Gen AI-Powered Solutions | ASU+GSV Summit 2024

[Music] thank you everyone for being here thank you to our panelists um you know I'm excited to host this panel I'll be I'm admittedly a little bit uh scared of it as well because uh I am far from an expert on AI but I would assume that a lot of people are probably in that same boat um so we have a lot to dig into uh you know I think there's still a lot that we don't know in terms of potential use cases risk that may come to bear as the adop of these capabilities ramps and evolves over time um but yeah really excited to dig in here with people that I really respect that that I think are with their companies that are that are I think at the Forefront of pushing the AI conversation forward um so looking forward to digging into it um but just wanted to start here as as we think about maybe just level setting a bit you know with the conference last year there was a ton of talk of AI generative AI um we're sitting here A year later I think the question is are we moving faster or slower than you would have expected sitting here a year ago and and you can maybe frame that in the context of your individual companies the industry as a whole however you want to take it I don't know if anyone wants to volunteer to start but yeah I'll start uh soen CFO at Doo I think uh just perspective our company's a learning management software company so we serve around 3,700 Enterprise customers in terms of solving the problem from a learning management perspective perspective so we the typical use case of our customers customer education partner education employee on boarding compliance um sales enablement so on and so forth um in my in our world I would say that you know there's real problems that are solvable uh using AI we've had a c one product that's been in the market for the last three years um which automates content creation which is a pretty easily understandable um you know perspective that content should be automated using AI uh you know we have a product effectively what it does is it takes the content of the customer so someone like an AWS uh which is a customer will utilize uh shape which is our product automate you know it will create a microlearning pill from any any output that you uh you include in the product so let's say it's a it's a 500 page operating mod um you know operating um uh document it will take that create a microlearning pill within seconds with the voiceovers the background converted into 50 different languages um in real time and now what you've effectively done is automated content for the customer and taken productivity you know 10x faster at the same time reduce the cost to produce content which is very specific to the End customer um by I don't know close to onethird or what it would have been initially so um where you can actually demonstrate Roi from a productivity perspective or reduction of actual cost that they would to spend I think that has been monetized and can be monetized I think the biggest question question that some of us all of us are are probably debating is certain features functionalities that AI is going to solve does that help you monetize it or does that help you win more customers and that's really where the game's at this year is that there's a number of features functions companies are releasing like shape shape we can monetize but if I'm doing semantic search within the ecosystem of my custo helping them search through the content that that sits in my my ecosystem that's more of a functionality that helps me en enable better win rates than actually monetizing it separately so that's kind of the biggest question that's come out and will play out in 2024 25 if you look at companies like CRM you know they're calling for a 9% growth rate then you kind of question well that's your uplift and where's the AI monetization coming right and that's really where is going to play out um at least in my world yeah I I kind of think about it on three Horizons in terms of how fast are we going compared to expectations um they're sort of what am I seeing and then how is course Sarah doing our or how's our organization going and then how where are customers and and how are customers adopting and was their understanding um for from my perspective you know when Chad GPD came out I totally freaked I mean I just it was it was more startling to me than any technology more than the iPhone for me at least I was an English major and I just had never imagined that a computer could produce strings of words that made sense like it did and I was like whoa this is crazy crazy and then when gp4 came out just a few months later then I was like oh my God I connected two dots and I thought it was going to continue at that rate we started talking about autog GPT and agents this was in you know March of 2023 and I just thought if it continues at this rate I don't know what's going to happen in the next 12 months I think a lot of what we saw in the early stages was people had the technology ready and it kind of Hit the World by surprise and so it felt like a really rapid rate of of Advance on the technology but it has down a little bit like I was panicked that by the end of 2023 agents were going to be actually buying things on Amazon and doing all these things and that has not really happened what I have been impressed by is the reasoning capabilities of the model is is not just open aiet like Gemini Pro is is quite good and and um CLA Claud is 3.0 is quite good so there's more models that are really good than I thought and I just got access to Google Gemini Pro 1.5 I've been a big believer in context grounding which is putting into the prompt the knowledge that the large language model needs instead of fine tuning and I was waiting for that window to get bigger and bigger because then I could say I could say here's what I need you to know to help me and if that got sufficiently big it could know a lot of stuff and originally it was like 8,000 tokens which is maybe 5,000 Words which is you know not bad and then it went to 32k and then 128k for for GPT uh for Turbo but with G with Gemini Pro 1.5 it's now a million tokens and I've been using this for like the last couple weeks I can put in everything about my company our strategy our goals the org structure almost everything and the reasoning capability is pretty freaking good so I am now routinely just talking to G Gemini Pro routinely I I have all my prompts ready and it's really quite