Trust, technology, and tomorrow: How AI will change media

Welcome to this week's episode of what's at Stake. We're your hosts. Ylan Mui, managing Director at Penta.

And Andrea Christensen, Partner and Head of Penta's AI Task Force.

We are joined today by my former colleague at CNBC, sally Shin, to talk about all things AI and media. Sally recently started her new role as a venture partner at Comcast Ventures, the corporate VC arm of Comcast. She was also the co-founder of Rave, a startup developing foundational AI models for image and video that protect intellectual property. Previously, Sally spent two years as a scout for Kleiner Perkins, concentrating on consumer AI, online marketplaces, fintech and social and online media, and Sally served as an executive editor at NBC News and the San Francisco bureau chief for CNBC, where we got to work together. On top of all of that, as if she weren't busy enough already, sally recently started a monthly podcast with her friends called the Great Chat, which focuses on developments in tech, ai, business and media. We're so glad to have her on our Great Chat. Sally, welcome to the podcast.

Thank you. Thank you, ladies. I know this is like such a fun moment to be reuniting this way.

Yeah, it's great to see you and to see all the different things that you've done in media, in tech, really leaning into AI. So I guess if you can just start, Sally, by kind of catching me up, catching the rest of us up on what you've been doing, because you recently took a position as a venture partner at Comcast Ventures and I'm trying to understand what it is you do there and how you got there.

Yeah, I feel like it's like a full circle moment for me, because the first decade of my career I worked in journalism with you, obviously, Ylan, and so got to see the front row of that.

After that, I joined a music tech company as an operator and then two years ago, as you mentioned, I started a company called Rave, which we do our own foundation models, and so really been in the crossroads of AI and media for the last many years, and so I got connected to Alison Goldberg at Comcast Ventures throughout my journey, and so we spent a lot of time together.

Now I am a venture partner where we look at every company, from seed all the way to growth in various verticals, from data and AI all the way to healthcare. We look at robotics companies that are strategic to Comcast and Comcast initiatives, so we get to talk to some of the most great founders in various spaces. I spend most of my time in AI just because that's the expertise I bring to the table, and I also I spent. I'm in San Francisco, where you know a lot of the activity is happening, and so I bring a lot of the deals, I help on the diligence front and just keep up with a lot of the things that are going on in the AI space.

So, sally, that's fantastic, and I was just actually at the Comcast building in Philadelphia and got to go in the little sphere and that was very cool, so awesome that you're doing all that cool stuff. So, as you obviously know, ai has been a disruptive force in media. Everyone's really focused on it right now. You know, and we've seen everything from AI generated journalism to deep fake controversies. From your vantage point, how is AI reshaping the current media landscape, and is it good, is it bad? How should we be thinking about it?

Yeah, I was just looking up.

I think it was 12 years ago that I don't know if you guys remember there was a hack for the AP Twitter account where it said that there was a White House blast and the stock market crashed that day.

It's been 12 years and now the technology has gotten so good. So if you look at some of the video content, image content, if you're just scrolling through Twitter, it's really hard to almost see with your naked eye, unless you're looking at it closely, if they're real or not. So I think there are a lot of those things that need to be, you know, guardrailed, of course, from the platform side, and there's a lot of companies that are helping combat this. Whether it is, you know, you have some sort of a fingerprinting for images that allow to show that it's AI generated, but ultimately, you're going to have those bad actors that are going to use it. So I think those are things that are we're still at the nascent stage. I think we're starting to see some of those pop out here and there. So I think people need to be a lot more diligent when it comes to news gathering, but also just consuming news or content.

Yeah, and that's really interesting. You know, I've had some conversations with watermarking startups here and there and there's a view among some that bad actors will figure out a way to get around the watermarking. So it may not be, and that may be the same thing as fingerprinting as you're talking about it, but what do you see as most promising out there? And then, in addition, from the kind of consumer of information standpoint, how should people think about interacting with information they see online?

