Dan Grech joins host Tim Regan-Porter again to explore the practical applications of AI in journalism. This episode delves into how AI tools are currently being used in newsrooms to free up time for journalists and better service audiences.
The conversation highlights the potential of AI to influence storytelling and marketing, its impacts on journalism and local news, and the changes it could bring about. Grech and Regan-Porter discuss various practical AI use cases for journalists, including processing police reports, automated translation of weather reports, and creating transcripts of public meetings. They emphasize the importance of “human in the loop” in AI applications and the need for journalists to adapt to new tools and methodologies.
This episode is part of a series exploring the opportunities and challenges AI presents for journalism, offering insights into how AI can be a powerful ally in the mission of truth and storytelling.
(05:27) – Practical AI use cases for journalists from AP
(09:24) – Use Case 1: Processing police reports
(13:12) – Lesson: AI can free up time to focus on key tasks (but automation is difficult to maintain)
(15:39) – Use Case 2: Automated translation of weather reports
(18:39) – Use Case 3: Transcripts of public meetings
(23:01) – Use Case 4: Classifying press release emails for assignment
(26:13) – The importance of “human in the loop”
(27:58) – The importance of working on *how* you work as a journalist
(32:02) – Impact, not just efficiency
(34:40) – Use cases for existing off-the-shelf tools: writing support
(40:22) – Use cases for existing off-the-shelf tools: interview prep
(42:43) – ChatGPT tips
(45:56) – Data analysis and other miscellaneous uses
<h2>Listen to the episode here:</h2>
- Dan Grech: LinkedIn, Twitter
- BizHack Academy: web, Twitter
- Local News Matters: web, Twitter, Facebook, LinkedIn
- Colorado Press Association: web, Twitter, Facebook
- Tim Regan-Porter: bio, Twitter
Dan Grech is the founder and lead instructor of BizHack Academy, which provides digital marketing training to corporations and marketing executives businesses across South Florida. He was the News Director at WLRN (Miami’s NPR station) and was part of a Pulitzer Prize-winning team at The Miami Herald. He co-hosted Miami’s first podcast, Under the Sun.
He’s worked at The Washington Post, Marketplace and PBS’s Nightly Business Report. He’s also worked as the head of digital marketing at two software startups and the nation’s largest Hispanic-owned energy company. Dan is an active member of the South Florida startup ecosystem, where he’s mentored companies at the Goldman Sachs 10,000 Small Business Program, Babson College’s WIN Lab, StartUP FIU, The Idea Center at Miami Dade College, the Innovation Hub at Broward College, the Watson Institute at Lynn University and the Inter American Development Bank. He’s taught at top universities including Princeton, Columbia and University of Miami.
Dan is a graduate of Princeton University and has a Masters degree in storytelling from FIU and journalism from Universidad Torcuato di Tella in Argentina. He’s a father of two, his wife Gretchen Beesing is the CEO of Catalyst Miami, and his favorite color is purple.
Dan Grech [00:00:00]:
So when you're thinking about automation in your local news outlet, make sure you're automating the stuff that will make a huge difference to your audience and any repetitive tasks that are high value to your audience, but don't really require human intelligence or human-to-human intelligence or data gathering, or really a lot of discernment from the reporter. There are always in every job, some mundane tasks. And so the more that you can find high value to your audience, but low value provided by the reporter tasks, that's a good task that's eligible for automation.
Tim Regan-Porter [00:00:49]:
Welcome to the Local News Matters podcast, where we explore pathways to stronger journalism, better businesses, and healthier communities. I'm Tim Regan-Porter, CEO of the Colorado Press Association. In each episode, I sit down with guests from newsrooms and others in the local news ecosystem to highlight the innovative work of local newsrooms and those that support them, as well as the crucial questions they face. Today, I'm joined by Dan Grech again, and we're revisiting a topic that's not just on the horizon of journalism but is already reshaping its landscape: Artificial Intelligence. In our last conversation with Dan, we delved into the broad implications of AI in journalism. Today, we're honing in on something more immediate and tangible: the practical applications of AI in newsrooms. This isn't about distant futures or hypotheticals; it's about the tools and technologies that are here, right now, and how they're being used to inform, engage, and sometimes, challenge our traditional notions of news. AI is a topic we'll keep revisiting because the technology keeps developing and because the implications are enormous and there's so much to untangle. I truly believe AI will be disruptive and transformational in ways we haven't seen since at least the invention of the internet. That is not hype. This is not crypto-currency or the pivot-to-video. Actually, it may not be all that dissimilar to the latter, in that while the pivot-to-video was fueled by some bad data and drove very flawed and sometimes harmful strategies from companies that didn't understand video—its core language, its appeal, its uses and audiences—or how it would evolve, it nonetheless pointed to a real change in consumer behavior. Not for all consumers, certainly, but there is a large group of people who get their information primarily from video, from YouTube, TikTok, Instagram Reels and the like. Meeting that moment required more than having a reporter grab a cellphone and tell the same story in fundamentally the same way. Similarly, there will be hype cycles with AI and many failed experiments, not only at the platform and big tech levels but at the newsroom level and really for all companies trying to use this technology. But the changes to consumption will be profound. It's one of the least talked about implications of this new generation of generative AI tools, perhaps because it's impossible to see the exact shape these new tools will take. But I think it has some of the greatest near-term potential to disrupt the news industry, and all companies who've built their businesses on the web.