startling so deep down I'm feeling like things are moving super fast because of some of these capabilities in terms of of corsera we'll talk I'm sure more about what we're doing uh all together I'm I'm pretty happy with that um I always wish we could go be going faster but customers I think are just kind of overwhelmed I mean customers are overwhelmed it's it's not that it's it's it's hype it's that it's hard people don't know what to do they they don't know where to start they don't know where the value is going to be they're not sure where to monetize it they're worried about risks and inaccuracies there's it's a complicated situation that's moving really fast and I think for a lot of folks this kind of paralysis yeah that was really well said what he's said uh no I so I I I had a very pedestrian answer which is I actually think we're sort of on Pace with what I thought which is to your point like the initial was wow this is crazy but then you know like with any new technology there's sort of a I think you framed it well there's like a pent up sort of release of that and then like we're seeing the work of years that that comes to fruition and then it you know slows down or perceives to slow down a little um I actually think there's a I I've been looking for analogies and I think a strong analogies if anybody who can remember back into the like early 90s when sort of the internet was starting to be a mainstream thing and every business was going to be massively disrupted overnight on the internet and here we are how how long have we been saying education is going to be massively disrupted overnight on the internet and so that's not a I don't represent that as a negative or pessimistic view but just as I think that gen is a really powerful technology capability and has profound implications for everything that we're doing but it's a technology that needs to be applied I'm sure we'll talk more about that too and so I actually think that you know we're not going to see overnight transformative like businesses go away and new ideas happen you know Etc I think it's going to be uh you know we're going to see bumps along the way in terms of stair step function changes as the models get better and have more capabilities but I think it's going to be we're in this phase now of like okay the cool new thing is out there and now what are all the practitioners going to do with it and how are we going to build it into applications that fundamentally transform specific workflows or specific Industries and yeah there's going to be disruption in particular jobs and and and whatnot but I don't think that's all going to happen like suddenly overnight every knowledge worker is going to lose their job uh I think we're going to see you know efficiency gains and etc etc so I'm sure we'll talk more about that too yeah I agree and I think um my mind is sort of divided into the the capability and capacity of the technology which has sort of blown me away for all the reasons that you just named uh and then the product productization and the productization perhaps predictably has trailed maybe it Trails a bit my my expectation but from a learner's perspective they don't sit down as Jeff does with all these questions and all this data and often the the characterization of the learner is that they that is their circumstance right they sort of sit down like thank God a chatbot I have all these questions for the chatbot and the interface right now and the productization of that capacity is really under delivering and so I think it's sort of like there's going to be pent up eventually there will be breakthrough products but right now I don't think any of them really has broken through from a learning perspective a lot of them are built on the assumptions held by the the product creators that that the learner is in their image and and if you've been in classrooms recently that's not what classrooms look like social social learning classrooms perhaps individual Learners alone in a room May sometimes look that way but most of social learning isn't doesn't look like that the only comment I'd add is in the enterprise software where you're serving enter I you know companies there is an element where who has the data and how you can you can utilize that data to make it very relevant for the end user which is in my case a customer um or enterprise software Enterprise company um that is that is extremely important the models are only as good as the customer letting you actually make it it much more useful for them and so you know whoever is sitting on that intellectual property that can make it much more useful for the customer is going to also benefit at least at least from an application software perspective that was going to be my next topic was focusing on the data side because that's something that anyone who was on the said on the last panel was a topic on the last panel too but you know as you think about data it seems like it's going to be a big focal point with what you're feeding these AI models um how do you think about some of the key proprietary data sets that you as a company have have you had to change your data strategy at all to think about building out better capturing data sets within as you engage with Learners as you track prog pressions maybe talk about the data side of the equation for I'll jump in so I I uh I'm the CEO of Quizlet direct to Consumer study platform you know millions and millions of people use this every day to study so we have lots and lots of data and we've been and and there's content there's the content of what the materials that people are studying but kind of the the the asset the data asset that we have that we think is really valuable is the data of actual study like what's the progression of things you're studying what are you getting right what are you getting wrong when you get things wrong how are they wrong Etc and we've been using ml like data science and and AI is not new geni is relatively new but we've been talking about you know we just keep calling it different things along the way but we've been using uh data science and machine learning for a long time and I think what's happened now is that we can start to really activate that data in a really meaningful and direct way that maybe you know before it took a long time and and really you know really specialized data scientists to do something use useful and that useful thing that we could do was a model that could do a small