Yeah, so I know there was an initiative started by Adobe on the image front and it's a consortium of a group. I don't know where it is today and where it stands, but there's certainly industry folks that are coming together to help combat it. I think you know, as a consumer you just have to be vigilant about. You know, bringing trust back into media is going to be kind of the integral part of all this. There's such a proliferation of content across so many different platforms You're getting bombarded with information all the time and you look at, you know content from both, you know Twitter or you know various blogs. The kernel of truth becomes so much more important. So how do you actually map those information back? So I think there's going to be some need for trust back into some of the traditional ways that information is disseminated and you know that. Obviously that that definition has also evolved, but I think that that's going to be sort of the important part of the news industry.

That's going to be sort of the important part of the news industry. Yeah, I feel like, as you were describing, 12 years ago there was a lot of concern amongst news organizations about, one, what AI would do to their own content, but two, whether that would undermine consumers' trust in the content that they produced. But now I also feel like, even as those concerns remain, there's also just a real desire to lean in and leverage AI in order to produce new types of journalism, new types of content, to interact in different ways. Are there developments in media that give you hope that these two, the very old school journalism shoe leather can be integrated with the most cutting edge AI technology?

I think I'll say it in the like how I use AI on a day-to-day um.

The info gathering process is so much more efficient and so if I'm curious about a topic, if I want to search, um, you know, companies like perplexity has really replaced my own search engine, so I use.

You know I go to perplexity before I go do a Google search, um, so I think, uh, you know the go to a perplexity before I go do a Google search. So I think you know the like the early step of news gathering. That's going to change, that's going to make it more efficient. I also, you know, some of the tools I like to use every day are, you know, claude from Anthropic or ChatGPT, and I just do a quick edit on my email just to make sure that it's grammatically correct, and now it's sort of an afterthought of like using it as those products. So I think everything across the office, the newsroom, is just going to get a lot more efficient in that way that allows for more time for reporters to spend time, you know, developing a story, doing some of the manual work that you can't do with technology, and so hopefully that arms them to do greater pieces of journalism.

One of the other challenges that we found with AI and news media is obviously one of copyright. There's a lot of concern about whether or not some of these models are foundational models or like hoovering up essentially copyrighted content intellectual property. You saw this potential pitfall of the technology and that's part of what led you to create your own foundational model, rave. Tell me about that and how it worked and why you started it.

Yeah. So the way that we approached our models was we trained our content on licensed content, and part of that was we wanted to work with enterprises and so the outputs of our models were commercially viable. So this was on the image and video side. I think when it comes to LLMs, we are seeing a lot of deals across publishers and platforms. So OpenAI did a five-year deal with News Corp. That was $250 million, I think. Axel Springer has done one Associated Press, but then there's also lawsuits with OpenAI and New York Times and a lot of the platforms don't disclose where they get their training data set.

That could change based on where the administration is going, and that could change based on where the administration's going. I think it going to be. There needs to be ways for publishers to be compensated for that, especially, I think, when it comes to like news organizations, where you're going to get net new content daily, and especially if it's like an investigative journalism let's say you're I don't know, you are the information and you had this exclusive story and you want to be able to get compensated for those. You know the work that went into developing those stories, the reporters that researched those stories, so there needs to be some sort of way that news orgs still need to be compensated for that, and that was part of the thesis of what we did at Rave quickly to something you said about media and trusted media.

I was recently at a AI dinner with a couple of people from the media industry, one of whom was talking about how they were thinking about using AI to better personalize content for readers. So it could be something like your local weather, local news, your local politics, economic news, whatever you might want it to be and one question that came to mind was in a world where we're already concerned about filter bubbles and people opting into the news that they want to hear, how do you think about AI being used for a narrower personalization and how that jives with concerns about polarization and filter bubbles?

Yeah, I think when it comes to local news, it's really interesting because obviously we've seen, you know, decline in local news, but it's also now AI is allowing us to get more hyper-personal local news across the country.

So there's a company, there's a startup called Hamlet that is going through various like court hearings or local um, local city council meetings and then pulling out information that allows for people of like local cities to get information um, that hasn't been like possible before, and you can actually track various hearings and these. These city meetings are like hours and hours of long, and it allows you to now summarize really quickly what you need and you can track. So I think there's a lot of benefit to that. I think, when it comes to political leadings as one area, I think it was the LA Times that announced that they're going to start using AI tools in the opinion page. So if you're served an opinion piece on something, they're going to serve up two or three other opinion pieces that have a counter view. So I think that's an interesting approach.

