[Chad Davis] 12:01:23 So thanks everyone for joining us today. [Chad Davis] 12:01:31 And this is going to be an exciting one. I know it's billed as content creators and journalists, but when I was talking with Ethan and Ernesto. [Chad Davis] 12:01:40 Before the webinar kicked off, it really struck me about how the approach that they are taking in the KQED is taking as a whole station approach. So I don't want to steal their thunder, but I think you're in for a real treat. By way of housekeeping, I'll just make a quick mention to thank our friends at NIDA who help us market these webinars. Always appreciate it. [Chad Davis] 12:02:06 If you are new to this, you can sign up and kind of be notified about what happens with public media innovators, our newsletter or these webinars and whatnot through public media learns links i think oh amber's good on the links too. Full service there [Chad Davis] 12:02:24 Next month, a little bit of a difference because, of course, we've got an annual meeting for the PBS folks hitting during the fourth week, I think, of the month. [Chad Davis] 12:02:33 So we are doing a different time. It's usually on Thursdays. We're doing it on Tuesdays in May. It's going to be Tuesday, May 13th, and it's going to be at noon Eastern, 9 a.m. Pacific. So registration link will be popping in there very soon. [Chad Davis] 12:02:47 Love it how that happens. And last bit of housekeeping, then I'll hit you with. Those who intend to use the chat, and we hope you do. [Chad Davis] 12:02:58 You will notice if you open it up that it's probably defaulting to host to panelists. That is Zoom being Zoom. And you can just manually set that to everyone. We notice every time we do one of these, people start sending comments and questions and the first round always end up just coming to us on on the host and panelist category. And so set it to everyone and everyone will get to see what you do. [Chad Davis] 12:03:22 So with that, we're going to kick it off here. Welcome to Ethan Toven-Lindsey, and Ernesto Aguilar. Ernesto, I want to just… give a shout out to. He is our first three-peat guest. So I was saying before we started, he's going to become the Steve Martin of our [Chad Davis] 12:03:43 Of our webinar series, we get at least one Ernesto hosting it every year. [Chad Davis] 12:03:49 But I'll let them kind of say what they do at KQED. I will just say that There are a handful of stations that I look to when I'm thinking about like, how could we be upping our own best practices and what does it mean for [Chad Davis] 12:04:02 For us in Nebraska to level up and KQED is one of those stations and you are about To see why. So Ernesto, Ethan, I'll let you guys decide who wants to kick it off and I'll see you back here when we get to Q&A. [Ernesto Aguilar] 12:04:18 Well, Ethan unmutes and queues up his introduction, I just want to say thank you so much, Chad. [Ernesto Aguilar] 12:04:24 And Amber and the entire team, we really appreciate being here and also Big shout out to all my colleagues I see in the conversation and also a lot of public media friends here really are honored by your presence. So Ethan, please kick it off. [Ethan Toven-Lindsey] 12:04:40 Okay. Thank you, Ernesto. And I agree with Ernesto. A lot of what we talk about will probably be informed by a lot of the other folks from KQED and around the country who we've worked on with this project or this just sort of thinking about AI. [Ethan Toven-Lindsey] 12:04:57 I'm Ethan Toven-Lindsey. I'm the VP of News at KQED. [Ethan Toven-Lindsey] 12:05:02 And we've been within the newsroom, and this is where Ernesto asked our playoff one another, we've been thinking about very particularly how newsroom journalists, content creators use or take advantage of AI and different AI applications and then have been interfacing with the rest of the organization about how they're shared [Ethan Toven-Lindsey] 12:05:26 Or differences in terms of those projects. And so you'll hear some of those individual as well as sort of like organizational [Ernesto Aguilar] 12:05:40 Fabulous. I am Executive Director of Radio Programming and content innovation initiatives here at KQED. That is a big title. It covers a lot. I also just recently finished up an AI and journalism leadership cohort at the City University of New York's Craig Newmark School and write a newsletter on LinkedIn called AI and Public Media Futures, which has synthesized a lot of the learnings from that cohort. [Ernesto Aguilar] 12:06:07 But is also servicing out more conversations that are coming out of public media around AI. And we have some really great things to show you. I am very much somebody who believes in scale, as does Ethan. While we do have a lot of KQED and we're able to do a lot Thanks to that, we also are understanding of how organizations have to be very, very nimble [Ernesto Aguilar] 12:06:30 In trying to ensure that we can get this work done. Sometimes as somebody who's leading this up as one person or two people and not only just that technical part of it, but also the culture part of it, which we'll get into in just a moment. [Ethan Toven-Lindsey] 12:06:49 Ernesto, do you want to share the screen or [Ernesto Aguilar] 12:06:52 Either one. I'll do this year. We have… This feels like live radio. We just kind of roll with this. I like it. [Ethan Toven-Lindsey] 12:06:58 Okay. It is exactly. Let's tap dance a little bit. [Ernesto Aguilar] 12:07:02 There we go. [Ethan Toven-Lindsey] 12:07:04 I'll do a little quick kickoff and then and then spin to Ernesto, who's leading a lot of this within KQED connecting. We also on the call have Tim Olson, who I assume many of you know, who is also leading a lot of our efforts. [Ethan Toven-Lindsey] 12:07:19 Within the newsroom. Gosh, it was probably when ChatGPT that weekend when it sort of like burst out onto the scene. I mean, you know, in the Bay Area, we've been tracking it and knew about some of the Gen AI efforts but [Ethan Toven-Lindsey] 12:07:33 Specifically, there was that one moment, and I think many of us remember it happened. And I think I got a lot of questions from my newsroom going, what do we do? How does this impact us? [Ethan Toven-Lindsey] 12:07:43 And, you know, sort of Interesting note, the current OpenAI offices, they're actually moving, but the current AI offices are directly across the street from our office. [Ethan Toven-Lindsey] 12:07:55 The early morning team who comes and produces forum often see Sam Altman walking into the OpenAI offices every morning. [Ethan Toven-Lindsey] 12:08:04 And people were like, what does this mean? How do we interact? What are we going to do? And at the time, in the newsroom, we had a a person who was, Lowell Robinson, who has since left to do AI within the industry [Ethan Toven-Lindsey] 12:08:19 But was the sort of senior producer of experimentation. I forget what his official title, but it was sort of experimentation and innovation. And so he ended up sort of pivoting a lot towards AI and focusing on gen AI very specifically and [Ethan Toven-Lindsey] 12:08:33 And our projects. And the first thing we did, and this is maybe we'll get into it later, was really talk about what are the guidelines What are the ways journalists should think about this and the way the KQED newsroom will [Ethan Toven-Lindsey] 12:08:46 Can, should, and will be using AI in the future. And that was a process we intentionally said, we're going to come up with a a document just for the newsroom. [Ethan Toven-Lindsey] 12:08:56 And in tandem, the org itself was also thinking about that as well. [Ernesto Aguilar] 12:09:01 And Ethan, you just set up what I think is really the groundwork and the foundational part of the conversation. [Ernesto Aguilar] 12:09:09 4k cutie because there were a lot of questions. And I think there still are a lot of questions and i think everyone in public media has a lot of concern about what this can be for their organizations. [Ernesto Aguilar] 12:09:21 What Ethan and I have always tried to remind all of our newsroom staff and all of our organization, frankly, is that Public media is fundamentally a relationship business. It is not about replacing content or voices with ChatGPT. [Ernesto Aguilar] 12:09:38 Those individuals are upon which Listenership is built, donor relationships are built, and long-term sustainability for organizations is built. [Ernesto Aguilar] 12:09:49 Our morning edition host, for example, Brian Watt, people tune in for Brian Watt. [Ernesto Aguilar] 12:09:54 They're not here to get a dozen stories that ChatGPT