ask them to go down the line, do a quick introduction of who they are, the company they serve and work for, and what it is that they're doing in their corner of this remarkable landscape. So, Shane, want to kick us off? Hi, everybody. I'm Shane Ehrlich. I'm the president here at Jasper. And so, yeah, what are we doing? I mean, for us, we feel like this movement is such a massive responsibility that we wanted to make it about all of us and this whole community that's really pushing what I can do and how we think about it and how it's going to change the world. So I think I'm most excited about the people that showed up for this event. You know, again, we thought it'd be this small thing and all of a sudden, you know, over 1,000 people and tickets are sold out. And I'm also really excited about the partnerships that we're forming. Like in a traditional software vertical. You would think our software industry would think of us as kind of more competitors than partners, but there really aren't any competitors in this space. We're all sharing information, we're all sharing our learnings, we're all sharing the models, the customers, like constantly, all day. And so I'm most excited about just how fast things are changing and how passionate the community is around using AI to deliver good things. Amen Amen. Everyone, remember what Stockholm founder and CEO stability AI. Our mission is to build a foundation to activate humanity's potential. And we do that by giving away our models for free open source. So we build the most popular open source models. And my aim is to build that to billions of people, because it's amazing how you can extend the capabilities with this. I'm John. Hello my name is Amjad Massad. I'm CEO and co-founder of rapid upload is the fastest way to make software. Our mission is to bring the next billion software developers online. Billion kind of sounded crazy when we started talking about it, but people can imagine now, given the power of AI and how you can sort of create a piece of software by just talking about it, that it's actually possible to have a billion people creating software. And so we're really injecting AI into every aspect of making software. Your software is going from this pretty industrial age type of business where you had to have this very large setups and people would spend a lot of money on these software systems to create, create software. But I think with AI, it's going to get way, way more accessible and in the same way that, you know, design or marketing copy is getting accessible, I think software is going to follow the same trend. Dario hi, I'm Dario day. I'm CEO of isentropic. Anthropic is no. One company with a special focus on safety and interpretability. We built a model called clod that in a recent post by a really good site at scale. I was called both safer and more fun to play with than chat. We're now working to offer that offer that model to customers in a wide variety of use cases. I've had kind of a front seat to the language model stuff for the last few years. I was at open the eye for five years, led the teams that did GPT 2. And GPT 3. And it's exciting to watch, watch the evolution of all of this. And, you know, I'm excited for the next phase. Awesome fantastic. Well, and to build off of that, I think the notion of the app, the applications that sit on top of these crazy, powerful, large language models have certainly opened up the general consciousness to what could be possible. It's why it's on mainstream media. The kickoff video that you all saw this morning, when you've got, you know, Trevor Noah talking about a.I., you know, we've made it to the mainstream. And I think it is because, you know, these large language models in this research have been going on for many, many years. GPT 2. And three been around for four years now. But it was only when GPT and the model in the app, the chat bots built on top of isentropic and that all of a sudden users can get access to it. What we've done with Jasper. You know, when you put it in front of an end user, all of a sudden it hits the mainstream consciousness in a way that it simply hadn't before. It was the grounds of researchers and scientists, and now it is in the hands of everyone. And so a lot of what we want to do on this panel today is talk about how we get the power of AI into the hands of users and businesses. I thought actually we would start, Dario, with you since you have been working from ground 0 at the link large language model layer at the foundation layer and maybe just share a little bit of what surprised you as you all were building these models in the early days and how it evolved in ways that maybe you didn't predict? Yeah Yeah. So I mean, you know, there there are many moments where something interesting happens. But what, really what really stood out to me is, you know, the moment when we were building GPT 3 and I think I had one realization that made me realize both on the technical side and the application side, how big this was going to be and really the nature of it, which is, you know, as we were training GPT three, we discovered as we were trying out the model that it had many capabilities that we never really intended. We just threw in a large data set and as we were training it, we're like, oh, this thing can speak Italian. It can, it