I think where the problem lies is that the technology is still so young, and so how you're training those models to respond, what is the benchmark for, like where the political leaning is, that's all going to dictate how and what kind of information you're served to the reader. There's also a company that's called Particle, from Sarah Bakefor, who used to work at Twitter and she has a news aggregation app. That's, I think, still somewhat in stealth, and so every article, every story is summarized based on what the political leanings are. So there are a lot of companies, big and small, that are combating it or that are trying to serve various viewpoints, but I think it ultimately comes down to how the models are trained or how the benchmarks are done. So it'd be interesting to see, but I think the approach is the right one.

But it's so interesting that you bring that up that AI can actually help, potentially help solve the problems that our increased personalized filter bubble world are creating. It takes AI to solve AI maybe.

I'll correct it by saying I don't think it solves the problem. I think it helps add a layer to the issue.

Do you, do you think that people trust the content and the answers or the information that they receive from AI, because we've been talking a lot about sort of trust in the traditional news media, which you know has has ebbed over the years. Do you think that AI has a high level of trust and how can? How can they improve their trust level with consumers?

Yeah, I think that's why I love platforms like Perplexity, because they actually source back where the answers came from, so you can actually, when you serve up the question, you know, let's say, hey, you know what happened at the White House briefing yesterday and it'll pull all the different sourcing from the various websites that they got. You don't see that with ChatGPT or with Claude, and there are sometimes, like, the reason why I would go look at event, ask a query related to a current event on perplexity over some of the other platforms is because I like the sourcing. I think it's also again like we're still just learning to use this technology as a consumer.

So I just have this like inherent way of like thinking about OK, where did the information come from? That's really just like our journalism background, like that's how we're trained to ingest information. But I think over time it's going to get very good that you're going to be able to trust the queries that you get.

What we've seen, alongside the rise of AI and all these sort of new ways of doing journalism, quite frankly, and of communicating with audiences, is also, of course, the rise of the influencer people who are helping to shape the news. They may not be traditional journalists, but maybe they're news adjacent and so they're still able to, you know, influence the way that people think about topics, the way that they shop, the way that they consume and the way that they digest information. Do you see those influencers being better able to leverage the technology that's out there more than traditional media publications have been able to historically?

I think, for one, the term influencer is sort of confusing because I think some of them, you know, spun out of big news organizations and are, like, have been doing great work within their own media orgs and then, you know, launch their own, their own media orgs and then, you know, launch their own. I think there's also the like new category of people that are working in the news that are doing influencer work. Of course, like, when you have less guardrails from you know various editors to like standards of reporting, you have a lot more flexibility and right now, like, growing the audience is kind of their key part. So you have, obviously, the people like the Joe Rogans of the world that are just vastly growing their audiences. Ultimately, where people are going to stick around is when the information is correct.

When it comes to news specifically, and that's the sort of the category I'm talking about. And so specifically, and that's the sort of the category I'm talking about and so you know, you'll see kind of a lot of like rise of creators or influencers, I think, that are hacking the system to just get a lot of audiences, a lot of subscribers, onto their platform. That's where the app dollars are coming in. But you know at the end where people are going to stick around or where they can start trusting the content or where they can start trusting the content.

So, as a venture partner at Comcast Ventures and a former scout for Kleiner Perkins, you have a unique view of the investment landscape, and so I'd be curious what are you looking for in investments?

Yeah, I'm particularly interested in various ways to like tell stories, and that could be in content generation. That's helping fuel the next, you know, film and editing. There's some really cool tools out there. I'm spending a lot of time in both voice and in world modeling, and so, you know, voices like companies like 11 Labs have been obviously doing a great job. They are you know. Voices like companies like Eleven Labs have been obviously doing a great job. They are, you know, when it comes to voice, voice is so like, deeply personal and so you can really engage with your audience. And it's a new tool of just interacting. Subtitles in real time, where you know maybe there's some you know films that didn't have the ability to actually get subtitling done, and now you can use these tools to help disseminate content in various languages in real time. So I think those are really fascinating. I'm also spending a time, a lot of time, in world modeling. So you know Fei-Fei Li, who is at Stanford, just launched her company. There's various companies that are doing this.

