The AI Revolution in Media & Social - Part 1

Aakash:

I think the biggest thing is it’s not a revolution it’s an evolution. Rexus has been around forever, we’ve had AI in media and advertising forever. This is just coming of age and going 10x on that.

Tarun:

I think AI feels different, AI feels different for the simple reason that everybody can actually visualize real world use cases

Aakash:

The pain point for OTT industry has been the economics have just been very brutal across the globe, it’s not just India, across the globe. If you bring down content production cost with all this technology again brings in a new phase of media.

Tarun:

Client doesn’t care whether this content was AI generated, the client cares about is this moving my fundamental metric.

Aakash:

Push more and more to consumer controlled content, them giving their own shape and form to what content they consume.

Tarun:

It will make creators do different things, it will make them more efficient so they will use this to create even more high quality content faster, cheaper, better.

Aakash:

I think a big pushback people say is AI generated or this, I don’t think consumers behave that way.

CV:

The summary synopsis is media and social requires a ton of content which is unique to this segment. AI allows for us to break the quality across quantity curve.

‍

CV:

Hi, everyone, and welcome to another episode of Matrix Moments. Today we’re discussing with Aakash and Tarun, we’ve got a very interesting topic for you. This is AI in media and social, we’re choosing to call this episode Feel Good Incorporated, for all you music fans out there that's a Gorillaz song and we’ll come back to why we named it Feel Good Incorporated. Before we start off I think it’s very important to say that it’s June 20th today so anything that we say might be irrelevant say two to three weeks later. You might say something and then LLaMA will get leaked and then people start hosting AI on their mobile phones, so that's the pace at which things are moving in AI nowadays. I think before we unpack what AI is and how it sort of impacts media and social let’s just take sort of a step back Tarun and first question towards you, you’ve seen so many fundamental technology shifts and you’ve seen that yourself over the last decade or two. So what’s your point of view on AI, is this a fad, is this going to remain, how important this is, what’s going to change for us, is this here to stay?

Tarun:

So, firstly, thank you for doing this. I think I'm glad you started with a disclaimer that everything we say may not hold true and should not be used against us in the future because that's the pace at which things are changing. So I think I was thinking about how many platform shifts we all have seen collectively in our lifetimes. It started with personal computing that is as real as it gets, next came mobile as real as it gets even more value creation arguably. Then came cloud and we’ve got massive, massive bellwether companies with the cloud. I think there was a time maybe 3-4 years ago we were talking about mixed reality, AR, VR, arguably not played out as much but who knows with Apple Vision Pro now that could change. I remember two years a lot of us had a lot of FOMO on Crypto, NFTs. You know, it was hard to argue against it but at the same time it was hard to see real world use cases. I think AI feels different, AI feels different for the simple reason that everybody can actually visualize real world use cases where the application of generative AI in particular can help actually move a business metric. So whether it’s content, quality, whether it’s cost of production, whether it’s speed of production one can see real world use cases of how does this actually make like business better. So who knows I think 5 years from today or 10 years from today we all will be sitting and the data will be in front of us but this one seems a lot more real. I don’t know what you guys --?

Aakash:

Yeah, I think if you look back at those shifts which happened there are two parts to it, one is can you imagine what happens, what are the experiences on business that you can create, what leverage do you get. And second is the readiness like you can start imagining and you run ahead of yourself, I think that's what happened with mixed reality or I would say even to a great extent with Crypto and NFTs. I think with AI there are both those things which are addressed. A, there is imagination, these things have been thought for a very long time, it’s not new firstly. I think the biggest thing is it’s not a revolution it’s an evolution. Rexus has been around forever, we’ve had AI in media and advertising forever. This is just coming of age and going 10x on that. I think which is what makes this shift far more real and I actually don’t doubt that within the next 5-10 years you would see a completely different paradigm of how we think of media and social.

Tarun:

Actually it seems a little bit to your point, it’s a great point actually, we’ve had rule based algorithms since like ever, we all have done that. And different people have called chatbots as AI, people have at some point Siri and Alexa were that and it seems like it’s all happened overnight but that is really just generative AI. Till then there has been a lot of rule based stuff, it’s for the first time with creativity and generative nature of it is playing out which is why we all feel like hey, where did this all come out from. But if you look back like every technology takes 10, 15, 20 years before it becomes bust.