can cough up. They're here for the news. [Ernesto Aguilar] 12:10:01 But they're here for the connection they have with the people. So we're trying to figure out ways that These tools can be used to elevate that as much as possible and try to give someone like Brian more space to do the work that they have [Ernesto Aguilar] 12:10:17 In front of them on a daily basis. [Ethan Toven-Lindsey] 12:10:19 One of the things we've tried to reinforce in the newsroom and probably broader as well. [Ethan Toven-Lindsey] 12:10:26 Has been the Actually, I think the role that or the promise and hope that AI presents two I think in particular, and I'm sure that this isn't true just for local journalists, but to local journalism and local media companies in particular because of that because of that [Ethan Toven-Lindsey] 12:10:51 Like, I mean, I use this as an example a lot. That week that OpenAI and ChatGPT burst onto the scene and became public and people started playing with it was also the week in the Bay Area where there was a levy breach in Pajaro about an hour south of San Jose, sort of Watsonville area. [Ethan Toven-Lindsey] 12:11:11 And we sort of pointed out that AI could definitely help journalists cover that. Maybe sort of synthesize or compile levy histories of levy breaches or use data in particular ways. [Ethan Toven-Lindsey] 12:11:24 But what it can't do is send four reporters, which we did, to Pajaro to cover it and talk to people in the area. [Ethan Toven-Lindsey] 12:11:31 And the promise that that represents is really valuable in terms of our ability to connect. And then secondarily, what Ernesto is saying is that I think our Our trust is built upon that. We've had a number of, Tim Olson, who's on the call, has brought a number of [Ethan Toven-Lindsey] 12:11:48 Groups from other countries who are talking with KQED and public media in general about How are we navigating the this environment? And one of the things that has come up a couple of times that I think other outlets in other places are doing is they're using it for avatars, either [Ethan Toven-Lindsey] 12:12:07 You know, visual avatars may be television anchors or audio voice. And I think one of the things that we've sort of drawn a line at is that that actually does Forget whether it's useful or not useful or not. It breaks the trust. [Ethan Toven-Lindsey] 12:12:24 That is where the trust and credibility that Brian Water, our other individual folks. [Ethan Toven-Lindsey] 12:12:31 That relationship that we have with the audience. And I think we saw everyone sort of has talked about coming out of this last election about the relationship and the personalities that that all media has and that connection with people is even more powerful now, or at least maybe we're realizing how powerful it is. [Ethan Toven-Lindsey] 12:12:48 And we don't want to water that down by making that more by providing a broader space between the actual people doing the actual thing. [Ethan Toven-Lindsey] 12:13:01 And the product and the audience. [Ernesto Aguilar] 12:13:04 Absolutely. So let's talk a little bit about some of the ways that KQED has applied AI in our content division, newsroom, and some of our operations. We're going to share out a number of examples for you. [Ernesto Aguilar] 12:13:20 Hopefully to get inspiration from, not to get intimidated by. Because AI can be a very intimidating and you can see this and go, I don't know how I can do that. [Ernesto Aguilar] 12:13:28 Ethan and I will be more than happy to talk with you about each of these tools one-on-one if you'd ever like to, but hopefully this will give you some inspiration about ways that your organization and you might be able to apply AI for your content efforts and to support your organization [Ernesto Aguilar] 12:13:46 In serving its community best. [Ethan Toven-Lindsey] 12:13:51 So I'll take this one. And this was actually, there are some of these um so the um My team sort of came to me and said, okay, we now have some guidelines and we'll talk about sort of how we thought about that later, because that might actually be [Ethan Toven-Lindsey] 12:14:05 Maybe a possible first step for a lot of our organizations. Like, hey, how do we dip our toe into the water or whatever, like just thinking about it and getting a team of people together to think about how you're going to use it, how it's ethical to use it, how it's not how. [Ethan Toven-Lindsey] 12:14:21 The guidelines has been as valuable as anything else. So that might be a first step. We'll talk about that later. [Ethan Toven-Lindsey] 12:14:26 But once we sort of did that, then we were like, what are some easy enough things Chad said it in the chat that allow our journalists to do more and don't take away the core functionality, the core purpose and what their role in the product is. [Ethan Toven-Lindsey] 12:14:51 And so there are a number of bots out there that you can just plug into Slack. That's how we did this, that will help suggest headlines. [Ethan Toven-Lindsey] 12:15:00 We had a Slack channel before this that was head chat that people did the same thing. They would put in a headline and then their colleagues would come and share and would would put back and forth. We just added this AI bot to be another voice in that. This was not a, you will get this headline and you will put it in your story. [Ethan Toven-Lindsey] 12:15:22 God, no, you know, one that is often inaccurate and two. [Ethan Toven-Lindsey] 12:15:26 It loses that sort of like human connection. But I think what this does is it's sort of a little bit of a yes and. [Ethan Toven-Lindsey] 12:15:32 It allowed us to put in here. A story and generate some headlines. [Ethan Toven-Lindsey] 12:15:37 And I think that that sort of making sure that there's like different ways of going about it You can do prompt engineering in building this to make sure hey, we're trying to sell it or hey, we're trying to make sure that the audience get it, or we're trying to be super Bay Area focused. [Ethan Toven-Lindsey] 12:15:56 You can sort of ask it in such, you can build it in such a way so that the headlines that it's generating are in line with your editorial strategy and your strategy that you're trying to connect with your audience. [Ethan Toven-Lindsey] 12:16:07 But then once those ideas are put there, it becomes a conversation. It becomes a, oh, what's interesting there that we could use or what's not? [Ethan Toven-Lindsey] 12:16:16 And I would note, we specifically sort of limited it to headlines. Other news organizations are using it right now, are using Gen AI to produce social media. They'll share a story that they've written, that they've produced in social media, and then they'll have [Ethan Toven-Lindsey] 12:16:36 Gen AI produced the copy within Instagram or whatever. One. I hope they're fact checking it and make sure that copy that is produced there is accurate. [Ethan Toven-Lindsey] 12:16:48 But also the thing that we sort of held back on is There are people that we employ that do that better, I think, than AI does. [Ethan Toven-Lindsey] 12:16:57 That is part of their job. There are maybe ways to explore doing that in the future, but right now, the way we've sort of synthesized it or settled on it is this headline generator. [Ernesto Aguilar] 12:17:08 I just want to underscore the oh god no part that Ethan just pointed out. Because if you look at what Tim has entered on this screen. [Ernesto Aguilar] 12:17:16 I always think about AI as the anxious intern. It wants to please you based on what little information you've given it. [Ernesto Aguilar] 12:17:23 Tim has given very little information and you can see the wild ideas for headlines that it has come up with here. However, I mean, as Ethan points out. [Ernesto Aguilar] 12:17:33 This tool is just an ideation. Exercise a resource. [Ernesto Aguilar] 12:17:38 For a reporter who may be just struggling with ideas. Many of us have been on deadlines. [Ernesto Aguilar] 12:17:44 We've been working on something for a very long time. We've got several pieces in process and coming up with the most brilliant headline idea without asking your colleagues to help you out with it, sometimes can feel a little daunting. So being able to throw out something [Ernesto Aguilar] 12:17:58 Like a piece of your story and trying to get a few ideas might be just a breadcrumb or two that you might need In order to kick off your own creative process. So this is always a very helpful thing for what we've done. Also, we created a tool through the work of Lowell Robinson and some of our