This is one beyond video, so you're trying to understand the 3D world and how to interact with the different worlds.

It really was a focus around gaming, but now you can think about how, you know, various people can look and watch films in different perspectives.

So me standing here and Elon you standing on that side, can we look at the scene in a different angle, so that your vantage point is different from mine, as if we are really in that world?

And so when you start to be able to immerse yourself in these films, it's really really cool. You start to be able to immerse yourself in these films, it's really really cool. And you have to understand that when you turn right and then you turn left, you're going to still have that like I don't know, there's a vase and a flower in the background that that's going to stay, and those are all you know parts of like where the training comes, that that is getting more advanced, and so, again, these technologies are still fairly nascent, but they're evolving very quickly. Just think about like, where, like early days of OpenAI's DALI was to, where some of the like image generation platforms have gone over the last two to three years, video is getting very good and a lot more cost efficient, and then I think the next frontier is going to be around world modeling here is going to be around world modeling.

So a couple of years ago there was this really popular moving film where you put the 3D glasses on and you had this experience of crossing the US border. I forget the name of the film and so I'm just more curious because this sounds so cool. Is this kind of a technology that could be applied to like past movies and you can suddenly go into the broader world? Or is this something that could only be applied to like past movies and you can suddenly go into the broader world? Or is this something that could only be applied, sort of, to content going forward?

No, I think, don't quote me on it but I think you can. You'll be able to take past technology and do it similar to right now. You can have a 2D photo and animated and bring it into a film. I think you can start to look at be able to do that with various scenes that you have where it can be interactive. It might need additional prompting just to give some perspective and depth, but that's part of what generative AI is allowing it to do, is it's allowing to predict what the next move is going to be in that scene, and so it'll help create that world for sure.

That is so fascinating. I will admit that I feel like I'm the boomer in this conversation here because I'm still at the beginning of my AI learning journey, if you will, but that sounds so fascinating. And there are applications beyond entertainment, of course. I mean you could think about trainings you can think about. We do a lot of in DC. We focus a lot on sort of getting policymakers and lawmakers out to see what's happening in their districts, visiting companies, visiting a manufacturing floor. You can see certainly amazing applications there. I mean something like that could truly be transformational. It feels like Avatar.

Yeah, and also now there's all these tools like Cursor and Lovable that allow people that have no engineering or coding experiences be able to build apps and different platforms overnight, and so you're going to just see so many different applications pop up so quickly, and it's going to be just like fascinating to see what people can create, like I. There's all these. You know, you might be a product manager at Facebook. You have no coding experience and you can start building products, you know, based on you know the world, you see. So I think we're we're just like at the very start of an application boom, and so that's. That's.

The other thing I'm really interested in and spending more time on is one the barriers to entry to create a new product is much lower now because it's much easier to build products today. So, lovable you can, you can build a prototype and with just a prompting and say, hey, I have this like vision of this idea of an app. You know, create a prototype for me and it'll do it in a very stylistically beautiful way. And cursor you, you can start, you know, interacting with different code and not have to like code yourself and be able to build products. So you really need to focus on making sure that you can execute on these ideas beyond, just like you know what the prototype's going to look like, because now that tool's in so many other people's hands to be able to create.

So, sally, you're on the cutting edge of all things. Ai how do you use AI? You've said you've used perplexity, but what else do you use on a day-to-day? Or how do you use AI? You've said you've used perplexity, but what else do you use on a day-to-day? Or how do you think about tools to use to make your work more efficient?

Yes, tell me so that I can be like you, sally. I need to find more hours in my day.