can, you know, it can code in Python. We didn't even know it could make websites in JavaScript. We didn't even figure that out. That had to be figured out. After it was released, someone posted out on Twitter like, hey, you can make no code websites with not perfectly, but you can do it. And so, you know, I've been in many fields. I've been a physicist, I've been a neuroscientist. I've never seen a technology that is produced to such precise spec. On one hand, where you put all this data into it, you know, you have to have thousands of GPUs. IT costs 10 million, maybe $100 million to train. And yet once you've trained it, you don't necessarily know everything it can do. You have to deploy it to a million people before you discover some of the things that it can do. And so that's very exciting from the perspective of there are many applications I think still of chat, GPT, quad, the next generation models that are, that are, that are just waiting to create value that are that, that, you know, have not been tried yet or we don't know that it can be done. On the other hand, there's a concern that, hey, I can make a model that's very good at like cyber attack or something and not even know that I've made that. So it's this kind of duality that's, you know, exciting, exciting and a little scary. Yeah and hopefully a lot of exciting. I hope there are a lot of entrepreneurs out there in the audience today that are thinking about exactly what you described, where all those white spaces where we can apply this technology, solve real problems for real users and real businesses. I'd love to hear I'm just building off of the head innovation code inside a can. And we were talking backstage about how there are a lot of businesses out there that are starting to use this. It always starts with the early adopters, but enterprises eventually realize, well, I better come up with my what is my strategy? Every c-suite on the planet of every large company is being asked, hey, what is our generative AI strategy? And we were talking about how, you know, if you're a Goldman Sachs or Morgan Stanley or fill in the blank, what some of these application development groups are larger than 98% of software companies. They have more software developers on their teams than most software companies. So if you were advising a Fortune 500 CIO that might be in the audience or listening in, what would you advise them to do? As it relates to use of generative AI for software development, there are fundamentally two ways this technology is going to change software. One way it's AI assisted. Programming, you know, popularized by co-pilot and now replica Ghost Rider. And that's going to make existing developers a lot more productive. We think that developers going to get 10x to 100 tracks more productive. But it is a rising tide. And basically people who do not know how to code will be able to generate software, will be able to generate some amount of code, will be able to learn how to code by talking to the AI. And so, you know, you will be able to go in something replicate and be able to generate code, iterate on the code and learn as you're going. So basically non software developers become software developers, expert software developers become super expert software developers. Like, I think that productivity gains are just going to be insane. Right and so that's one aspect where just like everyone at the organization will be able to contribute to the creation of software. If you talk to any business leader today, the main skill that they're missing is a way to create more software. And software is one of those things where the more you have of it, the more you want of it right? And the more software creates more problems. And these problems need to get solved, and that begets more, more software. Then the other way this technology is changing software is itself is software. So, you know, a sort of a large language model like Lord or GPT 3 could be prompted to be programmed to do any specific task, right? So it can become a chat bot, it can become a marketing copy generator. It can become like an image generator. It can become a search tool like Bing. It's what's called foundational model because it can be programmed using English language or whatever language you want to use it in to generate an application specific AI. Right so that means that the act of making new software gets a lot easier because you go from 1,000 lines code base. So let's say you want to create a parser. Let's say you have some kind of unstructured data. You want to scrape a website. It used to take weeks of a team of engineers to be able to sort of parse unstructured data or scrape something from the web. Now, you can prompt you pretty to understand the structure website and pass out the information that you need. So again, not only making software gets easier, but the software itself gets a lot more efficient. Yeah and and so, you know, we ruppert, we started this new gig platform. We call it bounties and you know, the, you know, at first we thought, yes, you know, this is going to be, you know, all our users going to be writing code using ghostwriter or they're going to be a lot more effective. But the surprising part is a lot of the tasks were accomplished using some amount of right instead of writing user code base, you just