CV:

Yeah, and coming back to a couple of our internal discussions as well we’re already finding sharp use cases and I know people have been developing this for close to a decade now but it’s suddenly mushroomed into seeing very sharp use cases, seeing applications where it has real world implications and I guess that sort of ties it in well. That being said it still feels like in a lot of sectors people are looking for what does AI mean for this sector and the very reason we chose to do this podcast on AI in media and social is we felt that it’s not a hammer looking for the nail. There is a nail that's begging to be hammered in some form and it’s very clear that the implication of AI on this sector is going to be massive. So let’s unpack that a bit but before we do that it will be useful for the viewer’s especially, what is media and social? I know in Matrix it’s sort of a catchall term for a lot of industries and, Tarun, it will be useful to get your point of view on what exactly is media and social and how do we define it.

Tarun:

So I think put simply it’s everything which is content and like you had said everything we waste our time on --

CV:

Which is why this is called Feel Good Inc.

Tarun:

So I mean it’s basically if I take a content backwards view of it it’s basically audio, video, text, image, every format that one can think of and then when I look at the usage of it there’s this whole creator/influencer economy around it and what they’re doing with content which has been actually really interesting how it’s played out in the last few years but one can argue that the creator economy still feels a little shallow because monetization and how it sustains is still unclear, maybe AI will change that. Second is all these content aggregators, we’ve got Dailyhunt in our portfolio, there’s so many different content aggregators out there. And then there’s third is sort of the OTT/media platforms where everything from Netflix and Spotify to everything else. I think broadly when I look at that universe it’s all of these things put together that when I say application of AI to media social here are all the individual subsectors/themes that I think are going to get disrupted through these new technologies.

CV:

Got it. Cash, how do you look at media and social, how would you look the splitting the sector.

Aakash:

I actually think it’s a slightly simplistic one and we’ve seen this play out. There are two dimensions, one is creation and consumption wherein TD mentioned the first phases of innovation of technology. Distribution got disrupted drastically on the creation consumption spectrum. You got consumers becoming creators and which is what led to what you see, we got at a point where more than 500 hours of content is uploaded on YouTube in a day, so that's one dimension. I think the other dimension is purely to do with the production element of it. So if you combine those two and you say that hey, technology can be a leverage on production, technology can be a leverage on this creator consumption distribution part of it and one which I usually think of when over the years people have called it new media, people have called it interactive media. But if you simply distill it to what Tarun said image, text, video, audio, you take all of these forms of media, package them in an interactive format you get what is called games which already have had AI in them for the longest time. So I think media broadly boils down to these three core ways of looking at it and companies can play at intersection of it across section of it, everything.

CV:

Yeah, that makes sense. So like I said everything that you waste your time on is media and social for us but it’s actually a broad bucket, we’ve got OTT, we’ve got YouTube, we’ve got all the sort of I think we would also classify parts of the rating app behaving very similar to this sort of virality K-factor sort of apps. I'm just going to bring us to the point on why we feel AI is going to change things in media and social and we’ve had several discussions internally. The summary synopsis is media and social requires a ton of content which is unique to this segment. AI allows for us to break the quality across quantity curve and it allows for that through something we call modality. I think the next section it will be great if we could sort of break that down, what does modality really mean, why is it important for media and social. And then TD it will be useful – I know we had a deep discussion on the quality, quantity arbitrage changing with AI coming in, so we’ll handle it one by one. Cash, your points of view on modality, I know things have changed from the first time we saw stable diffusion to what things are looking at right now, take us through that journey a bit?

Aakash:

So I think modality broadly means that what mode of expression are you using to translate an idea internal and out for consumption. Four primary modalities, text, audio, video and images. Now one could argue that videos is composed images but let’s assume it’s four for now and if we stick with that what we’ve seen over the decades is text to audio or text to images those two have been the first port of attack for Generative AI. I would say text to audio today has reached a point where maybe we’re a few months away where almost getting to a point of production grade.

CV:

Is there a reason for the maturity to come in those phases?

Aakash:

I think actually you’ll be surprised, it is counter intuitive, right, it is counter intuitive from the fact that – so text to audio is easier because it’s the same language being put out into a spoken form or an audio form. Text to image is slightly more difficult because you have to take a – it’s almost like saying if you go back when you were an operator and you would brief your marketing teams or creative teams. It’s like you would give them a brief --

CV:

And then something completely different would come out.

Aakash:

Something completely different comes out. So I think that's a difference in that modality transfer. So text to audio far mature much easier. Text to image has drastically improved, I think both with people like Midjourney, Stability, everyone or even the models from big tech I think text to image today is constrained by how well you can articulate, it’s as good as the brief some creatives will tell you that your brief was not great, so it’s at that stage. I think text to video is still mid maturity because very different technical challenges of how you solve for it. And we know and we’ll talk about it later when we get to it. It can still be solved, so I think one, people tend to cherry pick just the single open model or what is out there as a creation tool. There are companies who are creating almost like scaffolding on top of these models to enable creators to --

CV:

So are we in for a shock you think, is there a Chat GPT moment coming for text to video?