personalization feature or something like that right but now suddenly we can activate that data set in a really powerful way and so yeah we we think all the time about okay what are the you know obviously there's a privacy angle of this as well of like what what should you collect what do you need Etc um but like how do we how do we aggregate that data on behalf of the learner because now we feel like the cycle time for us to give it back to them in some useful form in in feature set personalization Etc is so much faster so it's it's just tremendous yeah I guess I I would add um when we first started using the generi like you said we we have a lot of ml models we built over the years which like put this message up you know predict the probability the right time predict the probability that this person will click on this message and if it's higher than put the message up like that's what it was doing and that has now turned into like a personal tutor that talks to you and helps you explain Concepts it's like okay that is a completely gamechanging Innovation or evolution of of what's happened we uh made a decision pretty early on right right away we were thinking data modes data modes like what could we do that's totally proprietary and we just saw the models changing so quickly we're like we're not going to we're not going to win this war by building all of our own models but trying to understand how to use the content to and the underlying large language model so that without building a lot of proprietary models we still have some you know we still have a data science team and they're definitely still building some models but they're not building large language models not right now they're not we're not even really fine-tuning very much stuff we are really trying to figure out how to for someone who could use some benefit of a practice quiz or an answer to a question or a recommendation on what to learn it's how can you find the right content from somewhere around corsera and send it with the question into the right llm to give a really valuable answer that's presented in a great way but we're not investing in llms a lot right I mean almost at all there's just it's going so fast and we're getting so much value out of it that we're trying to differentiate more on the context of the content around it as opposed to like the parameters that are in your ml models I think I think from from our perspective I'll say that you know as a learning management software company we have the privilege to sit on the intellectual property of the customer so then the next question is I'll just give you an example so it's it's it's relatable as a company we're working with and these are multiple customers we do the same thing with this they had 200,000 sales calls last year as part of their ecosystem this when they reached out to prospects now you have an incredible amount of data whether it's from gong and other places where you could utilize that data and say do I create a and this is a product we have virtual roleplay technology where now you're practicing how to sell that product and you now know from that 200,000 calls last year what's your competition how you should have sold what the product should have said how you should have said the initial words when you introduced to the customer and it's not practicing one of AAR it's multiple avatars at the same time and so now you're getting challenged back and forth based on you as the company and your intellectual knowledge that has fed into the algorithm of what we call virtual roleplay technology the one challenge I will say in the Enterprise space that we certainly have have kind of taken up uh at at the front is what we call AI Control Panel um most Enterprise customers are going to tell you to switch it off or tell them exactly what the what the algorithms are doing for you uh unless they know exactly where these apply so even a simple aspect of it if you're going to use that data even in a synthetic format so someone like an AWS doesn't want Walmart to even benefit even in synthetic format you have to make sure that you have an AI Control Panel that restricts the algorithm and the usage of that algorithm only for that end customer and that's been playing out pretty interestingly in our space we get a lot of large forun 500 companies that are specifically asking uh and we've looked at that we built a AI Control Panel to give them full control where the algorithm can be used cannot be used is a synthetic format or not it's becoming an important question and and another point I'll just add on to that because I think it's it's it's not obvious to everybody but if you make a policy decision to train a model on data that's not absolutely clearly your data it might be legal and it might be the right thing to do in the long term but the number of questions that you're going to get the number of hesitations you're going to get the amount that it's going to slow down your rate of learning and your progress we decided we're going to make some very simple statements we're not training any models on anything that you put into this y any further questions on that okay now we're going to go like crazy building these things but dealing with intricacies associated with whose data is it that you're using to train these models is often not worth the values of the models that you're training is is what we found got it um yeah that's it's awesome um yeah um maybe maybe moving on to the application of AI to personalize learning I mean that's that's a broad topic that could look a lot different between the type of learner that you're trying to serve you know as you think about your efforts to maybe personalize The Learning Experience in practicality in reality what could that look like over time I don't know could we start with what does it look like sure maybe we can say where does it go like what does it what does it look like today today yeah I mean so I was thinking about like what the the prospect I guess what I'm excited about where I think we can finally go that we've been help back for a long time uh in personalized learning there's basically as I've seen it sort of three domains where we've made progress already we've had uh in K12 um early literacy uh measuring like acquisition of language and phonics and so forth decoding uh uh World language learning products like Duolingo uh and math math is always the F sort of the first place that these things strike and the thing that