On a recent episode on AI from Ezra Klein's podcast, Casey Newton summed up the conversation I've been trying to have in the industry since ChatGPT‘s public release. In talking about some of the relatively minor updates to ChatGPT this fall, he said, “you start chaining them together, you start building the right interfaces, you actually start to see beyond the internet as we know it today. You see a world where the web, where Google is not our starting point for doing everything online. It is just a little box on your computer that you type in and you get the answer without ever visiting a webpage.” As he points out, that's going to take years to fully unfold, but we see the beginnings of it. That presents both huge threats and huge opportunities for new organizations. And that's something that I want to explore in depth over time.
Of course, AI and especially the potential for AGI, artificial general intelligence, raise all sorts of other huge issues. From bias to disinformation to its use in warfare to job loss and displacement to what some fear is the potential to end humanity. Those are little beyond the scope of this podcast and we won't explore those in any depth.
But for now, I hope that we're all paying attention to not only those large issues and demanding guidelines and accountability as this technology develops, but also exploring the very practical uses of AI that are available here and now. For all of the concerns and rethinking we need to do, we must also learn about the power, benefits and limitations of the tools as we have them. To not do so because of these other concerns is to stick our heads in the sand and hope these things go away, as many did at the dawn of the web era. And just as the web, and computers in general, have produced many negative externalities, they've also given us great power that we use every day. Only by understanding and harnessing these tools can we not only remain relevant but also lean into the unique strengths that local news organizations do have. And this is what is required if we want to avoid the mistakes of the pivot-to-video era and avoid the hollowing out of news from AI-empowered content farms. So, let's delve into specific ways journalists and news organizations are using AI tools with Dan Grech. Practical AI use cases for journalists from AP Regan-Porter [00:05:27]:
Welcome back, Dan. You are the first repeat guest on the podcast.
Excellent. I'm not even a working journalist.
So last episode that you and I talked, we talked at a pretty high level about AI and the disruption it was gonna cause for many industries, but journalism specifically.
Yeah. So today is gonna be all about landing that plane and really giving folks examples of how it's being used in the field today.
And some practical tips if you're wanting to get started. So you've been working with businesses on AI for a while now. You've had a master class you did with the city of Miami and Dade County that I attended, which was a fabulous class, and you're doing fractional CMO and helping companies apply AI, But you've also kept a foot in the journalism world. You've been working with emerging journalists, ICFJ and others. And I believe just this weekend, you're going to Puerto Rico to talk with some journalists about it. Can tell us a little bit about what's on tap there?
Yeah. Thank you. I've been developing a real expertise in how to use AI for marketing and sales whether you're a business or a nonprofit or a media organization. And you know, because I have nearly two decades of experience having worked in print and broadcast. Naturally, I'm sharing some of this with folks in the journalism space as well. So International Center For Journalists, the ICFJ has brought me in as a trainer there, and I'm gonna be speaking about AI as a journalism tool to student journalists at the University of Puerto Rico in San Juan. We're going to be giving a really nice set of examples, some of which we'll actually cover today, about how to think about and use AI as a working journalist, whether you are the kind of reporter, entry level doing the groundwork of journalism or whether you're a newsroom leader who's trying to figure out how can I use this effectively, responsibly, ethically inside of my newsroom.
And so I think a good jumping off point, you and I have been talking about practical uses of AI in local news several times, actually, in the past couple weeks, but you just got off a call that AP did on real-world use cases for AI. I was listening in while I was buried in a spreadsheet, so only half paying attention. But, I think you got a lot out of that. So why don't you kind of summarize, tell us about some of the practical examples you got out of that, and then maybe we can end with some takeaways that you picked up from that.
Yeah. The AP has a project that every working journalist should be paying a lot of attention to called Local News AI. And if you just Google Associated Press Local News AI, they have a really wonderful website where they describe what I'm about to share with you. It's funded by the Knight Foundation, and it includes work with Northwestern University, Medill Night Lab, and Professor Jeremy Gilbert. And pretty much all 5 of the examples that I'm gonna run through with you are all trying to use intelligent automation or AI-powered automation to save journalists time or to solve kind of a thorny problem within the newsroom and all of them are intended to be scalable solutions. In other words, solutions that while they're testing them in one local newsroom, whether it's a local TV station, a local newspaper, local radio station, they're common problems, or challenges or time consuming tasks for any local newsroom.
