Inspired Execution

Chet Kapoor: Alex, welcome to Inspired Execution. 

Alex Mashrabov: Hi, thank you for having me. 

Chet Kapoor: So you have a really impressive career at the forefront of AI including a successful exit with an exciting new, we were just talking about new venture, right? But before we get into details, what made you want to be an entrepreneur? 

Alex Mashrabov: So this is a really good question. So I think it actually comes from my parents. So my parents are professors of mechanical engineering and they are actually professors, so it's above PhD level, right? And they teach at a university. Actually, my grandfather was a professor as well, and he was maybe a top 100 scientists in his area, in the world. So the bar was really high and the idea was always to get to the place with the highest concentration of talents. After the Soviet Union fell apart, the level of science in Russia dropped down and my parents, they were always coaching me, prepping me to get to MIT, Stanford, or other top research universities. This is how I started programming. So starting 11 years old, I started programming, and the idea was to get qualified to the International Olympiad in informatics. And typically people who qualify to these Olympiads get invited to MIT with full scholarship, fully covering the tuition. Unfortunately, I was only number nine in the country, while only top four high schoolers get to the International Olympiads. I realized that I'm, and also I did quite, I think I did quite deep computer science research in terms of data structures and so on, and I realized that I'm just simply not good enough to do world-class stuff on the science front. So I decided to try myself on as an entrepreneur. Back 2015, it was relatively, I mean, for me, it looked relatively straightforward that there are two trends. There were two trends. One is neural networks and second, mobile devices. So that's why I really dedicated my career so far to making sure that those models can work real time on mobile devices. 

Chet Kapoor: That's awesome. That's awesome. And that's when you got started on the whole journey. And then after that, it's it's addictive, right? You want to go back and do it again because you really enjoyed it. And you like the hunt of, going off and saying, I believed in something and I executed on it and built a business around 

Alex Mashrabov: it. A hundred percent. And I was always thinking how to make video generative AI actually useful. So back 2017, I discovered that actually models can synthesize pixels. 

Chet Kapoor: Yeah. 

Alex Mashrabov: And for me, it was always almost a mystery of how to transition from pixels to video generation. Meaning so that the video actually has a high level of motion and awareness of objects in the video. And this is the premise behind, on the technology side, this is the premise behind Higgsfield as we were, as we made a bets on world model. 

Chet Kapoor: Yeah. No, that's a, that sounds super interesting. So let's, before we get to Higgsfield, we'll get there. Let's talk a little bit about your journey. You started AI factory in 2018 and then ultimately sold it to Snap. What advice would you give to people who are looking for a job in the industry, what advice would you give to a first time founder on that experience? What's the one thing that you would tell them that they should think about that you would have hoped you had that advice? 

Alex Mashrabov: I would say one of the most important things I think is to develop empathy. A lots of, very often founders, very often we hear, very often we hear from people, some motivational speeches that founders should have a really crisp vision that they should be in charge, that they should sort of command the walls and try to reimagine the walls. This is maybe the perception which I had when I was 20 years old, but it's just important to be bold in some sense. Although I believe great businesses are built when the real problems is being solved and with, and ideally when the success can be measured on daily basis. So that's why, and I think I didn't fully realize that when I was 20 years old, when I started my first company. With AI Factory, I think we had a, we had an amazing opportunity to bring the technology which we built to 400 million of daily active users on Snapchat. And so I think this worked out well for everyone. 

Chet Kapoor: That's awesome. So, AI Factory got bought by Snap. . And, and you hung around at Snap for a little bit, right? I mean, you . And so you, and you're delivering AI at scale. What was the biggest challenge that you faced, right? Because you're going from running a company, founding a company, running it, and then you're at Snap. What was the biggest challenge you faced? And more, you know, less the specifics, but more our listeners would like to know and learn from your experience. 

Alex Mashrabov: Yeah, a hundred percent. I think parts of the issue was, I think parts of the issue was also that I lacked empathy. Different, I think the reality, I mean, Snapchat has two co-founders. And one co-founder is technical, second is not technical. And there are basically two different engineering organizations in the company, right? And many other organizations as well. Each of those organizations has its own goals and really building this awareness that different groups may have different goals and different attitudes to work was really difficult. I mean, definitely Snap has a very rigorous privacy processes and Snap values privacy a lot. So sometimes it may feel that there are certain functions which are adding too many layers of checks, for example. And as a startup founder, there is always a desire to move fast. And I think I lacked this empathy, although I was really fortunate to have great advisors. So I think for me, it took around six months to develop awareness and maybe 10 months to develop a full understanding of how a company works. But I wasn't, I believe I wouldn't be able to do this just myself. And I am just fortunate to have great advisors. I think total maybe 20 over the four years, there were 20 different advisors slash mentors for me in Snapchat. 

