theCUBE + NYSE Wired present the East Coast AI Leaders Executive Series | Cody Coleman, Coactive AI

Clips
More from theCUBE + NYSE Wired present the East Coast AI Leaders Executive Series

Cody Coleman

CEO & Co-Founder

Coactive AI

play_circle_outline Coactive platform streamlines image and video content search and analysis

play_circle_outline Entrepreneurial journey from Stanford PhD to Coactive CEO

play_circle_outline The Evolution of AI: Implementing a Model Agnostic Approach for Scalability in Rapid Development

play_circle_outline Unpacking the Impact of AI on Human-Computer Vision and Enterprise Search Capabilities

play_circle_outline New revenue streams from AI capabilities

Info
Transcript

Cody Coleman, Coactive AI

Cody Coleman

CEO & Co-Founder Coactive AI

At AI Leaders East Coast Edition in New York City, Cody Coleman, CEO of Coactive.AI, discusses the democratization of content creation through AI. Coactive's platform simplifies the search and analysis of images and videos, revolutionizing the way enterprises manage visual assets. By leveraging multimodal AI, Coactive eliminates the need for manual tagging and annotation, making content easily searchable without metadata. This innovation drives productivity gains, reduces costs, and enhances content quality for customers in media, retail, and social platforms... Read more

explore Keep Exploring

What is Coactive doing with recent advances in AI, specifically multimodal AI? add

What was the journey like for the speaker in creating Coactive, and who were some key figures that influenced their work? add

What approach did the company take in response to the rapid evolution of AI technology and the need for scalability? add

What benefits does AI bring to how machines perceive the world and how does this impact the enterprise and customer experience? add

What are some of the possibilities for organizations in terms of utilizing images and videos for new revenue streams? add

bolt Powered by CUBE AI

Cody Coleman, Coactive AI

search

>> Welcome back everyone to theCUBE's coverage here, New York City for AI Leaders East Coast Edition. Earlier, two months ago we did it in Silicon Valley, connecting Silicon Valley and Wall Street with the NYSE Wired communities with theCUBE together connected and bringing all this content value. I'm John Furrier, your host of theCUBE. Obviously, we're at the balcony above the New York Stock Exchange, our East Coast location at the NYSE Cody Coleman CEO Coactive.AI. Cody, great to see you. I just saw you at the Madrona event, the IA-40, which is their signature event. They do a study every year. You were a presenter of one of the AI innovators. Congratulations. Stanford PhD comes into AI, makes a natural progression. Great to see you, and congratulations on the speaking at the event.

Cody Coleman

>> Yeah, thank you, John. It's great to see you again.

>> All right, so you've been on the road. You were just, I heard when you came in that you were at the NLA for the tech week. Was that the A16Z kind of event? How'd that go?

Cody Coleman

>> Oh, it was awesome. It was awesome. So I mean, LA Tech Week is a massive event, so A16Z is one part of it. I was actually super fortunate to be able to kick off LA Tech Week with AWS on a panel at Amazon Studios. So it was just awesome to see how the creative atmosphere and energy of LA is actually coming in and embracing technology and specifically AI.

>> What was the vibe of the show? Share your thoughts on what the vibe was like there.

Cody Coleman

>> The vibe of LA Tech Week was... I think a lot of people are curious about AI. We've kind of gone over... Or maybe taking a step back. When I think about AI broad adoption, it happened less than two years ago. And 2022 ChatGPT was released and everyone was rushing towards AI. And what we saw was that in 2023, there was a massive kind of rush towards AI in terms of consumer and things like that, but in media, entertainment, so especially in LA, there was a lot of fear around that. You had the writers and actors on strike and things like that. And even going into 2024. And now that 2024 has actually hit us, is actually starting to see this kind of acceptance and embrace of AI and it's going from this existential threat to something that actually is critical for being able to produce content more efficiently, especially now that we need to produce more content than ever.

