Nikita Roy: Why AI Literacy in Newsrooms Matters

[00:00:00] This is AI Inside episode 22 for Wednesday, June 19th, 2024. Why AI literacy in newsrooms matters. This episode of AI Inside is made possible by our wonderful patrons at patreon.com slash AI Inside show. If you like what you hear, head on over and support us directly.

[00:00:20] And thank you for making independent podcasting possible. Hello, everybody, and welcome to another episode of AI Inside show where we take a look at the AI hiding inside all sorts of things, including the newsroom, which is the topic for this week's episode.

[00:00:40] I'm one of your hosts, Jason Howell, still in Italy, actually, but by the magic of pre recorded podcast. I'm here with you also at the same time. Joined by Jeff Jarvis.

[00:00:51] How are you doing, Jeff? I hope you think of me as you have a beautiful plate of Cacio e Pepe. I think it's my duty to get some Cacio e Pepe while I'm there. Absolutely.

[00:01:01] I expect photo evidence of this. All right. Well, I don't want to let you down. So I guess I'm going to have to do that. I don't make who knows at this point. I may have eaten it already, but and every pizza and every pizza and pasta.

[00:01:15] And oh, my goodness, can eat so well in Italy. Cannot wait. But as related to this show, just real quick before we get to today's guest and start talking about today's topic,

[00:01:28] which is a I in the newsroom want to give a huge thank you to the folks behind the scenes making this show possible. That is our patron, Patreon at Patreon dot com slash a inside show.

[00:01:39] Joshua Mayer is one of our newest patrons. Joshua, thank you so much for your support. Thank you for deciding like, hey, you know what? I like what they're doing. I'm going to throw them a few bucks a month and support what they're doing.

[00:01:51] And we would love to see more of you because it enables this show. Patreon dot com slash a I inside show. All right. With that, let's talk about a topic that I wouldn't say that we haven't discussed this.

[00:02:06] I would say actually our earliest episodes when we first launched, we had a lot of conversation around the impacts of artificial intelligence on journalism, on the newsroom. And actually, Jeff, you alerted me to someone who you know pretty closely,

[00:02:22] Nikita Roy, who is today's guest, also happens to have a podcast called Newsroom Robots that focuses almost exclusively on how artificial intelligence is impacting journalism. And Nikita, it is great to have you here today. Thank you for being with us.

[00:02:39] It's great to be here, Jason. I'm a huge fan of the podcast. So it's an honor to be here. Thank you. And we and we yours. I've had the honor of being on the key to this podcast.

[00:02:48] Nikita is brilliant. She is a real leader in journalism and A.I. How to use it, how not to use it. But why don't you tell us about all the things you do around newsroom robots first?

[00:02:59] Yeah, please. Yeah. So Newsroom Robots, as you said, is a podcast about A.I. and journalism. We started it last year in April. So we are just crossed our first year anniversary with a grant from the Harvard Innovation Labs.

[00:03:13] And in addition to that, we also now just launched Harvard, our A.I. Academy, the Newsroom Robots Academy. And so this is just a place for people to just get a quick intro into Chad's GBT.

[00:03:27] I launched it with another brilliant colleague of mine, Jeremy Kaplan, who writes the Wonder Tools Newsletter, which is another great A.I. tools newsletter. And so we are doing a lot of courses and focusing a lot on A.I.

[00:03:41] literacy. So this year, a lot of my work has been focused on A.I. literacy as a Knight Fellow at the International Center for Journalists and also leading the A.I. journalism lab at the City University of New York.

[00:03:55] So you've been tracking A.I.'s entry into newsrooms now for more than a year. And I guess I'll start with the blunt question. Who's doing well and who's not? Who's doing well? I have really kept my eyes on Norway.

[00:04:13] And I think, Jeff, you know it. I think the Nordics are ahead of us. Shipstead, a huge fan of Shipstead. We had VG Media Guard was over here at the Common Crawl Foundation event that you put together, Jeff.

[00:04:31] And I had the honor of speaking to him for a really long time. And I was taken aback by the fact that they have already built 30 generative A.I. tools to boost productivity and efficiency in their newsroom. There's this other really small newsroom called iTrumpso.

[00:04:47] It's part of Polaris Media, which is a bigger media organization in Norway as well. And they just got honorable mention for one of INMA's awards over there because they built out this data journalism interface basically that every day is going and scraping all of the different

[00:05:07] government documents from all of the municipals and then creating summaries and then automatically ranking what might be newsworthy and alerting their journalists. And this is something that was probably not possible before gen A.I. and everything.

[00:05:22] But now they have really over the past year, they brought together all of their entire team and they're a small newsroom. They're about like 25 people in that newsroom. So it's not something major, but they again, it was partnerships.