teams [Ernesto Aguilar] 12:18:19 To generate transcriptions of our daily morning program called Forum. And so we'll enter in a url it will deliver back a transcript and can then also give us information about that program, topics. [Ernesto Aguilar] 12:18:38 Names of the guests. It makes a lot of mistakes with names, as you can guess, and also give you a few highlights from the show. [Ernesto Aguilar] 12:18:45 A human will be in the loop around that transcription to ensure that it is correct and that the spirit of the program is also represented correctly because sometimes it's going to look at it without context or nuance. So some part of the conversation that might have really been the most important part [Ernesto Aguilar] 12:19:03 It's not going to necessarily notice. So it's going to be crucial for the human to be in the loop. And then from there, we use this transcription tool to generate descriptions about what that this program might have been without having to provide a full on this person said this piece for five minutes [Ernesto Aguilar] 12:19:22 This person said another piece for three or four minutes. It can also be helpful in a lot of ways in the respect that This tool will also let us know who's speaking and how much they're speaking. [Ernesto Aguilar] 12:19:35 What cities might have been represented and what topics might have been represented. [Ernesto Aguilar] 12:19:39 We haven't really integrated this yet into source tracking. Ethan, you can fact check me on this, but it can be helpful for us to decide These are topics we might want to give a little bit more attention to or an area that we're covering [Ernesto Aguilar] 12:19:54 May not get as much coverage as other areas. And we might want to take a look at that, particularly if we start to look at that data in relationship to what our overall broadcast listenership is. If a community where we have a large broadcast listenership is [Ernesto Aguilar] 12:20:09 List is tuned in, but we're not maybe presenting their community as much as we could. [Ernesto Aguilar] 12:20:14 There was an opportunity. Ethan, I bet I'm missing a lot there, however. [Ethan Toven-Lindsey] 12:20:18 No, I mean, so… I think one of the things that Ernesto and I are trying to center in this discussion is that there are some things that um that both scale, time, resources that sort of all of us can do. One is sort of an AI guidelines, I think is a thing you can you can [Ethan Toven-Lindsey] 12:20:39 Sink your teeth into. I even think a headline generator, both in terms of Ezra asked about cost. [Ethan Toven-Lindsey] 12:20:47 That's a marginal cost. I don't know what it is, but it's like a marginal cost. That would not impact anyone's significant bottom line at all. This tool, the transcription tool, when we thought about this, this was, we won We actually built we actually built [Ethan Toven-Lindsey] 12:21:03 We took existing products that were out in the marketplace, but then built on top of it. So this is a little bit more of a, hey, if you've decided this is a thing you want to invest in and spend on. [Ethan Toven-Lindsey] 12:21:16 This involves like either hiring someone to build something a little bit more custom for you Particularly if you're going to make it public facing, I think that's the reality. As we all know, there's transcription that allows you to do a ton of stuff on the back end. [Ethan Toven-Lindsey] 12:21:31 For all of her newsrooms or for for anyone. What we did here was we figured our forum program has never put our transcriptions online. Different organizations have decided, you know, have either had human transcriptions or other transcriptions previously. We never done that. [Ethan Toven-Lindsey] 12:21:47 And we figured this might be an opportunity for us to sort of public facing explore whether this is a place to do it because both not just cost, but also the workload this adds onto the producers of our daily [Ethan Toven-Lindsey] 12:22:03 Talk show is already They don't have any free time. They will tell you they have no free time for us to add an additional task on them. [Ethan Toven-Lindsey] 12:22:12 So either editing or even QCing a transcript You know, the whole transcript is going to be difficult for them. So what we did is, this is to Ernesto's point, out of the human in the loop that we took a transcription. [Ethan Toven-Lindsey] 12:22:24 We said, this is AI generated. There will be errors, full stop. [Ethan Toven-Lindsey] 12:22:30 And we're going to pull from it three highlights. [Ethan Toven-Lindsey] 12:22:34 One thing that I was super concerned about early on, we said, could you, because you could, you could ask AI to do this. What are the three most important moments of this show? [Ethan Toven-Lindsey] 12:22:43 Well, that feels really qualitative to me that I am not willing to turn that editorial thinking over to AI. [Ethan Toven-Lindsey] 12:22:50 Or what are three places where this could change your mind? Any qualitative prompt felt very stress. I was not willing for the newsroom to make those decisions. But here are three moments that are interesting feels like, hey, if you have a producer that's looking, that's tracking to make sure that that moment, they go, hey, I produced a show. That was an interesting moment. [Ethan Toven-Lindsey] 12:23:13 And that is the correct, you know, that is the correct trans transcription feels, hey, that's a useful role for AIA to play to produce a public-facing product that is valuable. It's sort of, many of us know those fresh air highlight pages. This is an you know a [Ethan Toven-Lindsey] 12:23:33 Ai infused version of that, which has been useful in terms of trying to put some of our forum content into different platforms. [Ernesto Aguilar] 12:23:42 If you're looking to build a bespoke tool, as Ethan was talking about, there are a number of resources that you can use. There are tools like Riplet. [Ernesto Aguilar] 12:23:51 Which R-E-P-L-I-T, which is almost a chat GPT for code. Say you wanted to build something like this. You type in your prompt and where that that output is actually going to end up and it will build it for you. It'll code it for you. [Ernesto Aguilar] 12:24:09 And also deep gram. We use Deepgram with some of the forum tools. [Ernesto Aguilar] 12:24:14 However, I will also add that you can also use Gemini flash The advantage of Gemini Flash is that it will take the audio natively. It's not going to transcribe it and then turn it around and do search. It's going to scan audio [Ernesto Aguilar] 12:24:29 Directly and then look for keywords based on what's in the audio. And that's important because that adds on extra cost. There's an extra token that's involved in transcribing and then there's an extra token in trying to figure out what words go where or what you're exactly looking for in terms of your prompt. [Ernesto Aguilar] 12:24:47 Whereas Gemini Flash is going to just use it based on one token. So that might be an option. [Ethan Toven-Lindsey] 12:24:51 And I would note like the tool we've built that uses replet and Deepgram It's one of those ones where it scales up. So if you do it like for… individual shows here and there, the cost is actually incredibly reasonable. But if you build it as like an ongoing for every hour of forum or every hour of radio or television produced, it can become [Ethan Toven-Lindsey] 12:25:14 Expensive very quickly. And we've had even we've [Ernesto Aguilar] 12:25:18 Yeah, I think it's just a couple of cents an hour. I don't remember the exact figure, Ethan, but I think it was definitely under 10 cents. [Ernesto Aguilar] 12:25:25 Something like that. But as you said, it adds up a couple of hours of this will just go. [Ernesto Aguilar] 12:25:30 Ethan, anyway, sorry, I could talk. We could talk about transcription because it's such an important thing to the audience. On the broadcast side, we have so much content and all of you On the broadcast side of public media, understand we do so much ephemeral content [Ernesto Aguilar] 12:25:43 That the audience oftentimes remembers it, is looking for it. And so we've been really trying to figure out how to solve this problem, and this has been one tool we've been working on to do that. [Ernesto Aguilar] 12:25:55 And let's advance. Ethan, I wonder if you can talk a little bit about discord or is it this in my slide? [Ethan Toven-Lindsey] 12:26:03 Why don't I start, but you can take it apart. One of the things that we've been… We had… with a grant from CPB and the Knight Foundation, we had launched a forum to stick it with forum, to