>> Yeah, it's interesting. You see people hear, I hear words like we now have clear line of sight on blank or whatever they're in. So you start to see people get it, they understand where they can apply, they know how to get into it, and there's at least a path. Clearly some of the stuff still out there. Agentic systems, which we've been covering a lot of on SiliconANGLE and the key research still down the road, but we're seeing the preparation, the progressions happening, and it reminds me of the web. Online population grows and functionality gets better, so the adoption will drive certainly more capability. Obviously the chips are getting better, faster, cheaper. Clustered systems. So that's great news. Let's talk about your company because we've had lunch together, we have multiple conversations about what you're working on, LA and New York, two big markets for what you do. Media, you're doing some really cool stuff in AI. Again, we're at Stanford, you've been in machine learning, you've been in the AI world, you know a lot of the folks who wrote the papers involved in some of the deep learning tech, but now that it starts to bust out on the scenes. You got your startup and a lot more opportunities are coming your way and happening. So give a quick overview of what you guys do real quick for the folks who don't know, and then we'll get into some of the AI stuff that you're working on.

Cody Coleman

>> Awesome. So fundamentally at Coactive what we've created is a platform or almost an operating system for content that makes it easy to search and analyze images and videos. And taking a step back when we think about what is happening right now with AI, the internet, as you mentioned, the web, it democratized content distribution. And now with AI, we actually have the ability to democratize content creation and there's just a huge push to produce more content than ever. The fact that every single one of us has a camera in our pocket now, the fact that it's become so easy to capture information and things like that. But the challenge is despite the fact that we've made it so easy and cheap to capture content, to create content, it's still difficult to deal even the most fundamental and basic things with content. For example, we focus a lot in retail and media and entertainment, and we were talking with one large food company and they actually have 4,000 photos of strawberries, just strawberries. And it wasn't because they were trying to find the strawberries' good side or the right angle, it's because they couldn't find the other 3,999 photos of strawberries that they had already taken. And then the same thing happens immediate. It's cheaper for a production company to hire a helicopter and a photographer to take photos and videos of the New York skyline than it is to find the assets they already have. And the challenge there is that it's been an incredibly manual process up to this point where enterprises have had to tag their raw assets either through human or machine annotations, load those annotations into their systems as metadata and search based off of that, which is an incredibly slow, expensive and inflexible process that barely scales to the content that we're producing today. And with the rise of generative AI just won't be able to keep up anymore. And that's where Coactive comes in. What we're fundamentally doing is that we're leveraging the recent advances in AI, specifically multimodal AI, to flip that process on its head and go from this world of tag load search to being able to load and index your raw assets and make searchable with no metadata, no tags required, and then also being able to power all of your other downstream processes in a highly scalable and secure way.

>> Yeah, it's like you're taking the productivity gain, but also putting a platform in place to make the service better.

Cody Coleman

>> Yes, yes.

>> So how's it going so far? Who are you talking to? Give us some examples of other customers. Who are you meeting with? What are some of the conversations?

Cody Coleman

>> Yeah, so we focus on three large verticals right now. So there's community and social platform. So one of our customers is Fandom, where we actually help them do trust and safety. So Fandom is the world's largest fan-generated content and entertainment platform in the world with about 350 million active users, monthly users. Previously, they had to review every single image that was uploaded to Fandom to see if it actually satisfied their community guidelines. With Coactive, they were able to codify those guidelines, take their process that took them 24, 48 hours down to 250 milliseconds for that first review, reduce their overall cost by 50% in 30 days. Additionally, we also end up focusing on retail companies and media companies. So we're working with large Fortune 500 retail media companies. There, what we see is both in retail, one of our customers, they use this for their marketing platform. These retail companies, they have millions of assets. They're producing hundreds of thousands of assets every single season in order to capture the latest trends and latest lineup, but they struggle to find even the most basic things, like strawberries. So there we can come in, we can actually index and ingest all their content, make it searchable in less than a second, and then we can do the same thing in media entertainment, but we can actually take it a step further. So we're actually working with a massive media entertainment company, Fortune 500, Fortune 50, where we actually help them generate minute-by-minute cut sheets so that they can take their ratings information and they can join that metadata that we generate at a minute-by-minute level, and join that with ratings data so that they can improve their content week over week and day over day.

>> So data impact, content story creation.

Cody Coleman

>> Yes.

>> So talk about the dynamic that's going on between assets, content is assets and workflows. Because what you're getting at here is you're attacking the efficiency of the asset development and consumption and the workflows involved in getting the asset. That seems to be the secret flywheel here.