[00:05:34] They partnered up with IBM and put together and built out this tool that now I was speaking to the product manager and he was talking about how they actually broke record sales for subscribers and hit their subscriber

[00:05:49] goals for last year. And they think a lot of that is because it's not just they are more productive and effective, efficient with your work, but they're actually producing quality journalism because

[00:05:59] they were able to produce content that they could not cover before just because they didn't have the resource for it. What kind of documents do you know? Yes, so they're all like multiple documents that's every day each and every county is like putting

[00:06:15] out records and like meeting notes and like all of those different information about the particular county. And so they scrape everything and then summarize it for people to then for their journalists to then go and investigate further. So everything from like infrastructure, information about it

[00:06:34] to just like meeting notes that I think is something that their open data is there everywhere. And we just need to be able to crawl it, scrape things and be able to put it in a really well

[00:06:50] like a documented database and be able to extrapolate and work on top of that. Because it's not like this information didn't exist out there already. This just really seems like one of those situations, those really great examples of working smarter, not harder. Right? Because I mean if that

[00:07:12] information is out there and the journalists are doing what they do, they're going to find those information sources, they're going to synthesize, they're going to do what they do in reporting

[00:07:21] and collecting and all that kind of stuff. But if you can use an AI system to do a little bit of that legwork, what does that free you up to do in your job? Yeah, it's just time and time again,

[00:07:36] examples like this just kind of blow my mind when I see kind of the efficiencies that AI can bring to an organization. When we get out of the realm of, oh, AI is bad, I don't trust it or whatever,

[00:07:49] maybe it's not good for that, but it's really good for this. It seems like it's really good, at least in this example of this is the information source, this is specifically what I'm looking for

[00:07:59] and when you find it, let me know. And that frees me up from having to do that long form or longhand. In Norway, we'll stay on Norway for a minute, I think this is all trying to have an excuse to go do the show

[00:08:12] in Norway, Jason. I don't know if you've heard of a company called Inacode, Nikita, that its model is to enable journalists to help towns, municipalities share their information more readily and in more useful ways. The example that they give is that if somebody files for a variance

[00:08:37] to build an addition on their house and nobody responds to the variance and the hearings, but then when construction starts, the neighbors start screaming, how did I not know about this?

[00:08:48] Well, if there's a new business model available where we in news could help the town share their own information more effectively for their benefit, citizens' benefit and thus ours. And I think that there's ways to open up using AI, those kinds of things more efficiently.

[00:09:05] I love that idea, Jeff, and I think there's so many different tools as well that are out there and just New York City alone, we have like 3000 open data sets that's there. And then there's just so much more information that I think as journalists,

[00:09:22] and I think that's where our role is going to be, is making sense of all of that data. And we now have the tools to do that. In addition to the productivity and efficiency,

[00:09:32] I find it as another superpower for journalists who probably did not have access to be able to make sense of such huge amounts of data because you needed to have the technical ability to code and do all of that.

[00:09:46] But now with like Chad Gbd's code interpreter and then this other really cool tool that I'm a huge fan of the team called Wabi AI, and they are again working with a lot of the European newsrooms.

[00:10:00] They're a European startup. And what they do is connect to all of the open data sets in especially right now. They're focused on Europe in Belgium. And as soon as any of these open data sets get updated, they automatically start generating AI insights, for example.

[00:10:16] And as a journalist, this is something that if you ever have to deal with these open data sets, they get updated regularly. And then by the time you've done your analysis, you have to go back again and get the updated data set, then do your analysis again.

[00:10:28] And this but this is just something that speeds up like the entire process. And I was talking to newsrooms who were actually working and using them, and they were talking about something that it would take them like a couple of days to work with the data journalist.

[00:10:41] Now a journalist is able to do that like within minutes. And I think that's that's the superpower that now journalists are going to have to be able to make sense

[00:10:51] of huge amounts of data. And I'm just excited by the potential of the number of stories that we'll be able to uncover because it was all hidden behind these kinds of like extensive sources and data that not everybody had the ability to go and analyze.

[00:11:09] Arnab, I'll let you off the hook. Who's doing it badly? Or to ask it in a more politic way, what should American and Canadian newsrooms be doing that they're not?

[00:11:22] AI literacy. And I am actually just really shocked when I actually see I interview a lot of the European newsrooms right now. I'm talking to a lot of them and everybody has an AI literacy plan in place and it's for everybody in the newsroom.

[00:11:39] A lot of them actually started it like back in before Chagy Pity even came out. They are journalists were learning everything about what deep learning was, for example,

[00:11:49] and what what what's the basics of machine learning just so that they have the lingo because they know that their newsroom, they have their product teams working on it and they wanted everyone to be comfortable with the lingo.

[00:12:00] And then Chagy Pity came and then they have boosted that a literacy and everybody has some sort of like an AI literacy plan where it's not just for how journalists can use Chagy Pity, but like getting them to experiment with these tools

[00:12:15] and then building shared institutional knowledge. And so that's one place I think we are still lacking as an industry when I see it in North America. It's still just in the hands, I would say, of product teams who are experimenting with it.