Aakash:

The Chat GPT moment by the way was actually as Tarun said it was more about that generative part of it. I think that moment has already happened, the moment has already happened with Runway, it’s there. If you think of it as a spectrum – if you want to create a 2-second, 5-second quick one you can do it. Can an individual video creator use that as a point of leverage to augment how fast and well they can produce they can do it. Have we gotten to a point which is saying that hey, can I almost start producing movie grade, no. But which is also text to audio, so text to audio for your own use case you could use anything online, for a more specific use case, let’s say marketing or translating marketing videos you would need more tooling for something like movie dubbing, we need far more tooling. So it’s just a spectrum of how well or how critical is the element of accuracy. So it’s just that, and I think the spectrum will keep getting pushed, maybe we’re in for a shock and it might happen that tomorrow we’re sitting here and saying production grade requires tooling, maybe tomorrow it might not, I don’t know. I don’t want to --

CV:

That's why I said June 20th. But so let’s park that for a second and look at the other part of it which is the whole quality quantity graph just moving upwards. What does that mean, why is that important, why do we feel that's the second big technical unlock that can happen because of AI?

Tarun:

I'm actually going to pick up from something Aakash said here, go back to your time when you were an operator. You would give a brief, your team would come back with something and often it was something very different. By the way that happens even when you're using prompts, it’s just that at that instant you’re able to tweak it and keep fine tuning it with almost real time feedback. And the difference here is AI will actually then learn over time what is it that you exactly mean when you say something. So I think going to your point if I take a 2 by 2 with quality, quantity we have learned the efficient portfolio theory and sort of as part of our investments course in B school. I look at this as a simple – there is a particular curve and as you want higher quality it actually requires a trade off with quantity or vice versa. I think what AI is going to do is it’s just going to keep pushing that curve more and more outwards and an efficient frontier will just keep moving outwards. Where it will be now possible or all of us will have to kind of fine tune our own mental models to understand that if I need this quality is it going to come at the cost of either dollars or time or quantity. Right, and so I think that efficient frontier is absolutely moving and every day like I'm sure all of us see this on our social feeds. Every day I log in and I see the quality of stuff that is starting to come out it’s actually starting to blow your mind. And how quickly you're able to change your prompts and people are saying prompt engineering is now like a different field of science itself. So I think we will see some of that happen, I think the underlying point for me again just going back to the earlier point is see, end of the day I don’t think people are going to care like I was talking to a founder this morning and I was asking – he actually runs a company which is in some ways impacted by AI in a sense where it could either be a massive headwind and the company won't be there or it could be a massive tailwind depending on how quickly they’re able to adapt. And I was asking him, saying that listen, what are you hearing from your clients and he said you know what client doesn’t care whether this content was AI generated, the client cares about is this moving my fundamental metric, like is my engagement going to go up as a result of using this AI content. Is my cost of production going to go down, is my time of production going to go down, can I retain my users better, can I increase lifetime value of my user? That’s what the CMO, CEO whatever it is cares about. AI content hai ke nahe, like beyond a point if it’s there and it’s working for me great but if it’s not there I'm not looking to reduce my spend by 20 percent and that's not the metric I'm looking to optimize.

CV:

Yeah, I completely agree. So I know it’s a bit of a divergence but the amount of time I’ve spent on those Balenciaga viral videos is mad, is unnecessary.

Tarun:

Which one is your favorite?

CV:

The Harry Potter one, multiple of those. But I think what is surprising for me was I liked something and then the time that it took for multiple other types of content to come on the same theme it was just hours of just uploading, right, so in three weeks I had enough content that I couldn’t finish watching. I think that allows for – the minute the quality quantity sort of curve gets broken down it allows for a lot more speed, it’s easier to create content and it’s faster to create content.

Aakash:

I think on quality quantity and obviously that's the pushing the curve, I think with technology so just two interesting tidbits about it. With tech we’ve always seen this frontier push very differently. The curve changes like if you think of bandwidth and compute we have exponentially grown on output and exponentially gone down on cost. So I think that's the unique nature of our technology. The second is what is quality, I think in content we have seen this play out a lot that earlier was treated as very second grade people and media would dismiss it as hey, it has some role to play but it’s not there. Look at ByteDance, it’s almost created a if not for geopolitical reasons a trillion dollar empire out of it where it’s purely about virality and attention. And virality personalization attention which also gets factored into quality, right, no one today says that that is not good quality. So I think both those things will also be there, so I think a big pushback people say is AI generated or this, I don’t think consumers behave that way.