stay with Forum, we launched a project trying to build a digital community in discord [Ernesto Aguilar] 12:26:04 Sure, of course. [Ethan Toven-Lindsey] 12:26:26 One of the things we've learned about is that Discord is actually useful as a little bit of a sandbox for some of these to us. One, it allows for you to build bots and sort of some of these tools into it. It's like just built for that. [Ethan Toven-Lindsey] 12:26:41 And two, I think that you have an audience that's a little bit more digitally savvy in that space who's willing to sort of work with you and play along with you. And so we had built, again, this is a much more bespoke thing [Ethan Toven-Lindsey] 12:26:53 That was using sort of natural language to go, what was today's show about or specific questions about the transcribed version of that show. [Ethan Toven-Lindsey] 12:27:06 This was, you know, sort of debuted and used as a very sandbox sort of product, not a real public facing, not even really beta, more of an alpha. Hey, this is ways you could ask questions. And we've learned The responses are, you know. [Ethan Toven-Lindsey] 12:27:24 Accurate to a degree and we work with it, but it is a much more of like a, hey, what are the promises that that this show, what are the promises and pitfalls that this sort of tool can present to us. Because just like any of the chat bots any of us have used with any customer service things where it's like natural language and whatever. [Ethan Toven-Lindsey] 12:27:44 You can see how you can see how Oftentimes it is both useful and also very limited. [Ernesto Aguilar] 12:27:51 Absolutely. I wish I had more to add. Similarly to the forum bot, we've also been using AI to take our orgreens of local newscasts and trying to provide overviews headlines and topics for the audience, given, again, how much that content is ephemeral. [Ernesto Aguilar] 12:28:13 Once has traditionally been broadcast, it's gone. However, as many of you know. [Ernesto Aguilar] 12:28:18 The audience may have heard a story about loans or school changes or anything your newsroom may have covered. [Ernesto Aguilar] 12:28:25 But sometimes there may or may not be a representation fully online, if at all, for that story. However, for the audience, it is crucial. [Ernesto Aguilar] 12:28:36 We've developed a few workflows to try to take those local newscasts and provide those overviews And Ethan, I believe that's also where this was a jumping point for the latest our podcast that builds off of these broadcasts. [Ethan Toven-Lindsey] 12:28:51 So this one is the one that I most… interested in sort of in the future. None of this has been uh sort of this is sort of us sharing with you the ways we're thinking about it. But we actually haven't [Ethan Toven-Lindsey] 12:29:08 Put any of this public facing out, specifically using summary and metadata for the newscasts. [Ethan Toven-Lindsey] 12:29:13 The things that I am most excited about in it are one, and this is true for probably a ton of content all of us produce, that if it's, we have it automated, we pull it from a, you know. [Ethan Toven-Lindsey] 12:29:27 We pull the newscast. Now, I don't know if this is true for for all of us. But at KQED, what we basically do is we isolate a track and you know every hour during the newscast we We pull it out, we record it, and we put it, we feed it to [Ethan Toven-Lindsey] 12:29:44 Our podcast to our RSS feed. You know, I'm sure a lot of us have a little bit of a human top and tailing, but the same thing exists. [Ethan Toven-Lindsey] 12:29:52 You're pulling a distinct piece of audio or media that might be produced for broadcast. [Ethan Toven-Lindsey] 12:29:58 And then putting it in a digital place. As you can see, this Spotify image that ends up being like the metadata is KQED Newscast for 4.04 p.m. On Wednesday. [Ethan Toven-Lindsey] 12:30:14 You know, August 15th. What we've seen is that the more you put data against that, the more audiences are going to be interested in and are going to either seek it out or be able to disentangle what you're doing there. And so one of the things we thought is, could you use transcription to then summarize to put it in there, put it in as like the summary in your RSS feed [Ethan Toven-Lindsey] 12:30:42 To the point we've mentioned before, there's still concerns about accuracy. [Ethan Toven-Lindsey] 12:30:48 And appropriateness and all the other things. So we have not published any of that because it concerns us. Is that a bridge too far right now But we're thinking about it. It's a thing we're very interested in. [Ethan Toven-Lindsey] 12:30:59 And then the thing that actually is even more interesting to me is that we are really Over the course of the next couple of years, trying to figure out how can we build our content to be much more personalized to the audience. So when they open the KQED app. [Ethan Toven-Lindsey] 12:31:15 Or when they consume any of our content, we're feeding them the content that we know that they may be most interested in. [Ethan Toven-Lindsey] 12:31:21 And what you can do with this transcription is you can then go, what? [Ethan Toven-Lindsey] 12:31:27 Geographic regions in the Bay Area does this newscast mention? And then this we have not built this last mile, but then input that into WordPress or any of our CMS to go this piece of content talks about Berkeley. I'm in Berkeley. [Ethan Toven-Lindsey] 12:31:44 And you could see a world where we are producing 17 audio newscasts a day. [Ethan Toven-Lindsey] 12:31:50 That lives in very specific places, but doesn't really sort of feed to our app right now. [Ethan Toven-Lindsey] 12:31:58 Newscast does, but it doesn't go, hey, you might be interested in this newscast that aired at 5.30 a.m, even though you weren't listening because it mentions Berkeley. [Ethan Toven-Lindsey] 12:32:07 And I would love to figure out this sort of like the next stage is to make sure that we're Adding that metadata against all of this content that we're already making So that we can put it into the pipeline to the audience, go, here's all this content that may be relevant to you. [Ernesto Aguilar] 12:32:30 So one of the other projects, we're in the process of working on at this point. [Ernesto Aguilar] 12:32:35 It is a partnership with Stanford and Berkeley, local universities, regional universities around an investigative journalism website ultimately but it is going to be really a public service Seeking to extract data from incident reports related to police accountability in the state of California. [Ernesto Aguilar] 12:32:58 That is a big project. I think all of us would probably glaze over thinking about all of the reports in our state. We are one of the largest states, so there's probably a little bit of glazing over here when this project [Ernesto Aguilar] 12:33:10 Initially got launched, but it's utilizing AI to take in many cases hard copy reports, as you see here processing each one of those forms and then pulling from those forms select pieces of data that seem to be in common with every one of these forms. [Ernesto Aguilar] 12:33:30 The fact is that of the counties that we are pulling forms from. [Ernesto Aguilar] 12:33:37 The forms don't align. Different counties have different ways that they end up reporting. And so we had to isolate several data sets that we thought these are the ones that we have in common. [Ernesto Aguilar] 12:33:49 Begin to really build out a database to surface out the important stories And also what journalists might need to know about these reports and the arc of the history of these reports. [Ethan Toven-Lindsey] 12:34:06 I sort of maybe I'm in the same thing you said last time, Ernesto. I don't have much more to add except that the promise of what this can do with data extraction is incredibly valuable. We are working with universities and partners. We helped sort of launch this project four years back, and now we've sort of handed a lot of it off because it is like deep [Ethan Toven-Lindsey] 12:34:27 Data extraction and data work to our partners sort of like figuring out how there's reporting out of it is a thing we're super interested in. And I think the project, as much as anything else is working to… bridge that gap between the big data, the big data database, the big data that we're going to have, and the reporters, how the reporters are going to use it to turn it into stories [Ethan Toven-Lindsey] 12:34:54 To draw stories out of it. But this is, in terms of that sort of continuum of different projects, this is at the, hey, we're trying to figure out