[00:12:30] I go into a lot of these newsrooms and when I do workshops between 20 to 40 percent of the room has probably never even used a tool like Chagy Pity.

[00:12:41] And that shocks me. That number should be zero. If you're a journalist and if you're in a newsroom, you should have at least used a tool like Chagy Pity. Just out of curiosity.

[00:12:51] Yeah, just out of curiosity. So we are just like reading about tools like Chagy Pity and not really experimenting with it. And most people would have tried it out when it came out in November 2022.

[00:13:03] So they have not tried out GPT-4 and now GPT-4-0, which is out there.

[00:13:08] So I am like right now my big thing is if you have been tried Chagy Pity since a year ago, go and try it out right now because the free version is giving you access to all of the premium capabilities of Chagy Pity that have been available for the past year.

[00:13:23] So like the data analysis, you can try out a custom GPT, for example. You can try out the latest model GPT-4-0 and you can see how different it is and how advanced that model is.

[00:13:36] And I think AI literacy is so important because there's a right way to use a tool and there's a wrong way to use a tool.

[00:13:43] And we will not know what the right and wrong way is until we ourselves are educating our teams about it and then figuring out how to use because every journalist, every team member, not just if you're in the editorial department, if you're in the ads department, wherever you sit in the newsroom, you have your own set of expertise and experience.

[00:14:03] And when you combine that with a tool, a generative AI tool, you'll really be able to unlock a lot more use cases than just one person who's figuring it out.

[00:14:13] Like there's only a certain number of use cases that I as a person can know because I have a limited set of experience and expertise that is probably completely different to a photojournalist because I'm not that person. And so they would have maybe more understanding, more use cases.

[00:14:28] And so I really think that we need to be focusing on how do we build institutional knowledge around AI because it's a time about upskilling an entire workforce.

[00:14:38] Does does that apprehension that you're talking about that that sounds at least the way the way you just explained it to be pretty common here in North America when we're comparing, you know, kind of the approach or the mentality around AI tools to places outside of the US.

[00:14:54] But does that apprehension come from the journalists themselves or the people who would be working with them? Or is it a more institutionalized kind of apprehension that's passed down to them?

[00:15:07] And, you know, they might feel like, well, you know, it's this is discouraged or this is actually actively discouraged by the organization. That's why I'm not going to touch it. And why is that different here versus the other places that you're talking about?

[00:15:25] One of the main things I've noticed over here has been a lot of people comparing it to the pivot to video. Oh, OK. That's interesting. It's just another hype cycle. And everybody is going to invest in it.

[00:15:41] And just because it's the same social media companies, I would say, like a meta and Google and Microsoft, who are also now inventing this new technology. And at the forefront of that, people are hesitant about that past relationship and how they got burned by it.

[00:15:59] Then there's so many other complex issues about the fact that they have been scraped and people's hard work has been scraped and was used to train this model.

[00:16:09] So not most people in the industry don't see that as a fair way to be using it and being not being compensated for that.

[00:16:19] And then there's just a lot of those copyright concerns, which people tend to have the idea that if they are using it, it's just it's not ethical. And they are concerned about it.

[00:16:32] And privacy remains one of the biggest issues because people just don't know if they are using these models, how it's going to be trained. A lot of them, like I even talk about how it's actually for teens is there.

[00:16:44] There are this Microsoft Azure where you can have enterprise level access and that's private and secure. But sometimes people are just too hesitant. And I was like, like, is it really? And not having not being able to trust the tech companies.

[00:16:58] And I think it's because of the past relationship that media has had with tech and being burned by it or feeling like they were like that's just going to be they're going to make themselves reliant on the stick. But I think it's a completely different thing.

[00:17:12] This is not social media. This is a transformation and it's impacting every single industry. And this is a foundational technology. This is like your emails are on Google. It's the same thing as having Google Docs. You are reliant on Google as a company there.

[00:17:29] But now, in addition to Google Docs and Gmail, you also are adding now Gemini as a product of that. So they're going to be reliant on. That's interesting. Yeah. You have a stand on whether using news content to train models is fair use and transformative.

[00:17:48] I'll put rag and I'll put separately, but on the on the on the New York Times suit again, it's a contrast in views. It's litigation in the U.S. and its collaboration in especially the Nordics where Shibstad is making their own LLM. And everybody's cooperating.

[00:18:06] Everybody's putting it in to figure it out. The New York Times and Alden, the hedge funder suing OpenAI. Where do you come out on that question of fair use and the training sets? Yeah, it was interesting.

[00:18:21] I was I was at a panel with media lawyers just last week at the New York State Bar Association. And it's very interesting to hear the perspective from media lawyers on this particular issue.

[00:18:35] And I've spoken to a lot of tech lawyers and just how different both sides of the argument are. But I have to say, as a as a technologist and somebody who is a data scientist, I see a distinction between two things that are happening with this technology.