call into a model. So a model becomes part of the call graph of software. It's almost like a function, like a function call. Fascinating love it. And by the way, on the topic of this massive productivity enhancer, what percentage of developers in the world today. Do you think use something, use generative ai? And what do you think that is like? If you were to forecast out next year, how steep is the curve? Do you think of adoption of generative AI as a tool in the toolkit of developers? It's actually surprising how fast the adoption speed. There are some reports out there about Stack Overflow traffic dropping month over month, 15% in the month of December, the month that chatty came out. So that means developers are going to chat to get answers instead of going to the open web. Right interesting. And we can have a discussion about all the problems that brings. But the adoption has been like nothing we've ever seen. The adoption of Ghost Rider and replica has been crazier than any other product we've launched. So I think this is what makes this moment super special. It's for the first time, consumers, developers, enterprises are moving at record speeds to adopt this technology. And what do you think the applications are for? There are from my background it Twilio and SendGrid. And I remember we used to quote a lot. There are about 20 million developers on the planet as of today. I love your goal of we're going to put this in the hands of everybody. I have a billion developers, but there are 20 million developers in the world. Not all of them are working on the most complex architects. Whatever they're doing, lower complexity code. If you're the CEO of a BPO that's working on like simpler work, should you be worried, like what percentage of those developers will have their jobs automated? Well, the conversation that Kevin was just having. Yeah, we don't really. I have to speculate. I mean, if you look at the design world, for example, right, you go from Adobe, this complex sort of piece of software. They did a great job with it, but it's very complex, requires four years of college and some training. And then you had Figma Canva and you have an explosion of design tools. Now, everyone in the world is a designer. Yeah, every person in a company is in Figma. You know, whether you're a product manager, product engineer, and so I mean, figma, I think they've released this pie chart of their users. Only 40% of their users are professional designers. So I think perhaps only a third of people interacting with code in any given company will be professional software developers. But I think every part of the company, I think, you know, we're seeing that with Jasper as well. Every part of the company will be able to interact with the active creation along with the active I love it software. So a terrific use getting a great use case into the hands of the citizen developer all over the world and all functions in business. Ammad, talk a little bit about your world in using AI for imagery, the kinds of users you have using it, things that people would expect and maybe some customers. Examples of things that you were surprised by, people that are using this in a business context that maybe, maybe they'd be enlightened to hear about. Yeah so we release all types of models, but I think the most famous one is August released, a model called stable diffusion, where we took 100,000 gigabytes of images and create a 2 gigabyte file that you type a few words and boom, it generates any type of image. Originally in 6 seconds, the latest version can do 40 images a second, which is a bit insane. Wow I think it'll run on an iPhone in 3 seconds. This was a bit of a game changer because it was fast enough, cheap enough, and good enough to go everywhere. Just, I think, integrated it in 3 days. They're like, that's kind of cool. Pretty extraordinary. And again, it's this ability to take it removes the barriers to creation. I think that's the thing. These technologies understand our intents and I think you'll see the best technologies and implementations on the contents better and better. It's like go straight to understands what you want to do over time. Just we're going to building in these feedback loops that bringing in the brand voice and these other things as well. And we've seen people take stable diffusion to insane places from like someone took it and put in song spectrograms, riff fusion and then you can type in, I want a jazz song from New Orleans, but with rap and it generates the image thing again, the sound wave, and that generates music. Wow which no one expected against, I think. You know, Dario, we were discussing earlier, like if you look at the word surprising in the papers, it probably looks like that. Yeah, these models can do just about anything. I think another example. We work with a number of film studios and other things. There was a scene in a movie that's about to come out where they had to build a book full of lots of pictures of women in the seventies, and it would have cost $350,000, sagged daily rates, three days of shooting, movie like Anna Kendrick and kind of others. One of our leads on our film thing basically cut it up and integration of various components. He did it overnight for $60 and saved 350,000. Real problems real, real early. And then we released it