Chet Kapoor: That's awesome. By the way, my advice is, you know, having acquired companies and being acquired is always the two things you said. One is, don't be too hard on yourself. It does take time to get adjusted and somewhere between six and nine months at minimum, right? Six months at a minimum to get adjusted because you're going from running a company to being part of a company. And then the second thing is, make sure you get advisors along the way. People who've been at the company who are empathetic to getting you adjusted. So that sounds really good. Now let's talk about Higgsfield, right? You're focused on democratizing social media creation for everyone. So I want to use the platform. Can you walk me through a use case on how it works? 

Alex Mashrabov: Yeah, exactly. So let me maybe start with a story, like why we got there. 

Chet Kapoor: Yes. 

Alex Mashrabov: I think I had like this amazing experience at Snapchat where we were overall, we observed that social media shifted from being just a tool to get a very low CPM, meaning getting very low cost per view to actually driving their returns. So, there are many great businesses built on social media where social media helps to fuel its growth. Although we, although videos have up to 10x higher engagement than images and there are, and over the last year we have seen a great emergence of new tooling which helps to produce videos for advertisers from the assets which advertisers have. And most of the advertisers, especially commerce businesses, they have a lot of images, like typically one SKU may have six, from six to eight images. So one of the basic solution is just to do a slideshow. Then there is another layer of complexity is to add an overlay with a pre-recorded human saying how he loves the, how he loves the video. I mean, this is the influencer marketing. Yeah. Although we believe that with AI, it would be, it will be possible to create those videos from scratch. 

Chet Kapoor: Wow.

Alex Mashrabov: And this is what, and this is what Higgsfield is working on. 

Chet Kapoor: and you want to make it so I, as a lay person who is not a video person, not a sound person, but know how to tell a story, I should be able to type in a story from text and then you will create the entire short social media video. Is that a good way to put it? 

Alex Mashrabov: Exactly. A hundred percent. And think about that. I mean, success on social media requires to put five videos a day and you can also think that there is typically there is some level of A-B experimentation as well. So it's not just five videos a day, but maybe 10 videos a day. Oh, and this is the, and this is quite light version. It's hard to organize a physical production process to be able to produce 10 videos a day. I know this myself. I had a, we had a production team in Kiev in Ukraine when I was at Snapchat. And even with in-house team, it's hard to do. It's hard to do production cycle within 24 hours. With generative AI, the production cycle will take hour or less for those 10 videos. 

Chet Kapoor: And that's what you're trying to target for everybody, not just for people at Snap, but you want to do it across all platforms. 

Alex Mashrabov: Exactly. And when, exactly. And so there are two main, I think there are two main differentiators in marketing worlds compared to the entertainment. In the entertainment, very often people try to create something from scratch. They try to unleash their creativity. Although within the marketing teams, first of all, there are many different vendors. I guess on average, there are 80 different vendors. You can, one, there is one vendor for sending emails, like mail to vendors, like Google ads. 

Alex Mashrabov: But I mean, typical team would use from 50 to 80, I guess. And being able to ingest, take this data, existing data, let's say performance data, take existing assets and to plug this to the model so that the model actually learns about the brands  and the model can create videos, on-brand videos without much prompt engineering, right? So basically, building this whole technology stack, which requires, which definitely requires a model, but the model is not just the only piece. It also requires this layer to ingest the data and also the layer to augment the prompts, enhance the prompts. 

Chet Kapoor: For sure. So you're still in early days. When does the product ship? Yeah, 

Alex Mashrabov: so we definitely have seen a great interest from creative agencies so far, and we are working with creative agencies, too. And there is a huge component to that about user education, probably something which I sort of, still surprises me every year. So let me maybe tell a little bit more about that. So when I was 20 years old , my thinking was really like, there is a great technology, everyone should care about new technology. And so that's, it's just important to do kind of have a bold idea to deliver the value. Although the reality, I mean, every year I learn more and more than most of the people, even professionals don't care about technology that much. So I think one of the recent examples in Snapchat, it was that we launched my AI chatbot. And this was probably second largest consumer chatbot in the world, tens of millions of users. And the surprising fact was that it was, it's just hard to explain to teenagers, which is the predominantly the audience on Snapchat. It's hard to explain to them what is chatbot. It's like AI chatbot. What's a chat GPT? They don't know what's it, what chat GPT is. And I think here with Higgsfield video model, we need to do a much better job on, on the education side. So we have, we're working closely with agencies since agencies have this one to many relationships. So we need to educate agency and then agency can spread the words. But also we have a consumer application where everyone can go and try our model. These app grow pretty quickly to half a million users within just maybe one month. And now we are also seeing retentive use cases of people coming back. And now we are getting to substantial revenue numbers from just mobile app as well. 