[00:18:53] And the first one is a foundational large language model, which is like if you go straight away to like GPT 3.5 and GPT 4, not using a tool like Chagy BT, but like using the core model, what they have been trained to do is basically predicting the next word.

[00:19:13] And if you talk about the New York Times lawsuit at the heart of that lawsuit is it's pointing towards an error that large language models tend to do in addition to hallucination, which is one of the risks of making up information.

[00:19:28] There's also the other big risk of memorization where they accidentally tend to memorize certain pieces from their training data because they might be very unique or repeated a lot in their training data.

[00:19:42] So that's one of the main issues why sometimes you say don't input private information into a large language model because they tend to sometimes remember addresses. If you put your Social Security number, they could remember something like that and spit that out, for example.

[00:19:54] So they tend to memorize patterns and unique information. And when you talk about the New York Times lawsuit, a lot of the exhibits over there that they were talking about where Chagy BT,

[00:20:06] where Chagy BT was able to specifically replicate information, those were extremely unique Pulitzer Prize winning award articles. And so that's a fundamental flaw of the model that researchers are actively trying to find a fix for the same way for hallucination.

[00:20:23] We're not sure if they will find. So and I see it as transformational. So that's one that's definitely one you. So I get the point of how it can be.

[00:20:38] It can really impact our industry, but there's a lot more to that technology and it's not meant to replace news.

[00:20:47] But at the same time, I've gone around saying that I feel like tech companies are becoming news publishers because the other side to it is what can you do with this foundational technology?

[00:20:57] You can build something like Microsoft Bing chat where or like Google's Gemini, where you are accessing the Internet, scraping a particular news website and then providing a summary about what the news is.

[00:21:09] Meta is doing this. If you go into any of their apps and you go onto like WhatsApp and you ask a certain question, they are going and scraping a particular news website and then summarizing that and producing information to that perplexity is doing that.

[00:21:24] I use it a lot, but at the heart of it, what I'm seeing is it is going and scraping all of these news. I don't have any incentive and I don't usually go and double check everything.

[00:21:35] Only if I'm fact checking myself or the information over there will I go and look at the particular news website. So over there, there's already traffic decline that's going to happen.

[00:21:46] And I feel like because the people who actually wrote that content and now you're specifically going and summarizing that information, that is a direct link of what you're doing and you're replacing and you're not paying those people for their hard work.

[00:22:00] And I definitely see the deals that chat GPT is currently doing and signing is at least a good step forward because there needs to be licensing. And for publishers, if you are going to directly use their content to produce a product on top of that.

[00:22:15] But that's not what a foundational large language model does. And also to the other side, if we put a lot of legal and legislation around a foundational large language models, it's actually going to hurt the small companies.

[00:22:32] It's not going to hurt the large tech companies and we will be killing the open source community. I agree there. All right, we do have more coming up, but we do have to take a really quick break.

[00:22:41] Yeah, it's so the deals that are being made right now, especially with open AI, the Atlantic, Vox, FT News Corp, actual Springer. What I've argued is that those aren't actually content licensing deals at all.

[00:22:57] They're PR and lobbying deals. They're trying to say to those companies, you're big, you're powerful. If we pay you money, will you just shut up and not bother us? And the vast majority of valuable news is left out of that.

[00:23:10] And we've talked about this before. I've argued that the news industry should come together and create a news API. The chances of our industry cooperating are practically nil because they never have.

[00:23:21] But Nikita, I've gotten kind of a nightmare scenario in the last few weeks where I don't know that chat interfaces and agent interfaces after them will take over the world. But let's just say they do.

[00:23:32] And if they do, I think we see the death of the web as a destination. Just what you just said. You can get an answer and you don't go there. So what has to happen, I think, is that we need new structures of discoverability and citation

[00:23:49] and new business deals. But those have to be, I think, very broad deals that almost replicate the web rather than these small deals with the big guys on top of the industry. And I wonder whether, in a way, those deals are kind of ruining it for everybody.

[00:24:07] Everybody else, I should say. Yeah, that's an interesting point that you're bringing up. And I think I looked at Nick Thompson's when he announced the deal and he posted a video on LinkedIn about it.

[00:24:25] And I think he gave a really good insight into why they're dealing with doing these deals. And also was talking to Gabriel Brotman over at Axel Springer about why is Axel Springer doing these deals with OpenAI.

[00:24:41] And from that, it seems like they see tech companies as right now being the innovators of this technology and wanting to be ahead with them and not being against the innovation that's happening over there and supporting there.

[00:24:58] So but what we are seeing is only the giant tech companies who are going to have the bargaining power and be able to basically strong arm deals.

[00:25:10] And I don't see local news companies, what local news companies are going to be able to have that same bargaining power with a tech giant. And we are fundamentally going to miss out on that news ecosystem.