CV:

Yeah. Show me good content I’ll watch it.

Aakash:

That's it. Perception of quality is also going to keep changing.

CV:

Yeah, completely agree. So if we marry all those points together what are the broad themes that we like? So this is the kind of technology shift that we feel makes it super interesting for us to look at companies building in AI in media social. But if I were looking to build in that space what are sort of themes I should keep in my mind to say these are broadly the themes on the consumer side that's going to get unlocked and hence I should try looking at those spaces.

Tarun:

Yeah. So which ones would you, like what’s top of mind for you?

CV:

There’s a lot of stuff around just creativity itself. I know we had this discussion internally, is music production going to take a big hit because AI will create music and we came to the conclusion that it’s not. Any such technology advancement has led to even more creativity which means artists will be able to put out much more creative stuff faster and at better quality, it allows for you to engage with your user in a much more sort of personalized manner. So I think creativity itself gets unlocked, content being such a – if you think about it, the only core pillar of media and social creativity is a big part of content and why you want to watch it which means watch times can go up. Relevance for me would go up and that's what I'm most excited to look at, how AI helps creators and artists create better faster and allows for your consumption as a user to be more enjoyable in some form.

Tarun:

So I’ll tell you my – the way I look at it, right. I think when I saw the first time I think Chat GPT was released all of us spent hours trying to create articles, stories, I actually showed my son how he can write his essays on it. Then Midjourney came out , we were playing around in that and I think the realization at least at that point it was like that aha moment at least in content time and cost to create content is trending towards zero. And that basically is a fundamental paradigm shift which historically there’ve been incremental improvements in how long it takes like today we’re recording this, we very a camera set up, we have people. It requires some pre-production planning, there is a post actual recording there is a post edit etc. But all this can now be done in almost real time. So I think there will be what I think is going to be just automated content creation driven by the fact that the cost and time to create this at high quality because end of the day there is still a consumer of that content and that needs to be --

CV:

So what happens, that will lead to more content which could also lead to more fatigue?

Tarun:

Yes.

CV:

Versus more content could just lead to now I'm more specific on what I want to see and what I enjoy. You think that's the path it will take?

Tarun:

So I think two things will happen, so for example like let’s take – I was reading this term called Robot journalism which is today there are thousands of people in the world who are watching every single game and then are putting out an article on what happened in the game, the key highlights, the key sort of winners moments or whatever it may be. And then there’s a score board, there are some highlights or whatever it is, AI can do that for you. Like why do I need a human to do that, somebody is listening to a political rally and saying oh, in this particular rally here are the five key points that this minister hit upon. And I can just look at that content and say summarize it and say here is the key thing. So I think there will be a lot of robot journalism that will take off. The corollary to that actually and this is in my mind there are two schools of thought, one is this will basically mean that several jobs will become redundant and to our earlier point I actually think and I was thinking does this make creators redundant in some sense. Like if I'm on reels today we all heard about that AI girlfriend whose sort of today making or raking in millions every year on content --

CV:

Yeah, yeah, Cash wrote a super highly engaged post on that as well.

Tarun:

So I guess in my mind we will see some of that but I think the longer term where this is going to settle to your earlier point is it will make creators do different things, it will make them more efficient so they will use this to create even more high quality content faster, cheaper, better. I don’t think this means that tomorrow when I go open Instagram or I open YouTube shots or whatever it is my feed is full of like virtual creators who I have no sort of connect with. So I think I will still have those creators but the best creators will be very smart about picking up these technologies and use that to create even more engaging content at a faster velocity.

CV:

I remember one of these conversations we had when you also mentioned personalization at scale. What does that mean, how does that --?

Tarun:

So we saw Bandersnatch, right, and I was blown and although that was in my mind more rule based because based on what you click the output is different.

Aakash:

If this then that.

Tarun:

Yeah, that was more like if then else kind of statements. I’ve been always fascinated by this thesis and this is something that we had discussed like several years ago which is personalization where n is equal to 1. Where literally every single person sees a – maybe Facebook and Instagram today are some version of that, each of us have a very different feed. But often I like my wife is seeing something in my household tomorrow or two days later my feed is filled with that stuff. And I think a lot of my friend like Sid and suddenly I see it. I think over a time this is going to become n is equal to 1 like each of us will be able to see content which is tailored to us at a level that I don’t think we can even imagine today. And so I was saying that today I'm seeing content which is categorized in a particular way. If I like suspense, if I like family drama, if I like romantic whatever within that I'm picking certain content and Netflix is showing me stuff basis what I watched previously. I'm seeing within that piece of content can I actually fine tune what I see in that content depending on my history. And I think that's the level at which stuff will start getting personalized. And it will all be done real time and it will all be done on the fly.