Use Case 1: Processing police reports
And, I'm gonna actually start with the one that I liked the best, which is from the Brainerd Dispatch in Minnesota and because it actually is something I used to do. So when I was in group, a starting journalist in 2000 at the Miami Herald, hired to cover the local neighbors section, one of my jobs every week, twice a week I should say, was to go to the local police station for the 3 communities I covered and look at the police blotter reports.
And I would, I mean, back in the day, I would hand write what was in the broader reports because they did not allow me to make photocopies or they charged me, you know, $5 per photocopy and, you know, most of the reports were completely irrelevant, but some of them were really significant and I utterly hated this task and I was required to do it because it was the number one most read part of our local paper. So anyway, I did this for many, many months for several years actually and I remember I went to and I had to fill in for another local reporter named Drager Martinez. Drager, if you hear this, please reach out. It's been many years. And poor Drager had been doing this for 5 years and every time you go there, there is like a little log of his name every time he visited and I literally had to sift through 25 pages of Drager Martinez, like nobody else had been there in in 5 years but him and then put my name at the bottom of that list. And it just It really made me hate being a journalist. It really made me not want to have a career in this field and The Brainerd Dispatch is testing a system where they automate this process. So the first thing it does, is it ingests from the city, the public safety incidents.
And then one of the really cool things that it does is it, you know, these are PDFs that are being distributed on, you know, these are public records, so they're being distributed on the police's website. So they take the PDF, scrape the data, input it into a spreadsheet, you know, or a database. And then what's really cool is they developed an algorithm to rank what is most likely gonna be something that you use, that you would want to publish. So there were certain keywords and certain news judgments that you would make that they then programmed into the AI and so the AI would tell you these are the most likely versus least likely, you know, cat in the tree, ignore violent crime definitely mentioned and then it would summarize the incident in a little paragraph and then input it into the website CMS for a reporter to then review and published. So there were, the reporter stepped-in in 2 places. The first is anything that got put into the CMS, the reporter put a checkbox next to. So even though the AI was like ingesting the information and putting it on a spreadsheet, they still had to check yes to import it. And then once it was imported, those summaries were edited before they were published.
So that was one of my favorite examples, really concrete, really clear use case that, you know, the public loves it, but saving hours and hours of time that then that reporter could use. The reporter talked about how he could then invest that time in going deeper into some of these incidents. So rather than just like going and doing it manually like I used to do, If there was something really interesting, he'd make that second call. He would report it out, and then they would write a bigger story.
Lesson: AI can free up time to focus on key tasks (but automation is difficult to maintain)
And that statement right there is really one of the takeaways, I think, from that because we saw that in several examples of how reporters and editors were using AI to free up time to do the more impactful, the more important, the more analytical, jobs that the journalists could do. Does that summarize one of your takeaways?
Yeah. You know, it's a time saving tool. It's very difficult. One of the big takeaways for me is that it's very difficult to build these automations because not only do you have to build the automation, you know, with all the integrations with various systems and you know, AI doing a very variety of micro tasks that add up to a larger task, but then it breaks, Right? Like if the police department updates their website in a small way, it could break the API that's pulling in the data. And so this whole concept of what they call dev ops or or developer operations, development operations. You know, you have to not only fund the creation of it, but the maintenance of it. And they said that many of these things, even though they're built well, were breaking on an almost daily basis, in small ways, and you have to have somebody who's staffed to do that. So it's expensive and time consuming to build it.
But once it's built, it allows—this is not intended in any of these cases to replace a reporter, rather it's intended to do an essential but repeatable task that newspapers and media outlets, local media outlets are expected to do and do those more efficiently to free up the human capital of reporters to do higher value tasks. And one of the things that is the highest value task of a journalist is human-to-human intel, interviewing human beings and getting information. So if it's written down in a police report, great, you know, let's let AI handle it, but you know those police reports if you ever read a police reporter jargony and terribly written and miss a lot of important detail that brings the story to life, so that's where the reporting comes in and that's a good use of a reporter's time is to do that reporting to bring the story to life and to add the context that no blotter, police blotter item will ever have. But might as well like not have, you know, me spend years of my life, you know, manually writing down these items. That's not a high use of my time and that would then free me up to do higher value tasks. And that's really, I think, the spirit behind all of these projects.
Use Case 2: Automated translation of weather reports
What other examples stuck out to you?
You know, I'm gonna be going to Puerto Rico and I live in South Florida in the tropical climate. I live in Miami. To the one that another one that really jumped at me was, El Vocero in Puerto Rico, which is, you know, one of their main Spanish language publications. They have a problem which is every 3 hours, you know, they're in the middle of like we are in Miami, you know, the hurricane zone. And so the National Weather Service and National Hurricane Center reports need to be published and potentially news alerts need to be sent to people and this can save lives. Right? This notice of a couple hours can make a real difference between a life saved and not. So the problem is that these alerts are written in English and then the National Weather Service has a Puerto Rico office with translators, but they don't translate the alerts right away. So what was happening is every 3 hours as they would issue a set of alerts in English and then basically the translators got to it when they got to it.