Chet Kapoor: That's awesome. 

Alex Mashrabov: This is as there are different kinds of types of social media marketers. There are, and I would say there are lots of freelancers, which wouldn't be able to pay, which typically allocates a budget of a hundred, $150 a month for all the software. So those freelancers, they use applications like CapCut, maybe FataRoom, maybe Captions, and typically one application charges for $15 a month for a subscription. And we are definitely, we definitely started to see in June more traction specifically for this audience. And this audience is actually ready to pay. 

Chet Kapoor: That's awesome. That's awesome. I'm glad it's going so well. So we're going to, we're going to shift gears away from Xfield. There is no way we can talk about generative AI without talking about ethical issues, right? And so, specifically deepfakes, right? It's a hot topic. It came up in my conversations at TED. It comes up in conversations I have with the World Economic Forum, but, you know, and especially around, content creation, we've, the elections are coming up in the U.S. and much, much more. So what's your perspective on, on just ethical considerations or ethical issues around generative AI? And so that's your perspective. And then how are you thinking about addressing this at Higgsfield? 

Alex Mashrabov: Look, I think, and I'm, to be honest, I let them be like really direct. I wish, I wish , there is more, there are more initiatives to address those topics. I am, I'm a little, sometimes I just feel that people talk, I mean, media, they kind of use this topic to poke at some company, although I just feel there should be more constructive solutions. So there are, I would want to maybe specifically talk more about video, because this is the space which I believe I know pretty well, although happy to chat about, about, about issues in text generation. So in the video generation space, there are two major issues. I would say let's, with the outputs, there are two major issues. First is the copyrights. 

Chet Kapoor: Yes. 

Alex Mashrabov: Clearly there should be a platform where, which detects if the content is copyrighted or not. And there is no such platform right now, which is, we would love to partner. I mean, we cannot monitor manually. We cannot monitor manually every generated video, although we would love to have a solution which allows us to do that. And it's always good to have a third party solution which introduces high bar for everyone, right? Obviously we have internal systems, but I think we need as an industry, we need third party solution. There is no such solution just yet. And I think this will hold the whole, and this will hold the whole industry back. 

Chet Kapoor: I agree. 

Alex Mashrabov: And this is true, not just for video generation for, but for image generation as well. 

Chet Kapoor: Across the board. I would agree. 

Alex Mashrabov: This is piece, this is first piece. Second piece is actually misused and this piece concerns me even more. So we actually have a model which can produce quite realistic output. So if you're looking, if you're watching a horizontally oriented video on the mobile phone, on the mobile phone, then it's sometimes it's hard to distinguish real video from AI generated. 

Chet Kapoor: Great. 

Alex Mashrabov: Although there is no solution available, publicly available right now for companies like Higgsfields to mark the content as AI generated. This is a huge problem. So the way, how I think about that, that there should be some digital signature which is invisible and which is hard to remove. 

Chet Kapoor:: Correct. 

Alex Mashrabov: We know that there are a lot of, I mean, there was a topic about adversarial attacks of how it's possible to mine, to manipulate those models. And I wish there is more research on that space. And there are actually businesses which implement those AI digital watermark. Very likely, in my opinion, very likely large companies which own the distribution like Google, Meta, Twitter, those companies have to step up and introduce maybe open source standards as well, which may not be, and open AI, yes. I think Meta actually has one, Meta has some solutions, although on the technical side, they don't, they do not seem to be on par with the level of technology evolution. So I believe these two pieces around copyrighted content detection and the AI digital watermark, I think we are falling behind as an industry. Definitely there are open questions about the training data and if that's a fair use to use data from YouTube in training and so on. I heard that YouTube CEO said that if open AI trained their model on YouTube data, then it is illegal. I heard these statements, although it is still..clear how the space will shake up. And let me please expand a little bit more. So we are likely to see huge adoption of this vision of this video understanding model, GPT-4 vision and Google one and many others. Many of those models, I believe that many of those models actually were trained on YouTube data. So that regards we likely are at the point of time where adoption moves much faster compared to regulation. So I'm not sure we will be able to solve issues about training data. I think it's still going to be a mess. Although I believe all the leaders in this space have to step up to provide more security and transparency about the outputs. I think cause like training data is like internal kitchen. And I think there is in the United States, typically government doesn't go really deep in, in the, in like how company works internally, although companies should be liable for the outputs. I mean, that's my take. Cause right now, Meta is liable for spreading some political disinformation, other type of disinformation, the same high standards should be applied with AI as well, with AI generated problems. 