open source and it became the foundation of Lanza and all those avatar. The things that you see on the App Store. I think they were making 2 and 1/2 million a day. So half of the top 10 apps on the App Store were based on our technology. It's phenomenal. Talk a little bit about multimodality. We talked a little bit about that backstage. Yeah so like we released models of all types, so we released the open source versions of 83 g, t, j, neo and x, and those have been downloaded 25 million times. We've got kind of code models coming, all variants of an open chapter empty. But I think the most interesting thing is we just got 11,000 Bollywood movies. They're actually in a truck at the moment on the way from one cloud provider to another. They are these trucks that have hundreds petabytes, 100,000 terabytes of movies. We're training a video model so you can generate any type of video set for it will be Brown not wait no offense but then we've also got audio. We've got 3D and others coming. So we feature that's coming because everyone is putting so much resource into this particularly after last week that was a crazy week is going to be about Ready Player one experience you can generate anything dynamically and so I think again you saw initial breakthrough stable diffusion when we released it open source it became the most popular open source software in the world ever. So on github, stars overtook Bitcoin and etherium, the cumulative stars over 10 years and 40 days, 30 days, 40 days cumulatively. Now, the whole ecosystem we ever took Linux as well. And that's what happens when you put these new libraries, these new technologies, these new functions in the hands of people. And so. Yes and content. Incredible Yeah. Yeah that's how we know we're in one of those moments. This is, as we have seen before in previous generations, of a technical leap forward like cloud computing or mobile or the internet that we have seen in previous generations. That fundamentally changes the way that we operate. And in it, the adoption of it goes as fast as in every cycle. The adoption curve is faster. We adopted mobile phones dramatically faster than if you looked at the rate of adoption of televisions or telephones. They just they get steeper every time. And I think one of the interesting things as well is as we talk about how it goes to the future, so stable diffusion. When we released August 23rd, 5.6 seconds for an image on A100, a top of the range chip. Like I said, now 40 images a second. Yeah, orders of magnitude speed improvement. So you will see this as well with the transcripts and chords and others of the world. Yeah, because these models are actually highly optimized. Yeah, we're just getting started. Just getting started. Which is crazy to even think about. Yeah Yeah. If you looked at the outputs of the models for writing 12 months ago, if you gave it a prompt and you gave that same prompt. Yesterday it looks like the difference between a kindergartner and somebody in high school in terms of the output, in terms of its sophistication, it's remarkable. And you just imagine what's that going to be like in 12 more months and 24 months and what are the implications of that? Shane let's talk about that writing. Jasper one of the fastest growing software companies I've ever seen, the adoption of the platform on top of these models has been extraordinary. You'll have done a remarkable job of tailoring it to the marketers use case. Talk a little bit about what you all have seen customer examples that have been amazing or surprising places where it's shown up that you wouldn't have expected. Yeah Thanks. And Thanks to our partners. I mean, we're in this really unique position where we the better their tech, everyone's technology gets, the better outputs we can provide and the better experience we can provide for our customers. Yeah and so I think surprise is a great term for what we see every day. I mean, we're as you saw with the Jasper AI engine, fortunately, the surprises don't really come on the technology side because we're able to test all these models. Well before they go into production. Our surprises actually come in how our customers and users are actually creating content and where they're pulling us into their organizations. So even just looking out at this group today, I think if we did this a year ago, it would have been a bunch of people looking over their shoulders like they're cheating on something or really want to show anybody. I'm here. We're the secret weapon that people use to do their jobs better rate and create better content. And I think now it's becoming a tool that we're actually if you look around here now, it's ceos, it's heads of marketing. It's heads of course, it's heads of sales, talking about how we can leverage generative AI in their organization to do bigger, better things at scale. So that's been a really big surprise. Obviously, the content and the ideas that come out of the system are like truly breathtaking. I mean, one of my favorites, it's maybe a 50 person company. It's an auto shop. They do most of their work is actually window tinting. And so they were trying to figure out there was this stigma around window tinting that some people didn't want to get their windows tinted because they thought it had a certain look to their car or whatever. And