[00:25:28] And I agree with you that I don't think chat is the interface of the future,

[00:25:33] because when you go onto something like chat GPT and the reason why you have people like me who are going and helping people understand the tool is because you go into it and it's just a blank piece of screen for you to type something in it.

[00:25:49] And you don't know what you want to ask it. So what's going to instead happen? And I think what the news industry really needs to focus on right now is building experiences with this particular technology.

[00:26:01] So, for example, you have all of these like tech review sites and fashion articles. What kind of experiences can you create for your audience? So, for example, what kind of like what tools could you add on to?

[00:26:19] Like I am buying something over here right now and it automatically decides instead of like affiliate marketing where you're writing an article, it's automatically giving you all of these different suggestions of places and like your car, your table and your desk.

[00:26:37] Like if you want a standing desk, what type of options are there? And it's creating like an entire experience for you. And it's not an article. And that's what I keep on talking about.

[00:26:47] Maybe like more visual and it's something about I think as an industry we need to just like come together and just think about what that would look like.

[00:26:55] If you're having like a bunch of fashion articles and you can start suggesting to people what kind of outfits that they should be wearing that's personalized to them. And again, it's not an article, but people can come and decide what they want to wear for their Christmas gala.

[00:27:13] And it's all of that. We're seeing a little bit of innovation over there, I would say. But again, it's in that chat interface. But I'm thinking how can we go beyond that? Because as news, we have so much of videos, images, audio, multimodal AI. It's huge right now.

[00:27:29] How can we create conversational experiences with not just the news, but there's so much more than that. And so what happens when you start embedding and creating those experiences with other forms of distribution and not just text? Yeah, that's interesting.

[00:27:49] I mean, that really just harkens back to our episode two with Shipstead. Svendr Mertalo. There we go. Thank you. By the way, is leaving the new Shipstead today as we record this is split up into two companies.

[00:28:09] And the marketplaces that kind of classifieds and commerce that subsidized all this great innovation they do is now its own company. And the news company is owned by the Tiniest Trust, which is like The Guardian, a foundation over it. And Svendr is going over to the commercial side.

[00:28:26] But there's a lot of great leaders still on the news side and a lot of great heritage from the people like him. Right. And the kind of that legacy. Yeah, Svend talked a lot about exactly what you're talking about.

[00:28:37] Not just being a news site that presents the text, the information, but figuring out how you use this new technology to almost it almost makes it more of like an interactive kind of experience as a user or as a reader.

[00:28:54] I no longer go solely because there's an article there that I want to read. But how can I tap into the collective knowledge that that particular organization has curated and crafted over time and collected into this gigantic database that we can tap into to get much more value?

[00:29:14] Does that approach encourage the continued creation of content the way it is or does that encourage a new type of creation? Does that make sense? A new way in which we will create content, I think so.

[00:29:33] And we'll have to because I think Gen.E.I. fundamentally is a user experience revolution. That's what it is. It's changed the way you can talk to data. It's changed the way you can talk to tons of information and you can the chat experience has evolved.

[00:29:48] You can talk to it. It can understand your environment, as we have seen with the latest demos from both Google and JVD for all.

[00:29:57] It's it's the ability to in real time process information and be able to give you feedback at the end and talk to you about that particular information from video, from audio images and text.

[00:30:11] And I think that means that we need to evolve not just as a news on a website. And I think we haven't evolved from the print era. We just put a newspaper online on the web.

[00:30:25] If you go to any of the news sites, that's what it looks like. And how can we now change that, especially when we know the younger audience are addicted to the way the videos are the short form videos? They really like that. They like the text box rolling.

[00:30:39] They like YouTube shots. So how can we bring in all of those different forms of video content and at the same time make things more personalized? As you're saying, there's so much of information that news companies are putting out over there.

[00:30:53] How can we now start personalizing on a deeper level to every single individual?

[00:30:59] And how can we now not just have it as a place where you are going and getting your information and that's like it's more of you receiving it, but it's like you're interacting with that information.

[00:31:13] So if you, for example, are not you're unhappy with the particular particular news story that's out there, people people tend to have emotions around that. Maybe you can have a conversation around it with your news site, for example.

[00:31:29] And then you can go deeper and then it knows information about where you're living and that you're traveling a lot. And for me, if it's like if I'm like going, I'm currently in Florida and I'm coming over here, it's like automatically it knows.

[00:31:44] And if I trust my news company and I'm OK with like sharing data about what I'm doing or where I'm going, it can start to personalize experiences with me and for me.

[00:31:54] And so that's where I think we need to start looking into how can we already building trust with audience? How do you build trust so that we can now use that to help personalize experiences for people?

[00:32:04] If I get a better experience with my news app because they are now going to meet my need by safely using information about how and where I'm going or like what type of things I'm doing, what types of things I'm interested in when my garbage pick up schedule is, for example.

[00:32:21] And all of those like information that you need, it just gives you an interactivity to be able to build on top of that.