how it might be useful stage, not the, hey, this is how we're deploying it in our newsroom stage. [Ernesto Aguilar] 12:35:10 And finally, all of us have lots and lots of archives, more than likely, whether you're TV or radio and KQED is in our own process of reaching into our archives and getting a more comprehensive set of metadata around what we actually have. [Ernesto Aguilar] 12:35:32 We have everything from, as you see on the screen, a James Baldwin who was who appeared on KQED to important historical moments in the history of the Bay Area. [Ernesto Aguilar] 12:35:44 But in many cases, those have been in storage for a long time. [Ernesto Aguilar] 12:35:49 So we've really sat together and deeply thought, how can we utilize this history in a way that comes to life for the audience This is its own webinar in and of itself, so I don't want to spend too, too much time on it. [Ernesto Aguilar] 12:36:04 But if you're thinking about how you might be able to use your own archives. [Ernesto Aguilar] 12:36:08 Please reach out to Ethan and I because I think that we would have a lot to share about this and could connect you with some of our colleagues that are most actively working around the archiving systems that we have here and get you into some of the operational components of that. [Ernesto Aguilar] 12:36:26 And with that. Actually, I apologize because we had one last item, which was to give a shout out to those of you who work with your local schools. If you work with your local organizations, KQED also has an education department [Ernesto Aguilar] 12:36:43 That works with the nine counties within our broadcast area. And those schools like your schools more than likely are dealing with AI not as something you're introducing But something that's already entered the classroom. [Ernesto Aguilar] 12:37:04 Parents have questions. Students don't know how to use it. Some of them are using it in ways that are inappropriate. Teachers are figuring out how to cope with that. And parents are looking for ways to talk with their young people about [Ernesto Aguilar] 12:37:20 How to use AI and what it means for their futures because As with us as adults who aren't in middle school, high school. This has created some existential dread for young people as well. [Ernesto Aguilar] 12:37:33 So our education department has created generative AI guidelines as well as media literacy tools at education.kqed.org. [Ernesto Aguilar] 12:37:43 You're welcome to take a look at those, download any of those resources, attend the webinars that are being provided for teachers and school administrators as well as public media professionals about how to use AI effectively in the classroom. [Ernesto Aguilar] 12:38:00 And also for educators how to effectively talk about AI. Ethan and I are more than happy to connect you with any of those staff in our education department. [Ernesto Aguilar] 12:38:17 And with that, Ethan, I think this is the hard part of the conversation. We got a chance to show you some of the tools and talk to you about some of the ways that we're using it and some of the ways We got it to work. But the hardest part is around [Ernesto Aguilar] 12:38:33 Culture. It's around change management. It's not just resistance to AI, but it's also just indifference. Is this one more thing I have to do? [Ernesto Aguilar] 12:38:43 Is this one more thing that I have to try to figure out how to do when I'm not exactly sure if it's going to be of help. [Ernesto Aguilar] 12:38:50 So Ethan and I have really had to focus in quite a bit on how to talk to our organization about AI and the change that it's going to be for public media. [Ethan Toven-Lindsey] 12:39:04 I think you can hear it through some of the things we're talking about. I think one of the things we've intentionally try to do is like involve as many people as possible, be as transparent as possible and I mean, we're seeing it in the chat. We still continue to see it at KQD. You're going to see it anyway, this sort of change in awareness is [Ethan Toven-Lindsey] 12:39:25 Is difficult. It's totally understandable. And I think the thing that I've tried to base it around is this sort of through line that we've been talking about here and in the chat, which is the point is not to go, hey. [Ethan Toven-Lindsey] 12:39:39 How do we replace this job? Or how do we save money by doing? It's are there things that we aren't able to do right now because of resources or otherwise, that this allows you know, particularly in the newsroom, but I'm sure this is true in other departments as well, allows the content creators to stay involved. It's not just hitting a button and then moving on. [Ethan Toven-Lindsey] 12:40:05 To stay involved, but do it in such a way that allows them to better produce and better provide for the audience it's like a tool that allows them to do their job to add more to their job without adding more to their workload. And I think that has been where this change management has been most effective. Hey, you've long wanted to do this forum team. Here is something that helps you do it in a way that you feel comfortable and confident and ethically [Ethan Toven-Lindsey] 12:40:35 Journalistically, you know. [Ernesto Aguilar] 12:40:38 I'm just going to call out some things on the slide here, which you've probably noticed as well, and there may be questions. [Ernesto Aguilar] 12:40:44 I'm looking forward to getting to some of these questions in the chat, by the way, because there's some really meaty discussions. But KQED did create an organizational policy, but we also have separate AI guidance for content just because content is the front line to the audience [Ernesto Aguilar] 12:41:01 And how each part of the house, so to speak, should be looking at AI. We've also done a number of other efforts, including brown bag lunch types of conversations talking about AI with staff, introducing some tools, answering some questions, doing a little bit of hands-on work. [Ernesto Aguilar] 12:41:21 And then also just recently, we've only done two of these at this point, so it's not a regular occurrence, but tried to do some design sprints. And what a design sprint basically is is We sit down together, try to figure out what's a problem that we as stakeholders in the organization can identify we might be able to figure out [Ernesto Aguilar] 12:41:39 A way to help AI solve. And then spend a few days to a week putting some resources to develop and prototype some potential solutions. [Ernesto Aguilar] 12:41:51 So that can sometimes be very successful and sometimes it just doesn't really go very far. But that's, I think, part of the effort to try to ideate and experiment a little bit and also giving ourselves as an organization releasing the pressure valve, so to speak, and reducing some of that stress to say it is okay that something does not [Ernesto Aguilar] 12:42:11 Necessarily go right to market, but rather it's developing this comfort level with experimentation and with saying that something might or might not necessarily be the thing that vaults us into the stratosphere, but it's baby steps of experimentation for us. [Ernesto Aguilar] 12:42:29 Regardless of any size of organization, I know that that is a fear for all of us that we want to experiment a little bit and it might not necessarily fly. But we're trying to give ourselves more permission around that. [Ernesto Aguilar] 12:42:42 Among our content teams, we did launch through Lowell Robinson, who we mentioned a little bit earlier, and now I lead up our AI working group, which is a cross-departmental committee that looks over many of the AI initiatives that are happening around the organization and tries to also unsilo some of it because sometimes in any organization. [Ernesto Aguilar] 12:43:05 Silos do happen. And AI is no exception. We're also in the process of trying to familiarize staff with what the AI tools are and how we might be able to implement them a little bit more. [Ethan Toven-Lindsey] 12:43:16 I mean, I do think that this is, if I were to um recommend, and I sort of hate doing that, but if I were to recommend what other folks can do, it would be gather a group of interested and relevant and key stakeholders and write up an organizational guidelines for AI [Ethan Toven-Lindsey] 12:43:36 And then on my side, you could do this for individual departments, but I think especially audience facing news or content focused right up and come up with the organizational guidelines for For content and journalism. That has been both useful so that we can answer