And so what happened is the paper was just sitting there on the website, refreshing the page. When the new alerts came in, they just manually translated it themselves. And so this occupied every 3 hours, you know, half an hour someone's time, they would then send out an alert manually. So what they did is they created a system to ingest these, translate them, and then draft an alert that then the reporter could just hit send on. Very interesting, all sorts of problems. So one of the things that I found really interesting, you know, as a Spanish speaker is that it was translating it into first of all, Spanish is very flowery and, we say things in Spanish differently than we do in English and so, a lot of times the translations were just non idiomatic and then a lot of times they were using the Spain Spanish idiom and Spain Spanish is as different from mainland, you know, Latin American Spanish, which is what most of these folks in Puerto Rico speak, as, you know, British English is to American English. And so, it was just like it was accented, if you will, with Spanish words, Spain Spanish words that just didn't resonate with the audience. All that had to be fixed and, you know, but in the end, it's a really powerful example of the potential life saving benefits.
It also, you know, just saved a lot of time, and made a much more efficient process than having to do it all manually. So that was another one that I thought has a lot of, you know, usability. One of the interesting things is they had two sources of data, National Weather Service and the National Hurricane Center, and the data from the National Hurricane Center wasn't structured in a way that allowed it to be used for this project. So they basically had to only do National Weather Service. So they're now working with the National Hurricane Center asking them to structure the data better so that it's easier for them to use. So that was another one that I thought was really cool.
Use Case 3: Transcripts of public meetings
Grech [00:18:39]: Another one that is, probably got to me the biggest impact potentially for every community in this country is creating transcripts from public meetings and then building summaries of those. Now, why do I think this is so important? Because what has happened in local media is as local communities have, as local news organizations have had to cut back, one of the first places they're cutting back is coverage of town hall of the little municipalities and if you can create a way to, but most of those town hall meetings are recorded and those recordings are publicly available. So if you could create a system like they did in Michigan, Michigan Radio, WUOM FM, to ingest those videos, get transcriptions of them, get summaries of them and then publish those summaries. You can have an incredible public service and it's so scalable. And you know, they talk sometimes about news brownouts or blackouts, and the idea of a news brownout or blackout is that there's no one in city hall. So to me, this is like at the core of local news. And just really quickly, a couple things that really stood out to me. Number one, they didn't even get to do the summaries because the transcripts were really bad. Now these are multi year projects.
I think these are most of these projects have been going on for 2 years. So a lot of the technology is pre chat, pre GPT four. So they were using mostly GPT 3.5 and the transcripts that they were getting, I think they were using a Google cloud translation service and then they had to switch because it was 50% mistake rate, on the transcript. So the transcripts are really problematic. They were able to get that down, to I think less than 20% and then once you have a decent transcript, then they could have started summarizing it, but it took 2 years just to figure out that piece and so they never got to the summarization piece. So, You know, it's still very early days and a lot of major hurdles, but you can imagine the kind of service that that would be to be able to have those public, meetings, transcribed and then summarized. And one of the challenges, of course, is when you do get to the summarization phase, which they haven't gotten to. What's the important information that you should put at the top of the summary? How do you when you have a 5 hour meeting, how do you determine, you know, what is important. It's that's a tough task for AI.
And transcriptions as well as translation is one of the areas that I've been plugged into for a while, and the field has advanced so much and so quickly in the last few years. So, you know, I mean, we've had a tool like Dragon Dictation for a while and, you know, everybody knows how crappy sometimes Siri could be with translation. And as someone who did a lot of long form pieces, you know, I was always looking for the holy grail of something that would give me a good transcription. You know, IBM was an early mover. Trent came along, and it was okay. That was one of the tools they talked about. But it has gotten so good lately. And there are so many tools.
You can plug them into, you know, Zoom now with Otter.ai, although I think Fathom's a little better. And so the applications are just across the board. And to be able to you know, since the pandemic, especially, more and more city council, school boards, and others are putting their meeting online and to be able to quickly transcribe that, get the automated summary, and then Have reporters go in there and see what's worth digging into, is this gonna be a huge service, for the field.
Completely agree with what you said. Like, the technology is moving really fast and you can see that even over the 2 years that these projects have been worked on, The technology was really rudimentary even 2 years ago. Like, the speed with which these improvements are coming is breathtaking. I know Open AI recently announced enhancements to Chat GPT to allow it to recognize language, speak language. Spotify is talking about doing translation now of podcasts. So like Dax Shepard, you can listen to him in Spanish and it's with his voice. So the, you know, the speed with which genovation is happening. It's breathtaking. You know, ChatGPT also mentioned image recognition, is finally being incorporated into their pro accounts.