[00:32:27] And so I think there are like new experiences that we need to be thinking about that if I'm walking down the street over here and I've never been in Tampa before, I can just open up my news app and like scan the buildings that are around me.

[00:32:40] And it tells me because all of the different things that must have happened in that particular area, there's so much of history at every single intersection. Maybe that news that has been done in the previous years that the local news company over here has been reporting on.

[00:32:56] And I can at a glance walking through, get a history of all of the different things that have been happening down the road.

[00:33:03] And or maybe there was like last week something in something monumental happened at this intersection and I never knew about it because I'm just here for the first time.

[00:33:12] So like all of those different, different ways in which we can start to create experiences because that's what's going to build subscribers as well.

[00:33:21] Because in a time where a lot of different people are going to be able to create a news website, as we've been seeing, people anybody can actually spin up a news website and call themselves news if they're just going and scraping from somewhere else and rewriting that.

[00:33:39] And then what is differentiating you as a company? So I think that's where I'm really, really excited about the potential because it's going to give us a moment to really change the way news is delivered and consumed.

[00:33:52] I think it's a great question, Jason. I think we're so used to the story form. We decide what the story is. We write, we are the storytellers, we're the narrators, we decide what goes in it.

[00:34:02] We could end up creating a lot more experiences, as Nikita said. I think we can create more databases of information. We can create more dialogue among members of the public, dialogue with documents and with meetings and so on.

[00:34:15] I think there's a whole raft of ways we can look at news and what communities need differently. However, recognizing that the tools will screw up.

[00:34:25] Just today I was going over, this is going to age me terribly. I was going over Medicare forms with my wife and I needed to find out how many numerals or letters there are in a Medicare number.

[00:34:36] Okay, simple question. I ask, first thing I see at Google is 12. And then I look below, that was the AI answer from Google. The actual answer was 11.

[00:34:46] We know these things are going to screw up. We know they should be kept away from the news and creating news on their own. They can be very valuable for lots of things, but perilous in other ways.

[00:34:55] Nikita, I'm wondering whether, because you talk to different companies most every week and you see a lot of interesting stuff going on. Do you have some companies and their work that may be of interest to our audience?

[00:35:09] I really find it very interesting the way in which people are building out tools right now and the approach with which people are thinking through AI product management.

[00:35:19] And I think one company, which would be very interesting to look at is Epimedia over in Germany. He's just left Alessandro Albiani, but they had, he was the AI product manager over there.

[00:35:31] And I really liked the way in which they were experimenting with AI and bridging the product team and the editorial team together.

[00:35:37] So they would do like two week internships where there's somebody from the editorial team would be like pitching products to the product team, and they would bring them over for two weeks and they would be completely embedded in the product team.

[00:35:48] They would not have any of their responsibilities day to day that they were assigned.

[00:35:52] And then they would be seeing how the product is being built. And within two weeks, they would try to build a prototype of that particular product that the particular person from the editorial side of the newsroom had as an idea.

[00:36:02] So they have actually built out multiple different products, one of which was again on the app. They have a version where you can either read the entire story or click a button and just get a summary of that of the new stories and then go from there.

[00:36:15] So I think like how are we changing the way in which news is being consumed and how can we get to those experiences is by getting editorial and product folks together to start thinking through because the expertise is in the editorial side of the newsroom of like the experience of like how to create great news stories and how to find that and what the audience needs.

[00:36:40] And the product team is bringing in all of these technical experiences, expertise and trying to see how to bridge the gap with what the newsroom, what could be of better value to the audience.

[00:36:51] So they were definitely one of the people that I think have been very, very interesting to look at. How about some interesting startups? Wabi AI for sure has been a very interesting startup to look at.

[00:37:07] As I had spoken in the beginning about the data journalism analysis that they do with automatically with AI and open source, by far one of the biggest startups.

[00:37:19] Noda is the other interesting startup that I've been looking at because I find it very, I keep on talking about we need to go from a lot more videos and doing short form videos and that's something that Noda with your tool allows you to do.

[00:37:33] Yeah, you can create videos from it. You can create a newsletter or social everything from just a particular AI like from a text and AI converts that into a video that people can then go and edit.

[00:37:45] And you can help with your newsletters and your social media posts. So I find them really, really interesting.

[00:37:52] And then for people who just want to get a start into AI, one of the main things that almost most of the newsroom that I've spoken to have something for headline and SEO.

[00:38:01] So if you're a big newsroom, you have either built it in house or if you're not a big newsroom and you are maybe a small local newsroom.

[00:38:08] A lot of newsrooms are using this tool called Yesio. It's a free Slack based AI tool, and it was developed by a fellow from the Reynolds Journalism Institute.

[00:38:19] And so it's free and over 400 local newsrooms have downloaded in their Slack workspace and they use it to generate SEO and headline summary.

[00:38:27] So I think that's also another another great tool and app for people to at least get started with the AI if they're nervous about and scared about it.