the each of these three domains has in common is that they have a highly structured sequenceable uh learning learning trajectory and you can build assessments around that and I think the the power that I see being unlocked is in generating a higher uh a more powerful robust set of Assessments and that's going to unlock you're going to see and you've already seen it now with like the Advent of a lot of writing feedback that's sort of pent up uh o over time because that's been a a persistent need now it's unlocked and I think you're just going to see many more domains that are being unlocked thanks to uh the ability to create more finely more accurate uh responsive assessments and I think that's in contrast perhaps to where we are today which is we're not there well and I Jeff you actually gave a great example of like personalized learning in your in your opening remarks about you know you take all of corsa's information you dump it into the context window and then you ask questions so it's your experience with Gemini Pro is way different than mine because you're personalizing it so it's not the traditional way that we think about personalization but that's how that's exactly how we're thinking about it from a Quizlet perspective which is if you're a high school or college student and you're studying for something you want to study for what you think is going to be on your professor's test that's that's that's the job to be done is I'm going into my midterm I got to sit for an hour and a half with the blue book what do I think is going to be on that test because that's what I want to study and so for us personalization is well wow if I can take an llm based application and I can dump in everything we know about the course everyone who's ever studied for that course in the past Because by the way the professors taught this course for the last six semesters and then say hey generate a study guide for me based on all of that context that's like a really personalized so it's not personalized down to that individual per se but it's actually personalized exactly like the job that they need to be done for them and so that's that's what we're doing um and it and it's something we it's funny to your point of it's something we were working on two years ago three years ago four years ago and we were making like incremental progress and then we made like a big for Ward leap in terms of the art of the possible and is it's still not perfect it's you can't news flash you actually can't trust the output from any of these models you've got to do a lot of post-processing to make sure it's valid to make sure it's accurate to make sure it makes any amount of sense Etc and one of the other interest I don't know if we'll get to this but one of the interesting things we found at least in students they're like super skeptical about AI right now because authoritativeness and accuracy really matters when you're studying for something especially if you're studying for an exam or for a test or something and so there's this you know they're also like the people we would think would be the most eager early adopters of tech platforms Etc and they are but they're also somewhat skeptical about about AI so there's a really interesting Balancing Act there and when you think about that personalization do you think that skepticism is changed over the last year yeah it's gotten worse uh I mean seriously like when because like like a few of the folks here said like when when chat GPT was released what November of 22 we were all like oh my God that's amazing and students were like sweet all my essays from now on I don't have to write this this thing will write it for me obviously not true um and after using it even a little uh you know you quickly realize like oh it's good for these types of things it's not good for these types of things it gives useful directive information in this direction but I can't actually trust it to do this you know so I think I skepticism maybe sounds too harsh but I think there's a reality that is set in of like okay here's how I could use this new tool you know and here's where ites doesn't apply we we have many examples we give but one is uh what we call corsera Coach it's like this little personalized tutor and career guide and helps you with Discovery but just within the tutoring side uh so all 7,000 courses on corsera have this little coach which is kind of amazing that you can do that if you're not training the models and you just need the language of the course coach gets really smart on what's in the course because it when someone asks a question we find all the different parts of the course that might be relevant to answering it and then say to the llm give an answer based on on that stuff so the stuff that we see people doing I mean this is live now so we we launched coach in March of 20123 so it's been out to millions of people for a long time and uh summarize this video well people press that button a lot how does this help me in my job they it sounds really silly but like how does this help me in my job if if coach knows your job it says well since you're a software developer learning this skill will help you do this better if it doesn't know your job say well what what's your job and they like I'm a software developer okay well this video is about this and they're teaching you this and so in your job as a software developer you can do this people say if I learn these skills what job can I get like oh well these are the kinds of jobs that use like it's in just the number of things you can ask about why is this relevant to me there's another button that's really simple which is explain this concept someone doesn't understand something I like explain this concept it explains it and you say then you could I still don't understand it explain like I was a 10-year-old blah blah blah like it just keeps on answering your questions based on the context in the course until you get it there's another feature that that coach has which is give me practice questions and what Co which coach can do is it can look at the questions that are going to be you're going to be quizzed on don't ask me those questions because those are the ones I'm my go then it looks at the learning objectives and all the transcripts and on the Fly starts generating random questions not random but like personalized questions to you you give the answer it says well not quite right it can tell you what got wrong of what you need to study