questions within the newsroom. Here's what you can and can't do with it. [Ethan Toven-Lindsey] 12:43:59 But also, I think it's that change management. It's got people thinking about it as opposed to going, oh, God, that's over there. That scares me. [Ethan Toven-Lindsey] 12:44:07 I don't want to address it or I don't want to worry about it right now. And I think that it's something we all can do. [Ernesto Aguilar] 12:44:17 And I think Ethan just touched on, I think, the two biggest parts of what any organization I think should be thinking about because we've had to think about it and learn from very painful experience. [Ernesto Aguilar] 12:44:27 That coalition really does mean a lot. Having people that really care so it's not just one person. Even if somebody who half cares or one quarter cares. [Ernesto Aguilar] 12:44:36 That is going to be beneficial, especially if you've got it across the organization in different tiers. [Ernesto Aguilar] 12:44:42 Because those people have different networks potentially than you do and might have a different sphere of influence. [Ernesto Aguilar] 12:44:49 Also, I think, just building on what Ethan mentioned. Talking to staff about what they need as opposed to and how we can might be able to solve that need as opposed to why it scares us. You want to have that conversation for sure, but also really thinking about how we might be able to address something [Ernesto Aguilar] 12:45:08 That we as an organization identify we really would like to see us go. [Ernesto Aguilar] 12:45:14 Last thing I would… [Ethan Toven-Lindsey] 12:45:14 It also this also as a process this also Deb just talks about how it moves and develops so fast. This has been a learning for us, which is that we sort of came up with guidelines sent out an email and said, we're done here. [Ethan Toven-Lindsey] 12:45:28 And we realized, oh, no, no, this is evolving. We're going to have to change. We're going to have to adapt our guidelines to new and changing technologies and things within this field. [Ethan Toven-Lindsey] 12:45:41 And I'll kick it to Ernesto because that's actually a thing we're thinking about right now, which is now we review it every couple of months and go, does it need any changes? It doesn't always, but when it does. [Ethan Toven-Lindsey] 12:45:50 Then it's really useful. Right now, we're thinking about agentic AI, how different changes within might impact what we're doing as well. So I think there's a There's a process that is also valuable too, that it's not just like, hey, we solved it and we can move on now. Here's our answers. [Ethan Toven-Lindsey] 12:46:07 It's like, you know, this is continuing to develop. [Ernesto Aguilar] 12:46:11 Absolutely. No, I think that it's… I think it tends to be the tendency of everybody. I say the policy is done, yay, but in this particular case. [Ernesto Aguilar] 12:46:21 Because as Ethan pointed out, things are evolving so fast. It is a policy, whatever you write, is something you want to evaluate regularly. Agentec AI is a good example. It may not necessarily impact a lot of content. [Ernesto Aguilar] 12:46:35 But it may impact the rest of operations. And it is being employed in a lot of cases of operations. [Ernesto Aguilar] 12:46:42 Hr departments are having these products pitched. Lots of finance teams are having such products pitched and Just having some clarity about what that is organizationally and having that periodic policy review is going to be really essential. And I think the last thing I would mention on this slide is just starting small. [Ernesto Aguilar] 12:47:00 And I, at the experimentation, I wrote down things like asking AI to critique something, asking it to write something for you that would otherwise take you hours to write just as a test to see what it does. [Ernesto Aguilar] 12:47:13 Something I've done in a number of cases because I look at a lot of underwriting language I have a test my assumption about a piece of language and write a prompt that says something like. [Ernesto Aguilar] 12:47:25 You are an expert in underwriting case law. Tell me if my assessment of this piece of language is correct and give me citations as to why it is or it isn't. And it's not going to be gospel, of course. I'm not going to take it that way. But if it identifies something for me [Ernesto Aguilar] 12:47:44 That I wouldn't have otherwise caught. I've basically saved myself a lot of headaches. [Ernesto Aguilar] 12:47:50 And with that, I think we're decent on time for questions. [Ernesto Aguilar] 12:47:54 And we've covered a lot. And I think Ethan and I are more than happy to field your questions. [Ernesto Aguilar] 12:48:00 After this, but let's kind of get into the chat and get into some conversation. [Chad Davis] 12:48:04 Yeah, I've been pulling stuff, so let's dig in. We've got at least 12 minutes left. [Ernesto Aguilar] 12:48:09 Great. [Chad Davis] 12:48:09 Sometimes we go long. We'll try to end this one on the top of the hour for everybody. But I'm going to start with one from Maxi Jackson, who just uh raise their hand. So let's see if this covers it. [Ernesto Aguilar] 12:48:19 Excellent. [Chad Davis] 12:48:23 Maxi says, I'm most interested in how this can assist public media in reaching broader audiences. [Chad Davis] 12:48:28 Can AI be ethically, culturally sensitive? Which is, I know something, Ernesto, you have given a lot of thought to over the last couple of years. [Ernesto Aguilar] 12:48:35 Yeah. No, I think that it really is about just being very intentional about A, what you're looking for, specifically what you're looking for, and also how you prompt. [Ernesto Aguilar] 12:48:49 So if you ask a fairly general prompt for, say, I'd like an outreach plan for a community we don't traditionally serve, for example. [Ernesto Aguilar] 12:49:00 You're going to get a very, very general answer that's probably not going to get as deep as you would like to. So in order to get to more of what you're looking for. [Ernesto Aguilar] 12:49:13 Or at least some ideation around hopefully what you're looking for, it might be helpful to provide as much information as possible. [Ernesto Aguilar] 12:49:20 For example. We are trying to reach out to East Houston, which is traditionally a Mexican-American community from the 1920s into today previously was a working class Irish and American community and has three different parks and four community centers located on the north, south, east, and west of the neighborhood. We would like to also do some outreach to [Ernesto Aguilar] 12:49:45 The local library. There are two in that community which have these hours and these hours. And they do, for example, whatever it might be. [Ernesto Aguilar] 12:49:56 How would we tailor an outreach program to reach Parents of children two to eight. [Ernesto Aguilar] 12:50:03 Again, that's hyper specific but Going back to what I mentioned about being the anxious intern, it's going to try to please based on what you give it with your prompt. [Ernesto Aguilar] 12:50:13 So if you can give it as much as you possibly can around that prompt. [Ernesto Aguilar] 12:50:18 Is going to be very beneficial. The human needs to be in the loop if you know this community that you're trying to do outreach to and you go. [Ernesto Aguilar] 12:50:26 This is not going to work. By all means, honor that instinct, but do your best to try to zero in on what you're looking for most specifically. [Ethan Toven-Lindsey] 12:50:35 I just want to like plus one to that last part, which is the human element. [Ethan Toven-Lindsey] 12:50:39 We did a project, we had an experimentation sort of led by Tim Olson in terms of working with Google. [Ethan Toven-Lindsey] 12:50:46 The Google News Initiative on transcription. This was a couple of years back. [Ethan Toven-Lindsey] 12:50:50 And we were like, what are some things that we can learn from it? And one, I use this as an example all the time because I think it's very valuable. [Ethan Toven-Lindsey] 12:50:57 Is that we were trying to figure out how can, you know, transcription can be 99% accurate, but it will miss proper names a lot or places or things like that. [Ethan Toven-Lindsey] 12:51:06 We did a whole hour show on forum about misgendering people. [Ethan Toven-Lindsey] 12:51:12 And the transcription, the whole time he said miss Gender. They were gendering the term of the whole, they were misgendering The whole entire episode. [Ethan Toven-Lindsey] 12:51:23 And understand that if you're publishing that without having someone who's culturally sensitive or thinking or aware. [Ethan