it was hurting their business. They put it in. Jasper, give me alternative reasons why I should get my Windows tinted. And the number one answer that came up from Jasper was to keep the UV rays out of your car to protect your children in the back seat. Like, how would you ever come up with that? And that became their entire marketing campaign. It became all the ads that they're running around, keeping UV rays away from your kids and away from your arm in the window all day while you're driving. And their business, you know, went 10x or whatever it was. So that's something, you know, you would never think. And so now in especially in these large companies, the ability to have a new product that's launching, the ability to bring your whole entire organization, regardless of whether they're in marketing or they're in sales or they're in cars. But a marketing team to come up with a concept, a brand and image and be able to spread that out across the entire organization. Yeah and automatically create content for their teams to go to market instantly around a product launch. You know, it's incredible. It's awesome. Well, one of the things, as you said, you all are working together collectively to raise this industry. This whole event and conference is all about that. You all do share one thing, which is you're all competing with some Giants. You know, this is a David and Goliath moment, if ever there was one. There are there there are big tech players who will eventually will play an important role in all this. You all are innovating and creating and shaping this industry from the ground up. Let's make it a mix it up a little bit. And in the David and Goliath. Why does David win? We've got a bunch of Davids up here. Why is David win? To win again? Jump in. Anybody I can start? Yeah so? I don't know. Maybe I'll give an answer to the question. I'll give something that challenges the question. All right. So in terms of answering the question, I mean, I'm a I'm a big believer in I'm a big believer in talent density is the thing that generates everything else. Right we're a we're very small company, but, you know, we're the team that's done really great things in the past. And, you know, I think we've managed to hire a bunch of really great people. And, you know, I'll take the team of, you know, 70, 80, 100 people over, you know, team 1,000 people every every, every time. And yeah, you know, we've continued to innovate and you know, we were the inventors of are all from human feedback now we have this constitutional way I think that I'll talk I'll talk about I'll talk about in a little bit. And so, you know, yeah, I'm just I'm just I'm just a believer in the ability of really small groups of talented people to, you know, to change the world. Yeah, to change the world and produce something great in terms of challenging the question, I mean, I think this field and this industry is growing so much that there is room for many different players to do many different things. And I would say in goliath, what we've chosen to do is, you know, to try and push the field in a certain direction. And I'll give you an example of that with constitutional I with her RL from human feedback, which was the method that was used to train GPT and that we use in the early days of clod. Basically you take this base language model that you know, has trained on a bunch of raw internet data but doesn't have a sense of what it should be doing, or what its outlook should be. And you show it to 1,000 or so human contractors where the human contractors will give some response. You know, the model, you know, you say, what do you think of Donald Trump and, you know, give different answers and the contractor says what they prefer. Now, one of the problems with that method is that it's very opaque. Right if someone asked, well, why does this model like Donald Trump or why does this model hate Donald trump? You know, all you can say is, well, we hired 1,000 human contractors and the average of the human contractors generalize through some high dimensional ML manifold came out with this answer. That's why the model gives this answer instead of something. And so it's not it's not very satisfying to legitimate questions that people have about why? Why does the model think this instead of that? And you know, who, who made it think that? Yeah so the idea with constitutionally AI is you just write a two page constitution, you could think of it, you know, less grandiosely is just like a terms of service or something like that. You know, you could imagine like, you know, Facebook's moderation or terms of service or something, and then you train the model against itself to produce outputs that are in line with the Constitution. You have it critique its own outputs for whether they're in line with the Constitution and you feed that back into the model. So at the end of the day, you can say, OK, I have this model and you know, it had these set of instructions and you know, this is what it is. This is this is what it's supposed to do and what it's not supposed to do. And you can kind of separate the conversation of, is the model doing what it's supposed to do? Yeah and did we put the right thing in the constitution? Yeah it also enables individual customers to say, well, what do I want in my constitution? If my model is