Use Case 4: Classifying press release emails for assignment
I wanted to talk about one other use case from the Associated Press real quick, which is a news leader use case. So most of the use cases that we'll probably end up talking about and that the AP was piloting are really replacing tasks that a reporter does. But what I found really interesting was their most ambitious project, which, was a project for a, what's known as an assignment editor. So, you know, back in my Miami Herald days, I would walk in every morning and there was Mindy Marquez, and she was sitting, you know, in the most prominent seat in the newsroom and she was the assignment editor. And Mindy would basically say, Dan, go cover this. And then, I would run out and cover it. And, you know, whether you're in a newspaper or in a local TV station, the assignment editor is like a really, authoritative and hallowed person and that poor person is just being inundated by pitches.
Hundreds and hundreds of pitches and they're overwhelmed constantly. And so very important things, it's like a rushing river and, you know, fisher swimming by all the time. And so could we create a system to go through email and flag the important ones and automatically build them into the coverage calendar. So for instance, the mayor's office releases a press to really, you know, a, press release about an upcoming press conference. You know, we don't need to read that. We know we're gonna go, so let's just go ahead and populate and staff that. So they built they tried to build a system, to do just that from, WFMZ TV, a television station in Pennsylvania and it was really difficult. In the end, the system was only able to classify 20% of the emails that came in and the other 80% still have to be manually reviewed but they didn't save 1 hour a day on a task that takes 6 hours a day.
So instead of it taking 6 hours, now it takes 5. So on some level, it's a failure, right? Like not a failure, but it's like a mitigate, you know, qualified success is only 20% are they able to interpret. But on the other hand, you're saving that person 5 hours a week. So, it was an interesting one, but anybody's ever been in an assignment editor role, you know, discerning the news and then programming. So what's interesting also about that one is Every news station has every news local news outlet has different ways to determine if something is newsworthy and that's actually part of their secret sauce. And so in this case, they're not gonna share that algorithm that they built to Herman if something was newsworthy because they consider that proprietary. So the system can work, but each of the news outlets are gonna have to program it themselves and the way they do it is they went through thousands of emails and said, Yes, newsworthy, not newsworthy. They built a neural network and and and and and that's how they were able to figure that out. Very, very interesting.
The importance of “human in the loop”
One of the things that stood out for me, was also both in that example and in a couple of the others, they made a point of saying, look, this is not a set it and forget it. This is not let the AI run. That there's this real interaction with the human and a human verifying and checking and being a part of the process. So I think that's an important thing to keep in mind as we use all, you know, as we dig into all of these tools.
Yeah. The term they call it is human in the loop. Human in the loop. So the idea of automation is that the computer does automated things and this is this is these are all examples of AI enhanced automation. So automation, is really about data being transitioned and transferred into different formats, intelligent automation or AI enhanced automation is leveraging AI to make that stuff more valuable and what they said is that every key step in the process needed a human in the loop and every mistake that was published was because a human wasn't in the loop. And so This goes back to the main theme that I talked about in my training which is it's not about humans getting replaced by machines. It's about the human AI partnership. We have to become cyborgs.
We have to become able to work with the machines to do more work better. But what's really interesting is they definitely could not, even if they had tried Get rid of humans even if it's to debug the software or to edit the output. And one of the things I have said and I think is gonna be more and more true is we're going from writers to editors. We're going from coders to debuggers.
The importance of working on *how* you work as a journalist
Were there any other takeaways that you got out of that?
This is gonna be a little bit wonky, but here goes. There is a way you build software that's called Scrum, and Scrum is about short sprints and building like little mini deliverables and then iterating on it. And the motto of Scrum is twice the work in half the time and the Scrum methodology, what's known as an agile methodology, has become the way that all software is built, and it replaced what used to be called waterfall, where you would work in a closet for years and then boom, release the software. In this case, you're releasing something every week or every other week. And one of the things I saw, I'm Scrum certified, I use it in the marketing consulting that I do and one of the biggest ‘s from Scrum is that they spend an equal amount of time not doing the work, but optimizing how the work is getting done. Right? So in other words, It's not about the work. The magic of Scrum is about the conversations about how the work is done that actually lead to levels of efficiency. And it occurred to me listening to this that it is very, very difficult for busy journalists in demanding newsrooms to take the time to think about and work on how they do the work.
They spend most of their time doing the work and it fills up every second of their day. And so they were talking, some of the leaders of this project were talking about how difficult it was to find the time to work on these projects because these all are projects about how the work is getting done and they're not actually doing the work. This is a huge problem for journalists because what's happening is as social media is putting more demands on us publishing to multiple platforms and more quickly as copy editors and paginators are being let go and now reporters have to write headlines and cut lines and maybe even take their own photos and video. The doing of the work is filling more we're being asked to do more and it's filling an increasingly large amount of our time and brain space and it wasn't like we were not doing a lot back in the day. And so this I don't think almost any journalist really has time to think about how the work is getting done and how to do it more efficiently, which is very common in business and in software. Like, we're thinking all the time about how to do things more efficiently. So I think we're in trouble as an industry because we need to develop a discipline that is very foreign to us where it's not about the articles published, but the work behind getting those articles published and how to do that more efficiently. And at the heart of AI enhanced automation and everything the Associated Press is working on in its local news initiative is about getting the work done more efficiently and it's very foreign and difficult for journalists to do. And I I think to me, that is a risky thing for the industry.