and here's the clip to watch it's it's it's mind-blowing it's absolutely stunning and so yeah people are clicking on these buttons at first they think it's like a support chatbot and so they don't click on it but once they start using they're like whoa this is like I have a real tutor here who's help me just to give you a little sense of where it's going it's now grading essays it's giving you personalized feedback on your written submissions and there's something to work on now that I'm just super excited about which is when you do a written submission not only can It grade it against a rubric it does you know when someone gets their PHD and they submit their dissertation you have to defend your dissertation which is like your oral quiz of what you submitted so now we're having coach after you do the submission not only give you feedback on the submission against a rubric but it says hey let me talk a little bit about your thought process when you submitted this your thesis looks like it's this what else did you think about writing about and why what would be the inconsistencies between what you said here and what you said here so it it actually is evaluating your thought process not just your submission and it's not only doing that for academic Integrity but it's also helping you think through how did you approach that problem and I think that's pretty cool right um yeah that's exciting I don't know if that I've heard a rumor that there was a Harvard Business case study coming out on you guys I don't know if that'll be covered in that but it will um maybe moving on to I think we got 10 minutes left monetization I mean there's there's a lot of investment that's going into leveraging these capabilities you all operate companies there I would think that I guess how do you think about the monetization as you roll out these AI capabilities this is about remaining relevant you know will it help you with pricing does it drive more engagement how do you guys think about the ROI on investing Behind These capabilities I'll start because I guess I'm a CFO so start from the math perspective um I think the way I think about it is I mean we talked about capabilities any capabilities that you're bringing to the market you have to first think can you demonstrate clear Roi to the customer so you know in our world in corporate learning there's really four or five aspects that are going to dramatically change right you either automate content for the customer so you're reducing dramatically their cost of creating content out house we talked about that so there's clear Roi measurable and as you Empower social learning in an organization you also have much bigger and much more seats taking that the historic model was give it to the admins they create content they distribute it to an organization now you're empowering social learning by giving modules like doape which is effectively automating content creation you would certainly monetize that then you say well I've got semantic search because I sit on intellectual property of my customer all of their ecosystem sits with me that intellectual property is not available in the worldwide web so I got to give them semantic search capabilities and what that means is just as a practical example Lululemon is a customer if you're stuck on the till and you don't know how to do a refund pretty real time and you don't have to go to an LMS real time you can ask the question in their ecosystem system and the answer comes out we call it Tik Tok way of learning that's the next phase of where this is going video capability video learning is real and it's very applicable in the corporate world where quick bites type of learning where content is spun out hyper personalized to the end need of the problem you're solving for the customer or their end user is important so that refund example or you're selling shoes at the till and you don't know what the feature functionality is and you can pretty quickly tell that so video video your capabilities as part of embedding into e an ecosystem where you can actually also practice your skill sets is is is kind of monetizable but I'm trying to say is every time you're trying to figure out are you bringing a real Roi to the customer where you can make productive gains and tie it to some sort of numerical exercise at least in my world um whereas there's elements that are not that are not easily quantifiable like the biggest problem in the industry in the corporate world is um you know you have skills all over the place if if you have an ATM uh provider they'll they'll tag skills very differently to an LMS you think about a company like Intel or or or or others similar like Caterpillar where you're doing a lot of onshoring this is real problem for the customer where they're onshoring a number of their capabilities in the US as an example now I got to move thousands of people from here to this operation I got to have a standardized skills taxonomy and a learning path so that I can bring that human being from this department to this department Department because I need to build chip chip manufacturing capabilities in the US very different mindset and skill set needed so AI is also solving that problem from a skills taxonomy perspective to help easily create learning paths for organizations where they can move human capital around because that is the the human capital is the one thing that is you know as everyone knows is is in limited Supply right now so we we took a pretty simple approach we said out of the gate it was just like we need to learn how to ship value as fast as possible forget about pricing forget about making money on it just learn how to create value for customers as fast as possible so we immediately said uh we're not going to get tied to one llm because we expected that there's going to be a certain break even where the value that you're getting at a cost of the llm is going to make it so we can deliver it for free and create a lot of value for the customer so we we never intended to make any money off any Revenue off of it um we have to watch it on the cost side so we can't we're not using GPT the the most recent version of GPT 4 for coach we're using 3.5 turbo like for any of you who are wondering but we were able to bound it and use enough context so that we can give it away for free because it's fast it's high capacity and it's inexpensive and by bounding it the right way it can give high quality answers that's still valuable