[00:38:36] This is one thing that's being used a lot. So I would say these are definitely some of the tools that I've been looking at looking at. And there's also another very interesting startup that I've recently been getting getting into a bit more.

[00:38:55] It's called Satchel AI. And what they do is they they plug into all of the city council meetings, for example, and then automatically take the transcript, generate that into a news story that you can then go and edit and go from there and be able to publish on your site.

[00:39:13] And so they take all that process. And I was talking about this so much of data out there. So how can we take like all of the city council meetings and automatically start to put that out to people and get people to get that as news articles in front of people?

[00:39:30] And so I think this is also another interesting one that I have been I've been interested in looking at.

[00:39:37] One thing that comes to mind for me is we've been talking about all the ways in which kind of the the journalistic organizations can change is that that's really just kind of half of the half of the challenge is, you know, how the organization integrates or figures out these really innovative

[00:39:57] ways to use AI, how they get over the fear factor of, oh, well, we're not supposed to use that for this that or whatever reason. The other half of the challenge is what the people who are looking for news, what their habits are, what they feel comfortable with.

[00:40:12] And I guess the thing that came to mind in the last few minutes as we've been talking about this is the approach for integrating an organization's data set into a new kind of way to present this and make it a little more interactive.

[00:40:29] Like that will work for some. But for other people, it's gonna be like, no, I don't want things to change. I want things to say the same. And that seems and actually that's built upon as you would know, especially Jeff, that's built upon many, many, many, many decades of, you know, kind of reinforced, you know, tradition and standards and all that kind of stuff.

[00:40:52] How how can this kind of evolution continue to take place when people are so dang ingrained? You know, it's they're so kind of used to doing it the way it has been and getting their news the way they have been.

[00:41:09] Yeah, I think it's a I think wake up call for the entire industry because as we're seeing newsrooms are shrinking. We are a lot of newsrooms are feeling threatened by the large language models and generative AI search taking over.

[00:41:25] And I think it's a matter of time for us to just understand that this technology is transformational. And it's not just affecting the news industry, it's affecting every single industry we're reporting on it. We know how it's affecting. And so if we don't step up, it's going to it's going to not be great for our business.

[00:41:46] And if you're running if a newsroom leader is running a particular like a newsroom, it is I would say it is an existential threat to a newsroom. I don't think it's an existential threat to news. It's an existential threat to newsrooms that do not in a way right now.

[00:42:01] And before we even get into AI, you're talking about data and I think it's very important to highlight this all of this functions. If you have a good database to start from your data is the fuel of any AI model of any AI powered experience that I'm talking about.

[00:42:19] And it's a whole different form of data. So a lot of the newsrooms that I'm talking to right now, like you have they're thinking about how to effectively manage and store news because in order to power all of these AI database, AI experiences, you need to create something called a vector database, which is a different form of storing information than how it is stored right now.

[00:42:42] And being able to build them and create them, it is expensive. It's expensive to host. And how can we efficiently store so much of news data in these sort of new databases, retrieve information from it and build experiences with it? There's a lot of costs involved in that.

[00:43:02] There's a lot of experimentation currently going on in terms of what's the right way to use these databases. So it's not just a new AI. The engineers on your team are working with a completely new set of way in which to store data as well in order to build these sort of like information ecosystems and AI powered information experiences.

[00:43:24] So I think we have to that's where also right now a lot of the innovation needs to be and a lot of the focus needs to be on building and getting top tech talent into the news industry, which is tough because we are competing with a lot of the other the tech tech platforms.

[00:43:44] But I know a lot of people want to work in news as well, and I think we need to appropriately compensate and get them into the news industry because that's how we will be able to create great experiences with AI.

[00:43:59] But are we at the point, Jensen Wong has said that universities should stop training computer scientists and someone in one of our early shows said that the great line that English is the most powerful programming language on Earth today.

[00:44:14] We hope more languages soon. A point is human languages. Do we need tech talent or do we need are these tools getting so easy to use that what we need is what you're doing, which is training more of the newsroom and how to build the things that we need to use.

[00:44:28] So I think there are two layers to that, and I think you need foundational tech talent. Like when I'm talking about tech talent, I'm talking about getting AI engineers and do people who are doing really cool AI research into news and getting them to help with the news industry in order to progress AI.

[00:44:54] We need that tech talent. Computer science, I think is evolving right and I think Sal Khan had come on your show earlier and he was talking about this as well. Like computer science is sort of like like was a new major that had come out.

[00:45:09] And so it's something that changes a lot. And I did a degree in data science and most of what I learned is honestly outdated in a way like all of the tools and technology that I learned is outdated because new things have come out and we are learning as it goes in the entire industry.

[00:45:24] The entire tech industry has to constantly learn because the technology keeps changing so quickly. So tech folks are usually used to this sort of like rapid change in the industry. What I think is now needed is getting those AI research folks to do the fundamental level of building a new infrastructure for news, a new data infrastructure for news because that's where also the crucial changes.