And hopefully, that leads to more impact, and not just more efficiency. So I want I wanna dig into this just a little bit because, as you know, I have a software background and went from Waterfall to Agile, also Scrum certified. And, you know, I know you've applied this in your own business, but one of the things that is really powerful about some of these agile methodologies is It's not just about efficiency and getting more done. It's also about effectiveness, which in journalism would be about impact. So a couple a couple ways that manifest itself in a Scrum the Scrum world is you have these cycles, so you have a rhythm. It's not sort of a constant pace. You take time to to debrief, to look back at what worked and what didn't every two weeks, a typical cycle. You take time to plan, and you have a limit to your work in progress.
Impact, not just efficiency
So you're not just trying to do more and more and more. You're trying to get at a sustainable pace and get better at it. And you're always measuring yourself, in a software environment by is this meeting the client's need? Is this resulting in growth? And journalism, we you know, you translate that to impact. And I think as we think about what AI is going to enable and the work it's going to save, we want that to be more of the paradigm and not what email did for business, which took a lot of friction out, made communication more efficient, but it didn't make our life better, easier, or even that much more effective. It just genned up the rat race. And so I think if AI could go either way. So if we're using it smartly, and evaluating what we're doing, I think it has the real potential to let us have more impact and more sustainable lives as journalists. And if we don't, it'll just be like email where you've just got more and more and more to do.
You're—you know, Tim, you're pointing to something really profound, which is It really matters a lot what you're automating and where you're looking to find time savings. What I love about these projects is every single one of them, we're doing a mission critical or life saving summarization, or automation and things that all of them consider essential to serving their public and that they know their public wants. So whether it's alerts about upcoming hurricanes or summarization of the local town hall meetings or, police blotter items. All of these are essential duties of local journalists and local media outlets and all of them have been endangered by the industry struggles and the cutting of staff and the lack of newsroom people. So when you're thinking about automation in your local news outlet, make sure you're automating the stuff that will make a huge difference to your audience and any repetitive tasks that are high value to your audience, but don't really require human intelligence or, you know, human-to-human intelligence or data gathering or really a lot of discernment from the reporter. There are always in every job, some mundane tasks. And so the more that you can find high value to your audience, but low value provided by the reporter tasks, that's a good task that's eligible for automation.
Use cases for existing off-the-shelf tools: writing support
So I would like to move into to some some different use cases. So what the AP has It's been investing in helping newsrooms with practical uses of AI for a while. And, the webinar they just gave highlighted examples of particularly automation, right, and custom development, that they helped facilitate. But there are, of course, a lot of other use cases, and OpenAI has Given some funding to the American Journalism Project to develop some new ideas. They've paid AP actually for their content, which is great. But there are a lot of use cases that don't require custom development. The tools that are out there and and maybe more importantly, the tools that will be coming can provide a lot of value. So I'd like to spend a minute just kind of talking about those other use cases and those other tools.
On the business side, there's some obvious things that businesses are doing right now that you've taught people about. So there's, you know, from an audience development, audience growth standpoint, there's all kinds of optimizations and tools around SEO, SEM, tweaking, advertising, delivering ads, to sales research. We had someone speak at our conference about how he's using just, you know, ChatGPT to help him prepare for sales calls and getting ideas for what the client might need and who might think about that maybe wasn't, you know, obvious to him to begin with. And I think a lot of that could apply to reporting. So, if you're if you're working on a story, ChatGPT can be a partner. It can be like an intern or an assistant to help you think through some things. And and and I know that's a little bit of how you talk about with marketing professionals and how they can use these tools. And I think there's some obvious applications to a reporter, going in to approach a story.
Yeah. I think When you're thinking about how to use automation, whether you're a reporter or a newsroom leader, the AP examples are going to be out of reach for most news organizations unless you have pretty sophisticated support systems and developers and so forth. But there's like a million ways that you as a reporter can use this in very simple ways using free or very low cost tools like ChatGPT and there were a couple that I thought were really exciting. I think one of the cool ways to use it and I don't know how ethical this is, but is to take your raw notes from like a phone conversation or a series of phone conversations and have ChatGPT write it into, Associated Press style inverted pyramid article or to take the lead from what you need to delete is the first paragraph or a couple of paragraphs of your article and rewrite it. Or to take language that maybe you borrowed from another publication and you don't want to use the same words because that would be a copyright infringement and just have it Rewrite those words to mean the same thing using different language. All of those are ways to enhance the quality of your copy, and to get you at least a rough draft of an article. Another example is if you are a foreign-born person working as a journalist in the United States and your English, written English, isn't that fluent, you can use this to help clean up your language. And I know a lot of, you know, foreign PhDs have benefited enormously from being able to use ChatGPT to help them write research papers.