Aakash:

And I think if you look at the generative models on that personalization point what happens tomorrow and again going back from OTT and content space what happens tomorrow if a favorite show around dragons you want to actually see the character be played by someone else.

CV:

Or a better ending to the last season of Game of Thrones.

Aakash:

So I think it’s not just personalization which is about the recommended systems getting better at n equal to 1 I think control will move back to users. It will push more and more to consumer controlled content, them giving their own shape and form to what content they consume. So imagine if someone could create an idea, turn it on product but someone wants to do more. A lot of these things you always see in gaming long back, gaming has always had a culture of modding the game. And then those mods themselves are popular. I think you’ll start seeing those things seep in like personalization of n equal to 1 I think NVIDIA pretty much already did a big demo of saying where your non playing characters are actually just AI like actually talking normal natural it’s not just preprogrammed. So I think you can draw a lot of early signs from what’s happening in gaming, I would expect in video you will have modding like I would want to see the dragon queen be played by someone else maybe it can happen.

CV:

I'm hoping it happens.

Tarun:

Like we had all these cameo models during Covid which became like seriously viral and people were starting to send greetings and wishes and I remember all sorts of – we had a boys trip and you can imagine the kind of person who created a nice video inviting everybody for it. Today like you think of those and you're like shit, like overnight those business models are completely redundant. How do you then justify now somebody paying for that stuff? And so I think over time as this goes to zero I think there will be business models that we can't even imagine right now that will start making a lot more sense and we’ll actually talk about some of those. We made some investments so we should cover those as we talk.

CV:

Yeah, we’ll do that. So that means creativity tick mark, crazy exponential growth, we see different kinds of stuff happening, maybe different types of monetization, we’ll hit that in a bit. But if you were to move away from social media because there’s also this large stakeholder of OTT where also we expect a lot of things to be moving around. So, Cash, I know you have a lot of experience in that space, just the quality quantity graph moving upwards what does that mean for just movie production, series production. We also see professional media companies also change the way they work.

Aakash:

I think professional – so be it an individual creator, a bunch of rag tag team of few creators or a production company that is extremely professional, I think you would see this element of digitized production or virtual production be used by everyone. I would rather state that today the shift which is happening is that majority of the social production was limited to the larger professional outfits. They were already using and leveraging technology in a very different way. Now you're actually getting democratized tooling, right, the tooling is actually democratizing it to even that solo creator to tomorrow start creating and you even see those examples on Twitter and YouTube already where creators, solo creators, who are at the front end of this and who are actually taking this hands on they’re creating beautiful professional grade content now. Like content piece which you would imagine only coming out of a professional studio. So I think that virtual production theme the tooling is getting democratized. You’ll start seeing, so we have multiple things it’s not just AI. So there you have it’s like a conjunction or a culmination of multiple multiyear trend lines coming together. You have something called as NUFS which is which are where obviously we have a company in Matrix US called Loma which is allowing you to take a image, just use your phone camara, and turn any live setting into a virtual set that you want which you can bring in to a rendering engine and then keep reusing it. You have stuff happening around creating arbitrage. We have companies like Synthesia, like a multiple of those who are able to again say that hey, I don’t need the person or the actor again and again. As long as I can get some critical learning data I can then start creating content directly using just that data and the models allow that. Tomorrow this will drastically bring down the cost, if you think of at least if I look back at the OTT industry the pain point for OTT industry has been the economics have just been very brutal across the globe, it’s not just India, across the globe. It has just become a fight to optimize and just figure out where can you get margin. If you bring down content production cost with all this technology again brings in a new phase of media. So I'm actually super excited about it.

CV:

Yeah, I think I agree with you completely, I think there’s also a point on you might even see new kinds of models exist, it’s not just better content on Netflix it could be what else if not Netflix. You could now suddenly see a new player come and create put out great content and capture consumer attention without having to invest millions and millions of money on just content propagation.

Aakash:

We’ve already seen concerts happen in Roblox, right, so those are again – I think it’s very hard to pinpoint and we’ll try, that's our job as investors we try and find these but I think it’s all eventually the founders would come up with all these ideas and maybe they’re going to be an explosion of what happens now.

CV:

Thanks Tarun, Thanks Aakash, I’m looking forward to the second segment, where we cover stake holders. What kind of business models we like and more importantly what we mean by work flow depth.

‍