Cody Coleman

>> Yeah, exactly. Exactly. It's actually making assets live up to their name because in actuality, when you think about images and videos at these enterprises, it actually ends up being more of a tax on the organization than a real asset. Because when we think about image and video data, just storing the data is an incredible, super expensive because of the sheer volume that image and video data takes up. For example, I always love this kind of analogy. When you think about big data today, it's largely structured data, tabular data, thinking Excel spreadsheet. And 10 million rows of tabular data, that's roughly 40 megabytes. You think about 10 million documents, like 10 million pages from Wikipedia. Now you're talking 40 gigabytes. That's three orders of magnitude different. That's like going from the surface area of Lake Tahoe to the surface area of the Caspian Sea.

>> Take me through the scope and magnitude of how to get started. So let's just say that I'm a media entertainment company. Could be sports, could be something else. How do I engage? What's the engagement that you have with the customer? Take us through the process. Are there mechanisms involved? Is there deployment scenarios? Take us through how you guys engage.

Cody Coleman

>> So our goal is to make it easy to search and analyze visual content. So we actually meet our customers where they are at today. So by and large, when you think about image and video assets, they're stored in some form of blob storage. So if it's in the cloud, that might be something like S3 or Google storage bucket, and that's kind of where they're sitting at today. And then actually getting that through the entire workflows into their systems, whether it be for analytics or for dams. There's a lot of work that goes into it, but we've started to make it really simple and easy, and it's a three-step process. So first, people give us a name for their data and they can carve it up into a bunch of different kind of setups depending on their organizational structure and depending on the data, then they point us to where it exists, whether that's in the cloud or it can be even a CSV of URLs. And the last and most important step is that they give us read permission to access that data and it's revocable at any point in time because from the very beginning, these are called assets. These are IP. It's actually really critical to the business that they're protected. So we've invested in data privacy and security from the very beginning. So they give us read permission to their assets that they can revoke at any time. And then the system, Coactive's platform does all the heavy lifting, everything from pre-processing those raw image and video assets to actually to prepare them for ingestion and indexing by a foundation model, by an ML model to doing that process in a highly scalable and distributed way to actually making sure that the artifacts are in the right data system for them, whether it be storing it in object store, storing it metadata, relational database, no relational database or vector database, and then connecting it through our APIs to their actual end-to-end workflow.

>> On the customer productivity gain, what is the main value proposition? Just kind of cleaning house, getting the foundation set. You come in, look at their assets first, you assess. Do you then give them a recommendation or is there usually a predetermined workflow and workload?

Cody Coleman

>> Yeah, so with our customers, we see them driving value in a variety of ways, reducing risk, like in the case of Fandom, where they can take down this problematic content so that they provide a better experience to their communities and also ultimately can advertise against that content in those pages. And the case of a marketing team, and for a retail company, we're saving them from having to spend tens of thousands of dollars to go out and do completely new photo shoots because of the fact that they can find the assets that they already have. And then in the case of media entertainment where we can help them figure out how to improve their content week over week or day over day, we're actually driving top line. And in another case, we're actually even helping people do content licensing, where directly, the more content that you can find, the more content that you can sell, and ultimately the more money that you make on it.

>> What I like about your business, Cody, is that assets and workflows are where the value is you're essentially allowing them to merchandise their assets-

Cody Coleman

>> Exactly....

>> in a way that's efficient. But now the front lines, whether you call it the creatives or the business folks can deploy without worrying about lag.

Cody Coleman

>> Exactly.

>> Like the lag involved and just retrieval, Hey, where's that video file someone shot? Ah, shit. I threw that. Which drive is it in? Which cloud is it in? I mean, there's a lot of that going on.

Cody Coleman

>> 100%. Yeah.

>> Okay. So what is it like for you right now? Let's talk about your entrepreneurial journey, because right now you're living the dream, you're on stage, all the events, you're with a lot of the power players, but you've got a business to build, you get some funding. Your milestone is get the next round, get some traction, product market fit. What's the journey been like for you and take us through where you are in the progression.