acting as a lawyer, I might want it to behave very differently than if my model is acting as a customer service agent or a therapist. One might say, you know, you got to tell a customer that the truth. If they have a wrong idea about the law and the therapist says, no, you know, you need to you know, you're really treating you're really treating the customer. You you need to make them feel better. So I think that's all to say, that there's many different directions to go in. And you know what? One of our pictures of what we're trying to do it, anthropic, is just expand the range of things that are possible and push the field in the direction of these models being safer and more controllable, you know, without losing the aspect of fun and power. I love it. I love it. And obviously that's going to allow these models and the technology to be brought to a whole different set of classes of users than otherwise, with the fear of hallucination and everything else would not otherwise be able to get there. So terrific innovations here. Anybody else on the David Goliath stuff? Yeah, I think, you know, as Daria kind of intimated and you know, you've got Shane, you've got Amjad and others, there'll be infinite customizability. And it's a market creating innovation. Yeah, I have no doubt. I was a former hedge fund manager. $100 billion will go into this sector the next few years. Probably trillion. This is more important than 5G and there'll be so much value created that this space for everyone, incumbents and others. I mean, who would have thought Microsoft would have integrated already? And then there's again, there's complementarities. So like where we're based is it's Windows versus Linux. We are the open source. So all of you that have private data on prem data, you will most likely use our open source models and that's complimentary to these amazing generalist models that are emerging whereby it's customizable. Yeah, but it's not you're not going to own those models. You're not going to own those weights. Yeah so that's kind of where we found ourselves in this and that's why it's putting out to the developers. And the final thing is community. So we have communities of hundreds of thousands of developers that build around our models and contribute to them as well. You know, and we do literally everything we do protein folding, you know, we do video, we do audio, we do everything. And I think the community aspect also becomes important because like, again, we're building a community. Jasper is building a community and we're just scratching the surface of how these models are being used and how to interact with them. And again, there's really huge amounts of value for David and Goliath in this, I think. Love it. Yeah I mean, if you look at the history of computing, every sort of every project that starts with the Misfits and the outsiders and the open source hackers. I mean, the early days of the microcomputer, what we call the pc, was mostly, you know, these like Bay Area misfits and hippies, you know, where, you know, Steve Jobs and Steve Wozniak used to go to these meetups. And what did the professionals do? They were working on mainframes. They were wearing suits, you know, working at IBM. Right and so I think the tinkers I'm a big believer of the value of tinkering and the value of creating that community, that immodest, talking about the stable the future community raptured were announcing actually maybe I'm foreshadowing that later this week that we reached 20 million developers globally and our audience is very different than the sort of regular sort of developer community audience. I'll give you just one example. A friend of mine posted a bounty on our gig platform. He runs a large executive assistant virtual assistant service. And he wanted to use potentially to look into AI as something that could help his assistants become more productive or potentially be a front line assistant for his customers before it goes to a human. And he puts a bounty and he underpriced it so much. I told him that he really underpaid for it, but he boasted about he said, I want to train opening, eye opening. I has a fine tuning service on top of our data from the executive assistants that are talking to customers. And he got a 13-year-old kid in India that built him a training platform. For $50. And so he definitely underpaid. And and, you know, the market now is like going up in price, but for some, it's still sort of underpriced. So you want to go take advantage of that. But, you know, ultimately what I'm saying is the world is big. Yeah, the world is really large. And there are people all over the world that are waiting to contribute. You know, I'm originally from Jordan. I mean, I had to come to Silicon Valley to build my company. I try to build that in Jordan. And I couldn't I couldn't raise any funding. I couldn't get it, get a team together. Nobody would pay attention to me. But the next Amjad doesn't have to come here and he can build a business back home. Powered by AI. Powered by this superpower. Yeah, yeah, Yeah. I think I think the interesting thing about this David and Goliath fight. Is that the winner, there isn't going to be a single winner. I think we're all going to win if we work together and we're smart and we're focused on what do customers want? How do customers