so that's what we're focused on but we are worried about the cost like we we we cannot use the best models for everything we for really important problems that are really valuable and not super high volume we'll use them an expensive model for other ones where you don't need the expensive model we use a cheaper model but our anticipation is for the vast am of things we will not charge extra dollars for it where we're right now where we're focus on making money is if someone has a better learning experience and they're paying a subscription if they stay longer retention goes up so that let's just do that let create more value have them learn for longer will make some more money because they learn for a longer time yeah that's I mean that's how we think of it as well it's a it's a technology INF like infrastructure component that we build into product experiences and where it can it has a cost associated with it as many types of you know some things cost more some things cost less um but if we can add value and add meaningful value then we bake that into kind of our product economics of how we're thinking about product and value for the customer we're large scale premium service so we give tons away for free similar to you um and then we have premium products that we try to get people to subscribe to and we we look at the cost but I also want to highlight it's important point that Jeff made about the different types of models because you you want to be we we strongly believe in the model dependency independency of being able to move across and test and figure out what the right I think we'll see in the not too distant future there's a whole infrastructure layer that's springing up around that as well where you know even within a given product based on the workload you'll send it to a different backend model because you're you're trading off speed quality like the quality of the response and the cost right and so you know similar we use a lot of GPT 35 and other things like that not the latest because they're faster and so we can't wait constraints exactly and we can't wait around for the gp4 answer to come back when we're trying to do things in real time versus other things that are you know so that that's a really important like learning that that we had early on as well of like thinking about that but but at the end of the day it's just a cost of goods and you got to think about the product and the value you're building and and the monetization follows yeah we're we're in the same the same that great well we got three minutes left um I just want to you know as we think about maybe 3 to 5 years down the road like what are you hopeful that your companies can provide lever in AI that you can't right now and and maybe we've talked about a little bit with the personalized learning experience but just maybe maybe what are you hopeful that you'll be able to do that you can't right now I mean in in three years if I just look at what we've been able to do in the first year and the rate of change of the underlying things I mean I it's I could just start riffing off the top of my head it should be able to do almost everything for free in any language totally personalized I it it will be the biggest question really is will humans be able to keep up with it with the rate of change the organization is demanding from them and with the learning that that requires learning new skills but also learning how to do your job differently with different people I just the social change the the work change change management is going to be really really hard so that's why I and we so we're gonna help people not just learn something like be able to deal with change yeah yeah I think uh it's funny when when you asked the question I thought what I actually flipped to what won't we be able to do and so but the thing is in learning the thing that I don't think we'll be able to do is replace the educator to human like spark in interaction like I think if you if you look back all the amazing Innovation that's happened in the last hundred years around education and the one thing that nobody's been able to figure out how to bottle up is that spark between an educator and a and an individual um and so I don't think we'll be able to do that I actually don't think we'll be able to solve that but what we can do is give both the educator and the student a bunch of tools so that they can spend all their time Focus on that that spark that motivation that that the human part of the experience that we won't be able to replace with technology so that we can all the things that we can use technology for to make that better faster easier more personalized more fulfilling Etc so um but I don't think we'll be able to replace that human connection piece yet at least I hope not soft skills are going to be important yeah those too yeah I guess I worry that the the K12 policy uh landscape won't keep up we'll we'll basically be be chafing uh at every attempt that we try to introduce this and that the structures of of learning the traditional institutionalized learning is just not built for this um and something's going to break you know and for any any innovators out there exactly what you're saying our team is out there talking to all the people and there's a lot of people generating content from scratch with with these avatars and they can speak in different voices but they're not real people and they didn't start with any really distinctive uh insight and what I really want personally and what we're looking for is what I call text based video editing both additive and destructive where where if you take an existing video of a real person who really did say something and just let him say it a little bit differently or fill in a three sentences take their video clone their voice get their permission um but just help augment and keep up to speed what real humans are doing that that's where I think there's going to be a lot of value I mean it's fun it's fun to create something totally from scratch but that's not what we're focused on well I think we are out of time so thank you guys so much for joining uh thank you everyone for watching thank you thank you [Music]

Channel: Global Silicon Valley

Views: 280

Keywords: arizona state university, asu+gsv summit, conference, edtech, edtech conference, education, global education, gsv ventures, skills summit, technology

Id: 3TLVmt6Uxuo

Length: 34min 38sec (2078 seconds)

Published: Fri Apr 19 2024