Toven-Lindsey] 12:51:30 That's going to be an issue. And so knowing both the limitations that you're asking the tool to do and making sure a human is there to sort of catch those limitations and understand where you're going to be able to need to push will be is critical. [Ethan Toven-Lindsey] 12:51:44 None of these things, all of these things we showed you, well, even the ones that are public facing are only public facing with the buzzword, the human in the loop, that we are not doing anything [Ernesto Aguilar] 12:52:02 The last thing I would add to what Ethan said about humans and being aware of that. Also just be very cognizant of what outcomes that might not be intended. For example, I've talked to organizations that have wanted to use it for Spanish translation. [Ernesto Aguilar] 12:52:20 Because they don't have enough staff to do it. And I asked, well, okay, so what happens if someone who is monolingual Spanish speaking comes to your board meeting. [Ernesto Aguilar] 12:52:31 Comes to your office, comes to an event. Can someone engage them? [Ernesto Aguilar] 12:52:36 Well, if they can't, try to think about how you can scale that eventually. I think it's a great intention. [Ernesto Aguilar] 12:52:43 But also, you might have an unintended consequence that it works and then you have a lot of people who come to these events And it turns out nobody in your staff can talk to them. [Ernesto Aguilar] 12:52:52 So just be thoughtful about what you're hoping to get out of it. [Ernesto Aguilar] 12:52:57 Who you're hoping to target and then what you might get and how your organization is built to serve that audience too. [Chad Davis] 12:53:08 Maxie had a second question. We'll touch it really fast because I think it's a short answer. But instead of piecemeal approaches, is there a turnkey solution for AI that does all of these different things? And I think in this instance, we're talking about sort of the content creation. [Ethan Toven-Lindsey] 12:53:22 I think you're right. It's a simple answer. The answer is no. There is no turnkey crops out there. I think all of these… Companies are trying to build them. Adobe's putting it in all their stuff. And so I think you watch that and you see if this comes, but as it currently exists, what we've [Chad Davis] 12:53:23 Process. [Ethan Toven-Lindsey] 12:53:42 Really found is that you've got to pick and choose and build or use this on your own. There is nothing out there that's doing all these things. [Ethan Toven-Lindsey] 12:53:50 In one package simply for us. [Chad Davis] 12:53:53 Right. [Ernesto Aguilar] 12:53:53 Plus one to that, but I would also say keep an eye out, I think, in the next year or so, this is going to, if not by the end of the year, there are a lot of organizations Verso comes to mind and some others. [Ernesto Aguilar] 12:54:03 That are trying to scale this for nonprofit journalism and newsrooms in particular, but there's going to start to be coalescing and some shaking out. [Chad Davis] 12:54:13 Cool. Let's turn our attention to the idea of what AI is doing to search. Two sort of questions, one from Stane gruntles and then Jeremy Ruggles also chimed in just a second ago. And essentially, it kind of comes down to like [Chad Davis] 12:54:31 How you are all thinking about AI and AI's impact on seo Jeremy, well, we'll do that one first because I have a Shane and I'll kind of focus in on Jeremy's jailbreaking question. So what are you guys thinking about it? [Ethan Toven-Lindsey] 12:54:42 I mean, I think that they are connected together and it's, I mean, we can burrow down but i what i'm going to say start as the top level. We've given up On search. I mean, in a way, not really, but as a strategy, as an organizational strategy. [Ethan Toven-Lindsey] 12:55:01 The goal is now to build a direct connection with our audience through an app, people go to the homepage. The homepage used to be dead. It's not dead anymore, everyone. Encourage people. [Ethan Toven-Lindsey] 12:55:13 Building that direct connection and having people come to you to consume what you're producing because they trust you, because they know you. [Ethan Toven-Lindsey] 12:55:22 To me, that's the only viable strategy going forward. It is not going to be how do we [Ernesto Aguilar] 12:55:33 And to what Ethan said, I would say for all of you who are on the operational broadcast side or the content side, by all means, take the time, the attention you have with your audiences because chances are it is your biggest audience beyond digital, beyond podcasts and everything else. [Ernesto Aguilar] 12:55:49 To try to introduce those broadcast audiences to what you're doing online and to start establishing, as Ethan said, that direct relationship to digital properties that you've got, whether even if it's just your website. [Ernesto Aguilar] 12:56:03 Constantly or at least once an hour regularly mentioning the website that what they can get there, where they can go, why they should go there, and building that habit and that relationship just a little bit more with the audience is just going to make that [Ernesto Aguilar] 12:56:20 Effort to scale that much easier. [Ethan Toven-Lindsey] 12:56:23 Maybe as an example to Shane's question, if you see that your audience is asking what are the 10 most important stories every day. [Ethan Toven-Lindsey] 12:56:30 You need to write a newsletter that answers what are the 10 most important stories every day and send that to your audience. There are ways to use some of this audience data of lessons of what they're interested in and how they're trying to use your content. [Ethan Toven-Lindsey] 12:56:43 To produce content that you connect directly with them as opposed to hoping that it will bubble up in some search or some other. [Chad Davis] 12:56:52 Now, are you guys advocating then Jeremy says, teaching journalists kind of jailbreak techniques Which for those who are not super, super techie jailbreaking is what people used to do to put altered at iOS or altered OSs on phones to get them to run you know and [Chad Davis] 12:57:10 A way somebody would prefer. That's not an exact definition, but jailbreaking techniques to or like running models locally [Ethan Toven-Lindsey] 12:57:20 I mean, we've talked about that. We're thinking about it. A lot of this stuff is like… that sort of like, hey, that question of putting anything in the model, they're using it for training and then vice versa, anything that they're going to then suppress [Ethan Toven-Lindsey] 12:57:36 These are questions that were sort of like grappling with, but we don't have any answers to. And I don't think anyone really does, but certainly even at KQED, we have not. [Ethan Toven-Lindsey] 12:57:44 Solve that. [Chad Davis] 12:57:47 Next question from Carson Risser. How have you all handled privacy concerns about transcription software or other AI software using the content that you feed it. Have you found software that handles your data privately instead of using it to train the software kind of almost ties a little bit into Jeremy's question about locally. [Chad Davis] 12:58:08 Local models. [Ernesto Aguilar] 12:58:11 No short answer. I think… Ethan, you probably would be the smartest person to answer this one, but I think we've also been very careful about educating staff about what kind of data you put in there. If this is a sensitive source, if this is a sensitive subject, if this is something you don't want [Ernesto Aguilar] 12:58:31 For example, just… put online or shared out with a model And the source is confidential, don't put it out there and be very mindful about the fact that all of our data has probably already been used in training, frankly, but there's no reason to give it something that's confidential or a source that [Ernesto Aguilar] 12:58:49 Would prefer to remain anonymous if you can do that. [Ethan Toven-Lindsey] 12:58:52 I use this as an example in response to this question. My wife's an academic. She does big data surveys with like personal identifiable data that AI could help incredibly with And she can't use it because it is personally identifiable data that she cannot put [Ethan Toven-Lindsey] 12:59:09 In any of these models. And there are, yes, you can build your own sort of personal ones. Those are expensive, hard. We're not anywhere near that. [Ethan Toven-Lindsey] 12:59:17 And I think these were sort of in the trying to figure it out stage of that, that some of these is just you have to earnesto's point. You can't use it for that stuff right now. [Chad Davis] 12:59:27 Cool. I think we will loop over