So it can help level the playing field a little bit. When I was a journalist, a lot of journalists were good journalists and not good writers. It seems to me that if you use this tool effectively, you to become a better writer. You're not gonna become, you know, the beautiful florid style of, you know, a Rick Bragg, but at least you can get your, upgrade your writing to a better level. So from a writing perspective, this could be an incredibly powerful tool. Just Be really, really careful that the data, that the stories you're writing are accurate to what actually happened. But if you did the reporting, Should be very easy to see areas where there's hallucination or making things up and and to correct for those. So that's one area that I'm really excited about that is, I think has an immediate use for many journalists.
Absolutely. And I wanna categorize some some ways to maybe think about the different ways you to use it. But maybe first, let's talk a little bit about approaches to the tools that are out there, in particular, ChatGPT. So I I gave a talk to a group of of business people, gave them several suggestions. So first, like, get the paid version. It's $20 a month, And it is so much better than 3.5. In a roundtable that you facilitated in Colorado, we saw that live where we Ask a question. We were in the 3.5 by default, and it was very stilted and robotic sounding.
And then we did the same question in 4, and it had a lot more style and fluidity to it. So definitely, you know, upgrade, and don't use it like Google. This is, treat it like an intern or have a conversation, ask advice, have a back and forth, give it some goals and some structure. Anything you would add to that?
Use cases for existing off-the-shelf tools: interview prep
Yeah. Well, you know, to your point, I'm really excited about the idea of using this to help with interview prep. So, I'd love to chat a little bit about that. So, You know, one of the hardest things when you're preparing for an interview is to think up creative questions. And so you could feed into Chai GPT or a tool like it background information that you gathered about your interview subject. You could even feed into them a list of the questions you've thought up and then you could invite them to think up other questions or creative questions or questions you hadn't thought of or questions nobody's thought of scheme. So brainstorming interview questions, I think is a very exciting use case. You could also say, ask JetGPT to take on the persona of the interview subject.
And as long as they're public, a public figure, they could actually conduct the interview with you. You could say, You are Taylor Swift. I am going to interview you now. Let's go. Where were you born? And you can actually practice interview using a chat, as well. I think one of the, you know, I used to teach interviewing as a skill and I called questions precision instruments. And so this is giving you more tools to get out of them information they've never shared before unexpected questions. You know, Terry Gross.
If you ever listen to a Terry Gross interview, her methodology is really simple. Terry is famous for preparing like mad for her interviews. And so she would read every book they'd written, you know, every interview they'd ever given. And then in the interview, If you listen, she'll summarize what they've said in the past and then ask the follow-up question that nobody had thought to ask. So ChatGPT can be like but you know, Terry Gross had no life like every day, you know, she would leave work and spend all night reading in preparation. So ChatGPT is really good at doing that kind of thing, reading stuff fast and ingesting it. So I could imagine a world where you could have like a question generator where you could say to it, what is a question that this celebrity or this person has never been asked before in any publication.
And, you know, embedded in some of what you're saying, there are some other advice for using something like a ChatGPT. One, you can give it a role. So you can say act as, you know, my subject or act as my editor or act as my financial adviser or my marketing guru and help me optimize this. You can, you know, ask it for a tone. So you tell it to be formal or casual or informative or persuasive or humorous. You can ask for a format, q and a, outline, bullet points, in the form of a script. Give it a purpose. Tell it who your audience is, and you can also, tell it how to respond in the style of or in a certain structure.
So, you know, those are just little practical tips as you're using it. Anything else you would add, you know, from your masterclass and just general tips on using a tool like ChatGPT.
Yeah. A couple little tidbits that I learned by listening to some prompt engineers Talk about how they what they've learned. So one of the most surprising is if you tell ChatGPT to take a deep breath and then do a task. It will do it better than if you don't. Now why this works? Nobody really fully understands. We know humans do better when they take a deep breath, but if you're not getting a good response, you can try that. It's kind of a crazy one. Another one that's really interesting is if you want them to kind of build out like a step by step process because a lot of times, you know, human beings and and and robots skip steps when they're explaining things.
So you say if you tell it just start your answer with the word first, comma. It will do a better job, of giving you step by step instructions. So there's a lot of little tricks. One of the things that's a little bit challenging is as the models get better and they learn and learn, a lot of what worked, works now or really helps it now, will will not be needed in the future. So what these tricks of the trade of prompt engineering or writing effective prompts are gonna evolve with time. One other kind of really important point that I wanna make is if you give ChatGPT or a tool like it, the exact same information to work from and the exact same prompt, It will give you a different answer every time. And so one of the things that I highly recommend is the regenerate and let it give you versions. Even if you like a version, you might as well go ahead and regenerate and get a 2nd or a 3rd.