actually want to use a.I., not just building really Cool Tech that's going to accomplish something that never sees the market, but how are we focused on what do customers actually want and what's going to make their life better? Yeah, what's going to make it easier for them to do their jobs, be more productive, spend more time with their families? All the important things that I can actually deliver if we're really focused on that. This space is so massive. It's bigger than anything else. It's bigger than it's assets, bigger than it's really big. And we're firm believers that every single company will have AI, every single software product will have AI. There will be all kinds of different pieces of AI within a software product that will be generative. There'll be all kinds of different things. And so I think if we're all smart and we all work together and we all obsess over how our customers want to use our technology, this is going to be a massive opportunity for everybody in this room, really, and and for society. And then we've only got a handful of minutes left. But I do want to bring it back to a slightly off topic conversation. But I'm glad you mentioned the Amjad and Jordan that gets to use AI. Now, there is something wonderfully powerful about the democratizing of this technology and knowledge and getting into the hands of lots of folks. So two questions. One I'll start with is we have a history teacher, art teacher, English teacher in the audience. I don't know if there are any parents in the audience like me with a high Schooler and a middle Schooler. What do you tell the teacher that's out there that says, I really still so really important to me that they learn how to think critically and write persuasively and do all these things. What do I do when I show up in class next week? Do how do they use this? I do. They try to fight it. Do they try to embrace it? What do you what do you all think about how they use this technology in the classroom today? I think the history of fighting technology has not sort of fared well, like even like in Jordan, up until recently, you weren't supposed to use a calculator to it. So I, you know, so I felt that I was so enraged by it because obviously the computers are going to be much better at like, you know, multiplying six digit numbers, right? Yeah and so I think what we're seeing because we have a significant amount of teachers on regular upload is actually the number one platform for learning how to code and what the innovative teachers are integrating ghostwriter into their workflow. And so if you fight it, they're going to go and use it behind your back and they get cheered and do all that stuff if you teach them how to use it. And if you put the correct godly guardrails. Yeah and I think there's going to be companies that would want to work with teachers to kind of bring this technology in a productive way. I think there's absolutely a productive way to use it. Just just as an example, the AI could actually be do work to like critique the student's work as opposed to write it for them. So there are ways to use the AI that's not just do the work for you, but actually give you feedback that is useful. For example, yeah, I love it a lot and I were talking at dinner last night about AI for good and how it could be used for good. His company has been doing quite a bit of interesting work here. You've got a nonprofit that focuses on this. Maybe share a little bit more of that story with the room. Yeah so my co-founder Joe runs imagine worldwide the org. So we've been deploying tablets with adaptive AI technology into refugee camps around the world. And from the latest, asked for 76% of kids in the worst environments in the world, they get to literacy and numeracy in 13 months on one hour a day of tuition. To give you an idea, to give you an idea of that out of 10 kids in Africa can't do a sentence by the time they're 10. So we just got the remit to basically educate every child in malawi, nearly 4 million kids. And we're going to basically bring this technology to hundreds of millions of kids over the next few years. For those who read the diamond age by Neil Stephenson, this young ladies illustrated primer, that's why we're focused on the edge and bringing these models democratized. Because I the education system would be the best in the world and need to work with you and augment you rather than replace you and do your work for you necessarily. And I think that's the future. I mean, I love it. The reason I ask the question was there's always lots of concern and fear or anxiety when a disruptive technology of the scale that generative AI comes out to the market. It's super important to remember this technology is going to be is currently and will continue to be used for an awful lot of good in the world. And we should not fight. It We should embrace it. And I love that simple summary. All right, parting, parting thoughts. We're here in a year. What's happened in the last year? Make your predictions. What's going to happen? What are we going to be surprised by? I didn't I didn't prepare them for this one. So it's going to be fun to see and think on our feet. But what do you think a year from now, as you think about what's transformed in this generative AI industry, what looks different from where we are today? Good