Cody Coleman

>> Yeah, I mean, it's been an incredible journey for me. It's been a dream since I was a little kid. I was one of those kids that I'm a nerd at heart. I was glued to the science channel or Discovery Channel growing up and just love the idea of building something that would have an impact on society. And then I was super fortunate to go and do my PhD at Stanford where I was advised by Matei Zaharia. So Matei is well-known for creating Databricks-

>> Spark, Databricks. He's a genius. theCUBE alumni, by the way. Great guy, Matei, good to see you.

Cody Coleman

>> Yeah, exactly. And it was great working with Matei in order to actually understand from him how to think about technology and also to think about how do you go about this process of doing category creation. Because fundamentally what we're doing at Coactive is that we're really trying to unlock the value of these images and videos and turn it from being a tax on organization into an asset that generates new revenue streams. So that journey, it's been interesting because taking a step back in time, we started in 20 21. I defended my PhD on June 7th, 2021, took Friday and Monday off, and then we were off to the races. But back then no one cared about AI. No one knew about embedding vectors. We talked to VCs and we were like, we want to do this kind of embedding-based approach to being able to process content based off of the experiences that I had working at Pinterest and Meta. And VCs were like, who cares about embedding vectors?

>> And by the way, back then it was mostly a fight about undesirable content or spam or some sort of thing that you had to clean the mess-up. It wasn't a proactive issue, but the tech is still the same. I mean, this is kind of the background. Facebook, Meta, I mean, they spent tons of tech on that. Pinterest, same thing.

Cody Coleman

>> Exactly. That's where the original inspiration was because my whole PhD was trying to democratize AI because back in 2016 when I started, I could see the future. I could see that intelligent applications were the future and that AI would power the workflows, the applications of the future. But the problem back then, and we're still facing that now, is that the barrier to being able to create those applications was so high. And it was awesome to see at Pinterest and at Meta, how they were actually able to leverage AI in order to drive business value with their content, whether that was recommendations, search, advertising, analytics, copyright protection, protecting the safety of online communities. But what I also saw is that it took a tremendous amount of resources in order to do that. First in terms of human resources. So a bunch of PhDs like my co-founder and I, as well as computational resources to be able to process this data at this scale. Images and video are like six orders of magnitude larger than the tabular data of the big data movement thus far. It's like going from the surface over Lake Tahoe to the surface area on the Pacific Ocean.

>> I mean, the petabytes and the zettabytes are off the charts because your surface area, as you said, it's also harder to ingest that. You need horsepower.

Cody Coleman

>> Yeah.

>> You need a lot of cash.

Cody Coleman

>> Yeah.

>> You guys have a big CapEx budget? Are you using the cloud? What's the back end look like for you guys?

Cody Coleman

>> Yeah, so that's a really great point. So during my PhD, I was actually able to see the front lines of computer systems. So my dissertation, of course, was even called resource and data efficient deep learning. So thinking about how we can do this very efficiently, and I was super fortunate to be one of the co-creators of MLPerf. I created a precursor called DawnBench and was able to see just how expensive training and inference were on this.

>> You can get a dollar for every download of MLPerf. You'd be a zillionaire by now, by the way. Congratulations. I always love that project. That was a success.

Cody Coleman

>> Thank you.

>> I just always like to point that out. That's definitely impacted a lot of developers. So again, the point in time, but at that time that was tooling. Things are getting going, okay, developers are hot. Now, you have now platform specific scale. So this is what I like about what you're working on. You've taken on a hard problem from a data size perspective. Never mind the business logic behind it.

Cody Coleman

>> Yeah.

>> I mean, that's a big issue.