the top of the hour a little bit. I know some people are going to have to go. Thanks for everybody who's been sticking in. [Chad Davis] 12:59:35 But we've got maybe three or four more questions. I want to try to get to those. Christian Knitson says. [Chad Davis] 12:59:41 Are you concerned that the use of AI headline generator Our use of an AI headline generator for ideas, even if only at that level could incentivize the atrophy of skills of journalists and editors to use their brains and be creative on their own. So like, will this subvert creativity and journalistic skills? [Ethan Toven-Lindsey] 12:59:59 Yeah. [Ethan Toven-Lindsey] 13:00:03 I think this is an incredibly great conversation to have. I don't think so. I think it's a yes and situation where just adding more ideas If you were like, this is the headline and you can only pick from one of these, maybe. But I think what I've seen in our headline channel is that there's more conversation and more [Ethan Toven-Lindsey] 13:00:21 Creativity, actually, because it's a little bit of a spark of a yes and situation. [Ethan Toven-Lindsey] 13:00:25 I will note, though, this is actually like a bigger, broader concern about atrophying, you know, sort of journalism skills I've talked to a lot of business journalists who say, hey, AI could be great to write earnings report stories. [Ethan Toven-Lindsey] 13:00:41 About earnings reports. You just take an earnings report, spit out a story. [Ethan Toven-Lindsey] 13:00:46 Every business journalist that I know got their start writing earnings report stories and learn how to be an incredibly strong business journalist By doing that. Bethany McLean, who broke the Enron story, talks about how earnings report stories like allowed her to figure out what was going on at Enron. [Ethan Toven-Lindsey] 13:01:02 And so those questions about could this pull away key skills that we like train our, not train the AI, train journalists, train creative workers on is an incredibly important. [Ethan Toven-Lindsey] 13:01:16 And, you know, sort of like must be answered and must be addressed question. But to the headline generator, I think because it's like. [Ethan Toven-Lindsey] 13:01:24 It's more of like a spitball popcorn thing. It doesn't feel that it's doing that. But I get your point. The question is a good one. [Ernesto Aguilar] 13:01:31 Yeah, I'd be really concerned about it for commercial news organizations, frankly, because they're trying to scale so fast with so many different stories. I have a friend of mine at politico who said, you know, we do a legislative tracker where we're tracking 500 odd bills. [Ernesto Aguilar] 13:01:48 We've done this via AI and I'm terrified of it because we need to represent all these stories because people expected of us But we don't know if it's accurate until someone flags it. And so he's just like, well. [Ernesto Aguilar] 13:02:03 What can we do? At Public Media, fortunately, I don't think has that kind of problem. [Chad Davis] 13:02:10 This will be the last question, I think. And it's from Hannah Rothblatt. And this goes back to when you were talking about the archives. Hannah writes, what is the purpose of using AI to pull advanced metadata from archives? Can you dig a little deeper into into that. [Ernesto Aguilar] 13:02:29 So I know that our colleague at KQED, which would be happy to connect you or colleagues rather at KQED, I'm going to give, I think, probably the most Crayola explanation. Ethan, please don't let me color on the walls here, is that we are processing, we're taking our old recordings. So in many cases, this is [Ernesto Aguilar] 13:02:51 Actual tape, digitizing them, digitalizing them, depending on where you come from, different people have different ways of pronouncing that. [Ernesto Aguilar] 13:03:00 And then feeding the audio components of that through AI, through transcription. [Ernesto Aguilar] 13:03:07 And then asking for themes and asking for some summaries. And then going back and rechecking that. [Ernesto Aguilar] 13:03:15 Ethan, did I get that right? I feel like I should be getting the buzzer now. [Ethan Toven-Lindsey] 13:03:19 You know, I think this is one where we are we probably we should connect you with our with our archive. We have like a deep archive staff who's looking at that. [Ethan Toven-Lindsey] 13:03:28 I do think one of the thinking is, is there are ways that you can resurface it to like you know, to create This James Baldwin clip is very relevant today because of something happened that relates to it and then sort of like resurfacing it both [Ethan Toven-Lindsey] 13:03:45 The archive itself and then maybe doing content around it. But I think that there might actually be archival uses for it that [Chad Davis] 13:03:56 All right. I think we're going to start to wrap it up then. [Chad Davis] 13:04:00 Join us. I mentioned this at the top, but join us May 13th, a Tuesday, which is different for us, at noon Eastern, 9 Pacific for our next webinar. We're going to focus this next time is going to be on marketing. [Chad Davis] 13:04:14 Ezra Wall made a point of kind of thanking us for examples of actual tools that are being used. That is going to be The focus of this next webinar. We've heard that from you before. It's a bit of an extension from a webinar we did, I think it was in December. [Chad Davis] 13:04:29 And people wanted to hear more, wanted us to go deeper. So we are bringing that around. [Chad Davis] 13:04:35 In May. If you are headed to the PBS annual meeting, I will just mention, no one asked me to do this, but I'm interested, so I'm going to just tell you. There is a session, I think it's on the 20th. Well, there's one by Mikey Centrella where they're going to get in, get kind of dirty with the code and on how people are coding for For AI use. [Chad Davis] 13:04:58 And then there's one with Richard Traylor who does search engine marketing and SEO for PBS. And they're actually going to start to talk about the AI's impact and what people are thinking about how the adaptation needs to start happening, how our SEO practices are going to [Chad Davis] 13:05:16 Are going to need to change. I don't want to over promise on their behalf, but hey, somebody's having the conversation on a national level on And that's something we should support if that session works out for you. [Chad Davis] 13:05:27 Time-wise, and you're going to be in Atlanta. We're talking with them about possibly bringing that to the webinar group later this year because I know not everyone can spring for that reg fee. Went up a little bit this year. [Chad Davis] 13:05:42 And I think that's it. I'll close with just echoing Deb Herbold's quote, because I thought it was such a good one. [Chad Davis] 13:05:49 Oh yeah, David says stick around for tech summit on Thursday. If you can. [Chad Davis] 13:05:54 It's a smaller pool of people, but get on that wait list if you haven't yet. Deb said it all moves and develops so fast. Some of our move relatively slowly. So a challenge for us would be how do we use develop AI effectively, also be flexible [Chad Davis] 13:06:09 Enough to grow and change as we go forward. Ethan cited this. It resonated with him. It resonates with me. We're all kind of in that boat. [Chad Davis] 13:06:19 I think that we try to present through these webinars people who have taken a particular focus so that we can kind of inspire. But, you know. [Chad Davis] 13:06:28 Nobody is like that far out ahead of anybody else when it comes to how AI gets adapted, especially in the media landscape. There are no turnkey solutions because no one's figured that solution out yet. [Chad Davis] 13:06:41 And I kind of joke sometimes, those of us who get up and do these webinars, we just happen to be the ones that either live across the street from OpenAI, as you all do, or happen to be like, had some free time the weekend that ChatGPT dropped and we're just five minutes ahead of everybody else. So keep that in mind, don't uh you know uh [Chad Davis] 13:07:03 Don't let that kind of despair take over and cause you to not dig into this. This is the future and this is, you know, the future public meeting. It's also the future of your careers. So I just wanted to call that out. So I think that's [Chad Davis] 13:07:17 I think that brings us to a wrap. So Ethan, Ernesto, thank you guys. Really appreciate it. [Ethan Toven-Lindsey] 13:07:21 Thank you, Chad. Thank you all. Thanks for the great questions. [Ernesto Aguilar] 13:07:22 Thanks, everyone. Thank you. Thanks, y'all. [Chad Davis] 13:07:25 See everyone next month.