Honestly, for me it's really these searchable tools. So perplexity, obviously, um, top of top of my list. I use chat GPT for a lot of different things, including like contracting, so, uh. So, for example, we talked about, you know, our podcast when we were working with a sponsor um, we could have used one of our legal partners, but I just drafted a, a sponsorship agreement on chat GPT and it provided me a format and then we worked off of that. So that's already taking away like two to three hours of like legal work.

That was, you know, built to somebody else that we could do on our own. So those things are really been like a great tool for me as an individual. Um, I think I've done that with like invoicing and things like that. So, um, and it's only going to get better, uh. So I think some of those like menial work that takes, you know, a few hours of like man time, I've been able to do on my own so that I can like quickly, you know, flip the comments rather than like having to wait for someone else to do that. So I love that Um, I think that's a really fun um. It was a very efficient tool for me to use Um and then I I like to play around with, like various you know, image and video products, just because I again, it's the the technology is evolving so quickly and it's just fascinating to think about something that's in your mind, take a snapshot of it and be able to create that on these different platforms.

And so I know a lot of my friends use it for blogging rather than getting an image from Get Images or pulling something from Shutterstock. They'll create something that's a little bit more custom. And you know, some of the consumer products are also kind of interesting where you have new generations of like social media companies where it's a lot more customized. So you know, there was YC Demo Day last week and then there's you know I'm going to another, you know Demo Day tomorrow and a lot of people are experimenting with that so that you can generate, you know, a group photo with the three of us even though we're in, you know, two different cities and you know right now it's still a little bit uncanny valley aesthetic, but that is also getting much better and I think by the end of the year we're going to see pretty good technology where it's going to be almost impossible to like tell the difference between a photo and an AI generated image.

To like, tell the difference between a photo and an AI generated image? Is there an AI tool you wish existed, that you wish someone out there? There's some, some entrepreneurial coder, or not even a coder anymore, but some entrepreneur would create so that you can invest in it. I'll share mine, which is I had. I had tried to use some of these AI calendar apps to organize my life. There's all this sort of mom AI out there and it sort of halfway did the job but didn't quite do the job. But if someone could manage my calendar for me, I would be so incredibly happy.

Yeah, I would say like similar category only because, like selfishly, I'm on the road a lot and you miss a lot of connections, and so how do I organize my calendar so that I know my friend's calendar, if we're both in New York, that we'll be able to connect and hang out. I don't know if that really necessarily needs AI tool to do it, but certainly organizing some of those calendars for sure. I know of Williams from, who was one of the co-founders of Twitter, is trying to do this with his app, Mosey. Um, it is not an AI tool. Maybe they use some some AI stack, but, um, where you can, you know, connect with different people? I will say also, like selfishly, like there's um, because we both come from um media and journalism background, journalism background, you know there's.

I'm always interested in like new ways of storytelling and content when it pertains to news and you know, going back to, how do you connect the like source of information and is that a way to like rebuild the building blocks of trust when there's so much content? You know, if there is, you know, an influencer or a content creator that wants to create content in this world, is there a way that they can like connect to like. Hey, here's the actual source of like, where this information is coming from, and maybe this is more of like a post, but you know, I think I, I think Twitter what they've done. I'm sorry, I keep calling it Twitter. I know it's X, but I don't know, I'll just Twitter. They'll call it Twitter.

Like, get rid of it.

Um, their community notes, I think is a really kind of a fascinating tool in a fast speed, fast moving news environment. It's obviously not perfect, it takes a long time for it to get there, but the fact checking part of it is bringing some semblance of trust back into a platform that has a lot of eyeballs, and I know Facebook has adopted some of that as well, or they're releasing their form of that. So I think just how quickly you can go back to being able to confirm where the news is coming from If there's a tool that could do that. You know power to them.

Well, we are all looking to restore trust in media. I'm so excited to see what you end up doing at Comcast Ventures, excited to see what new technologies and new developments that you can bring to the table. Sally, thank you so much for joining us. We really appreciated the time with you and the conversation today. To our listeners, remember that you can like and subscribe wherever you listen to your podcasts, and that you can follow us on Twitter at PentaGRP, and on LinkedIn at PentaGroup. I'm your host, Ylan, and, as always, thanks for listening to what's at Stake.