Chet Kapoor: I would agree. You know, what's interesting, that's very well articulated in a very complex problem. Right. And I think, I actually am a little encouraged that people are at least talking about it and I'm not talking about the media, I'm talking about the tech companies themselves. And I'm also encouraged that some governments are thinking about this the right way and starting to engage. What a lot of people don't realize Alex is that, each of the states in the US have actually started thinking about how they use Gen AI for their own state run apps. So people are, the legislation is starting to actually follow through. And so I do think it's going to be, this is not just going to be, you know, meta open AI and Google get together and do a standard. I think there's going to be some regulatory stuff that also comes from, from the country itself. So I think it's a super interesting space to see how we move forward. 

Alex Mashrabov: Exactly. I'm a little bit worried that, I'm a little bit worried that it might be applied state by state, because then it will make the governance much, much harder. 

Chet Kapoor: No, no, I would agree, but here's my take, it is better to get started than not, right? I think, and that's what makes a difference, right? First step, one step and things like that. This is really, really cool. So Alex, we're going to go to the rapid fire stage now, where I'll ask you a question. And if you can give me a short answer, as short as you possibly can. I'll go through a series of questions as we go through. So what's a misconception people have about video AI? 

Alex Mashrabov: I think people, I mean, many people think that this will go, this is going to disrupt the Hollywoods next year, although we have not seen any practical applications so far. That's why we are focusing on social media being one of the simplest form of the contents of the media contents. 

Chet Kapoor: What's one unpopular opinion you have about leadership? 

Alex Mashrabov: I am definitely on the side of radical transparency. And I believe many people are trying to avoid hurting another person, although all these delayed feedback typically causes more harm than good. 

Chet Kapoor: Yeah. In my experience. Who and what have been your biggest inspiration in AI? Just one, who or what? 

Alex Mashrabov: I mean, definitely it's Snowflake. Definitely it's a Snowflake. Snowflake moves quite high in in AI in AI world. And definitely Frank Zlutman is one of the greats one of the greatest sales leaders of our generation. Definitely definitely Snowflake is a very interesting company to follow. 

Chet Kapoor: That's awesome. What Gen AI product or service you think is overhyped? 

Alex Mashrabov: This is a really good question. I would say perplexity. Yeah. Think they claim to have 20 million in revenue and probably they claim to have $2 billion valuation. And I would want to see, and I would want to see clear wins against Google. As I know that many people say that they prefer perplexity over Google. I'm not sure I agree with that. 

Chet Kapoor: That's great. That's a good one. So last question, what global challenge do you hope humans and AI can solve together? 

Alex Mashrabov: Look, I mean, the way, how I think about that is is like 30 for the last 30 years, the brightest people in the world were spending eight hours a day answering confirmation in the computers, basically. I wish, and basically we were giving the inputs we, and I hope that now machines will start to reason and I will, and eventually people will have to spend less time like with entering confirmation in their CRMs and so on, updating spreadsheets. And we'll, and that's humanity will be, we'll be able to focus on key issues. Global warming is definitely one of them, but there are, I think there is a huge spectrum of issues. I mean, lots of new diseases and so on. I just wish we can get a little bit back. So from virtual vaults and value created in virtual vaults towards many more people focused on creating value in real vaults where we live in.

Chet Kapoor: Yeah, no, I agree. That is very, very well articulated. Alex, this has been a blast. Thank you very much for joining us. I think I think the listeners will actually enjoy this. I wish you the very best with Higgsfield. I think you're doing some really cool stuff. I'm certainly going to go and check it out. And look forward to talking to you again. Thank you very much again. 

Alex Mashrabov: Thank you. Bye. Thank you. That this is amazing. Thank you for having me. This was amazing opportunity.

Chet Kapoor: Thank you.