question. So I think in our world, Jasper is truly everywhere. Everywhere there's a text box and you're creating content. Jasper's there. Any system that you use throughout your day, any piece of content, any group, any person in an organization, Jasper is there to help make creative people more creative. And I think it's just part of how we live our life. I don't think it's an AI thing anymore. I think it's just the tools that we use make our lives easier in certain ways and it's just become part of the fabric. Yeah, I mean, 10 weeks ago, you know, no one really was talking about generative. That was November 30th, right? It came out. No one's talking about generative AI it's only been 10 weeks and now the world looks different. So you look out a year. I think this technology just becomes part of the fabric of how we live our life. And I think it's everywhere. And I think, by the way, our kids are just going to use AI like a calculator, like everything else. And I hope the smart educators embrace it. And it just becomes a new baseline that we can then push our teachers and push our students above and hope make things possible that weren't possible before. So that's what I'm I love it. Yeah love it. Going around I sometimes think we not really reinventing AI but time machine like people say Web3 moved fast. This is insane. The pace of innovation, the pace of adoption. We would have thought $2 trillion companies would basically be generative at first in kind of Google and Microsoft. That is how they orienting. I think in a year from now, I think language becomes even more elegant, like prompting and all of that. It just incorporates and it learns from you. Image is solved. We're having a stable diffusion moment in audio as well. We cracked music, speech. I don't think video will be quite so. It's going to be far out. Yeah and so again, being able to create anything you can imagine seamlessly is going to be an amazing thing. And I think it's a step forward in humanity because the Gutenberg press allowed us to communicate through text. Yeah, Jasper and others are removing the barriers for us to communicate audio visually. Anything Yeah. And you think about what that means for society. I think better stories for everyone, better communication. And in a year we're going to be there where this will be the big thing. Yeah, but not hype, real love. It comes from, I think it's hard to understate the magnitude of what's happened here. You know, Abracadabra. It's like this magic wand, right? Like now we have spells. Like, we literally have spells. You can, like, put a spell in two, T3 or quad and get an output of it. That's actually useful. Useful for your business, useful for your life, useful for your art. I'm focused on, on creating software and coding. And I, I think a year from now, we're going to see a whole new crop of entrepreneurs that weren't possible before. Perhaps we're going to see like $1,000,000,000 solo solar printer company in the next year. But at minimum, there's going to be an explosion of entrepreneurs building amazing businesses with very, very small teams. Yeah, I think in Japan the word for a prompt is magical word spell. Perfect absolutely. Is that magic, dario? Yeah, so agreed with everyone else. The pace is just going to pick up. You know, I think there's a dynamic happening that, you know, until maybe until maybe a year ago, most of these large language models were research projects. So the most you could. Justify spending on them was a few million dollars, which sounds like a lot, but really compared to the scale with technology is not. Now folks are spending $100 million. I think within a year we're going to see people spending or starting to spend. Of order. Of order. $1,000,000,000 and everything I've seen is when you keep scaling these up, they keep getting better. And now now that you can justify them with revenue, that cycle is just going to turn faster. In terms of what specifically I think will happen and I'm excited about, I think in a year it's realistic to imagine that we have models that almost never hallucinate when we don't want them to hallucinate. Sometimes we want models to hallucinate, we want them to tell stories. But you know, other times, you know, if you're wanted to read a pathology report or something like hallucination is definitely not what you want. And I think that's blocking much broader applications of the technology than are possible today. Right people look at the hallucination problem. They're like, oh, man, this thing might do the right thing 90% of the time. But I just like like, what is this? I can't I can't trust this. And I think that problem is I think that problem is going to go away. I think along with it or going to get substantially more reliable models, yeah, the accuracy is going to go up. But you know, as we see wider deployment, the standards for that are also going to go up and we're going to face new challenges that, you know, that we are not even on the radar today. Yeah, fantastic. Well, on behalf of Bessemer venture partners, Thank you to Jasper, to Shane and Dave and the whole team for bringing us out here. I on behalf of this whole audience, I'd like to Thank all of these four guys for incredible insights and sharing your perspective. Thank you all very much. Thank you, guys. So Thank you so.

‍