[00:45:52] It's in order to build AI systems, you need to have a new data infrastructure and that's only possible by people who are doing that sort of research. And we have some people who are already doing great work and we have like people like Nick Diocopoulos or at Northwestern University who with this computational journalism lab over there has some really great students working on stuff.

[00:46:12] But we need a lot more. When you say don't need English as the like universal coding language, now I agree with that even though I am somebody who has studied coding, I actually go to chat GPT a lot of times and get it to write my code.

[00:46:27] I get it to write my code and I get it to write my code in different, different languages that I'm also not so familiar with. But I have the basics of coding and building apps that I can now understand what's happening behind the scenes and then be able to create that.

[00:46:43] So I think you need to have people who have the foundational knowledge. But yes, you can get a lot more people to do stuff that they couldn't do because of the capabilities of generative AI. So you can do a lot of those data analysis tasks that I was talking about.

[00:46:59] But before it ever goes to publishing, you need somebody who knows the code who can go and fact check through that entire thing. So that's you need those you still need that talent. And in order to build the next level experiences, that's going to come also by getting those tech talents to be able to push the way and create those experiences and push boundaries there.

[00:47:21] But I'm really hopeful that I think we might see a lot of people getting into coding as a result because the barrier to coding has lowered tremendously. If you get stuck, the biggest thing with anything in code is that you get stuck the first few times when you're learning. It's really, really challenging.

[00:47:38] But now, because of tools like charge, you can walk you through all of the errors that you're facing and walk you through creating and getting the basic steps of how to code. So I'm really hopeful that we will see a lot more people trying out and getting into code and hopefully a lot more women in engineering.

[00:47:58] Amen, sister. Indeed, indeed. Well, and you're talking about chat GPT a lot. I should also point out folks should definitely check out the latest episode of newsroom robots podcast where you dive into kind of the as you say the potential of custom GPTs in the newsroom very interesting stuff.

[00:48:19] I was listening to that and I thought, you know, that I think we're going to see a lot more of this not just in newsrooms like right as we as we continue to understand kind of the value of I have this data set. How can I put it to work in a way that's very useful?

[00:48:33] We're going to see this not just newsrooms but everywhere else. So newsroom robots.com is the site for Nikita's podcast. And then I think while you're also there, you can also check out the Academy. There's a little link to take you to the newsroom robots podcast.

[00:48:48] Newsroom robots Academy that Nikita mentioned a little bit earlier. Nikita, this has been really fun. Thank you so much for letting us turn the table on you because you are a podcaster. It's always interesting right when we have the opportunity to have the tables turned and we become the guest instead of the host.

[00:49:05] This was this was a lot of fun. Thank you so much for having me. Yeah, anything else you want to throw in there as far as people who want to follow the work you're doing obviously they have the podcast in the Academy. Is there anything else you want them to know?

[00:49:17] Yeah, feel free to connect with me on LinkedIn and on Twitter by Nikita Roy. And so yeah, I'm always on lookout for very great interesting work happening in AI. And so feel free to let me know about that.

[00:49:32] Excellent. All right, folks should do that. And thank you, Nikita. Really a nice pleasure to talk with you today. And yeah, we'll talk to you soon. Best of luck with the podcast. Thank you again.

[00:49:43] All right, and that is it. Jeff, we have reached the end of this episode Gutenberg parenthesis dot com I think for for you, sir. Yep, that's work that does the trick that does the trick. There we go. So people can take a look at your your handy work magazine and the Gutenberg parenthesis. Thank you so much, sir.

[00:50:03] And yeah, anyone wants to get the new Spanish version it is out. Oh, really wonderful job. Wow. With with illustrations galore in here. Really nicely done from the Spanish publisher and no kidding. That's very catching cover. Beautiful cover. Yeah, beautiful cover. Excellent. Great work.

[00:50:25] As for this show, we normally do the show every Wednesday at 11am Pacific 2pm Eastern on and if you wanted to catch it live normally I should say with an asterisk. You would go to the text floater YouTube channel to youtube.com slash at text floater. But of course, I am in Italy right now. So these are all pre records. I think it's the first episode in July is our first return back to live episodes. So, you know, put that on your mental radar. And if you missed the live version, you can go to the live version of YouTube.com slash at tech sploater.

[00:50:55] And if you missed the live version and you get it in podcast, hey, that's the way the most that most people do because we published the show to the podcast feed later that day. Anyways, be sure to like rate review subscribe wherever you happen to listen. Also, as I've said, support us on Patreon so we can continue doing this work with you.

[00:51:12] So, as a result, I call them out at the end of the episode. Dr. Do Jeffrey Maraccini and WPVM 103.7 in Asheville, North Carolina. Thank you so much for helping us do this show each and every week. Everything you need to know about what we do can be found at AI inside dot show. Thank you much, everybody. We'll see you next time.

[00:51:42] On another episode of AI inside. Take care y'all.