Cody Coleman

>> Yeah. Yeah. I mean, the data processing again is incredibly difficult. It's like when we think about the tools that we have today for data lakes versus data oceans of visual content, it's like having a rowboat or a canoe. It is fine for crossing a lake, but if you told me to cross the Pacific Ocean, I'd say you're crazy. That you need a bigger boat. And that's fundamentally what we're building at Coactive is that bigger boat. But in addition to that complexity, just from the sheer systems and scalability piece of it, you also have the fact that AI is evolving so rapidly. So actually from the very beginning, we decided to take a model agnostic approach. Rather than building kind of a base foundation model, we designed our system to be modular so that we could actually keep up with the rapid pace of AI development. Because even back in 2018, there was this awesome blog post from OpenAI about AI and compute where they made a comparison back to the seventies or eighties when computer chips were improving every 18 to 24 months. We were doubling the amount of computation that we had, and Moore's law was created. They saw the same doubling happening when they looked at the amount of floating point operations and these training models where it was doubling not every 18 to 24 months, it was doubling every 3.4 months, basically roughly every quarter. The amount of computation that was going into these foundation models was doubling, and I think we should call it Altman's Law after that. And we've seen it play out since the study was started in 2012 to where we are today, where these models have gone from thousands and millions of parameters to tens of billions of parameters and more.

>> I always say tokens are like electricity for AI. They run everything. The innovation and cost per token has been plummeting. Context windows are opening up on number of tokens. The number of parameters are in the trillions now. So on the consumer side, the large language foundation models are getting bigger. Computer vision is hot right now too. Starting to see that. Now you have the small language models in the enterprise where you have these domain-specific assets like video, audio, et cetera. Multimodal have to be also ingested in and connected into the other models. How do you see that market emerging? Because at some point the data has to start talking to each other. I have to index things specifically. I mean, we've seen on our CUBE business, in 2021, we moved everything over into neural network format with vector embeds. Again, we didn't have to pitch anyone because we don't have any investors, but we knew that we wanted to have a content creation vehicle for the video that would be easy to help us produce content faster, simple use case. But now it's opening up more avenues. What are you seeing when you start seeing your customers start to identify this need for value unlocking? And two, once they get there, everyone seems to have an aha moment. Well, I have other too, what are you seeing there? Because I like the fact that there's low hanging fruit easily to get some productivity gains, but then when they realize they have a scalable platform, they usually pop up with a, wow, I can do this now too. What are some of those things that you're seeing on your side? Because that really is where the dots connect to the customer. Okay. Okay. Pilot worked. This is great. Now I want to apply it to blank. What are some of those conversations?

Cody Coleman

>> Yeah, that's a great point because it's such a special moment right now because it used to be that as humans, we saw the world through vision and machines saw the world through bytes. But because of AI, those worlds are blurring together and machines can finally see the world like we do. And that's giving us a whole new set of superpowers when we think about the enterprise and the customers that we serve. Being able to find a clip, a piece of content in less than a second, not hours or days. Being able to almost develop a spidey sense to detect content that violates the safety of online communities or editorial guidelines, to, of course, being able to turn these images and videos from a attacks on an organization into an asset in new revenue streams. So we really see kind of initially for us, people will see the night and day difference in terms of search or the fact that they can now bring structure to their unstructured data. And that opens up a whole world of possibilities where we see these enterprises pulling us into a bunch of different things. They might start out with images and then they want to process video. We're seeing people where we start with this content engagement, but then their advertising team is like, hey, let's take all of these commercials that we have and create a predictive thing to say, is it better to have the logo of the brand and the commercial at the first third, the middle third, or the last third of the commercial in order to drive engagement and to be able to give people feedback in the content that they're creating. So it's opening up a new set of superpowers as the digital world and the physical world are now blurring together.

>> Well, Cody, great to have you on theCUBE, great to see you, and congratulations on your success. And I still think that you're in this small little beachhead that will grow like this because multimodal media is going to be the lingua franca now that we can speak machine language basically through prompts and soon to be voice. Find you that picture of the strawberry, the best one that I like.

Cody Coleman

>> Exactly. Exactly.

>> Not the one that my colleague likes, the one I like, the one from the side. Now this is where it starts to get really cool, and I think you guys are on something really big. Great to see you. Thanks for coming on theCUBE. Thanks for coming to our East Coast AI Leaders with the NYSE Wired, Brian's team there. It's great stuff. Thanks for coming on.

Cody Coleman

>> Awesome. Thanks.

>> Okay, thanks. I'm John Furrier here on theCUBE. We are at the NYSE on the balcony above the show floor. A closing bell will be coming in an hour and a half, so we'll get that for you. And of course, all the best content here on the East Coast in New York. I'm John Furrier, your host of theCUBE. Thanks for watching.