It might even ask you which version do you like better. This is a really important tip because you shouldn't ever if it's really important, you shouldn't just be satisfied with the first. You can do it for the 2nd. So you can send them Like, if their response is off and you weren't specific, you can refine the query. But if your query was good and the response is still off, rather than giving up, just regenerate and the 2nd response might be better than the first. I do this all the time now, and what you'll See, when you hit regenerate in ChatGPT, it'll just say 1 of 2, 2 of 2, you know, and then you can cycle through the versions.
Data analysis and other miscellaneous uses
On the few minutes we got left, I just wanna tease people with some other ideas that some of which you and I have discussed and then see if you wanna add anything. So we talked a little little bit about using it for research and preparation for a story or an interview, and that's where having a conversation with it could be really helpful. Content creation and editing, I really love your idea of just throwing notes and see what you get back. Transcribing interviews, we've talked about AI assisted writing, copy editing. One of the areas that I think has the most promise is in data and document analysis. So whether you're trying to do something as simple as manipulate a spreadsheet and do pivot tables and you need some help or some new, maybe some new ideas for things to notice, or you're gonna throw PDFs at it, and have it do some crunching. I think you can really supercharge that. Even things like have it generate freedom of information request for you and do that in bulk could really be helpful.
Translation, we've talked about that. And then just in in the more conversational pattern with ChatGPT, I think you could really use it to dig into, find patterns, spot malfeasance, combat disinformation, and, you know, I think it'll become a real tool in some of the automations too and particularly around mis- and disinformation. And then you could also use it to build chatbots, for your audience. We were, I think I've talked told you that we're looking at building one for the Freedom of Information Coalition, where we'll upload it, best practices documents, maybe some court cases, and people can can go in there and chat with here's the response I'm getting from this government entity. What are my rights? How should I respond? What else can I do? Are they right about this matter of law? And it's you know, not all it's gonna be a 100% accurate, but I think it's gonna be a lot closer than whatever we're however, we're doing it now for the most part.
Yeah. One other really huge opportunity is computer assisted reporting using Bing spreadsheets and data. So one of the things that really held a lot of us back from doing data, database reporting, as computer assisted reporting is that we don't know SQL or we don't know how to query the data. And we are now at a place where you can use natural language query. You don't need to use SQL or another programming language to ask the spreadsheet for insights and information. You can, there's actually a function inside of ChatGPT. It's a lab function where you can upload a CSV of data, and it does advanced data analysis and you can even say so you can ask it like, okay, you know, here's a list of data, you know, please let me know, you know, what the average, salary for this group of employees was, but you can also say you are an HR manager and you're doing it, preparing a presentation to your boss about the data set here. What queries would you recommend that we make of the data and then please present to me that data and then visualize it and they can do that.
It's quite extraordinary. What this does is it allows all of us to become data driven reporters and computer assisted reporters. And You know, back in the day, there were entire conferences, Nitecar and, you know, entire dedicated people who were the like computer assisted reporter and they were always oversubscribed, they always had a long list of people they were waiting for to help. Now you can do that yourself and there's really very little stopping you. You do not need to be able to code. All you just need is a structured dataset and the CSV file, and you can do this yourself. And this kind of functionality is gonna soon be built into Excel and Google Sheets through Google's Duet and Microsoft's Copilot. So you are already even gonna need to, go outside of the spreadsheet themselves.
The spreadsheets themselves will become natural language queryable. And that is a game changer for computer assisted reporting.
One other thing I wanna mention is, I know we're out of time, but really quickly, The other thing that's been really, difficult is how do we use image generating software like DALL-E or Midjourney and my recommendation there is it's I don't think it's going to be appropriate to use it to create photo realistic images, but there's no reason you can't use DALL-E 3, which just got released, or, Midjourney to create cartoon or illustrations. And so I think another opportunity that journalists could do is to use it as an illustrator and then touch it up in Photoshop.
Well, thank you, Dan, for your time. Hopefully, this has prompted a lot of ideas for people, and, I look forward to—no pun intended—I look forward to talking to you again.
Thank you so much. Really appreciate it. Hopefully, I'll be your first three time guest as well.
Thank you for listening to the Local News Matters podcast, and thanks to Dan for more of your time and thoughtfulness.
A final thanks to our production partners at Pirate Audio. Pirate Audio believes every community deserves great things to listen to, and they are on a mission to help local newsrooms reach their communities via audio. If you're interested in starting a podcast or need production support for an existing one, let me know and I'd be happy to connect you.
If you like this episode and what we're doing more generally, please follow in your favorite podcast app, leave us a five star review on Apple podcasts and tell your friends about us. I appreciate the support. You can find past episodes, full transcripts and relevant links, and sign up for our newsletter at localnewsmatterspodcast.com or for lazy typists like me at lnmpod.com. You can also follow us on most social media channels @lnmpod.
If you have recommendations of others doing interesting and innovative work in local news, let me know through the contact form at the website lnmpod.com.
Past guests on the Local News Matters podcast include: Sara Lomax and S. Mitra Kalita (URL Media), Elizabeth Hansen Shapiro (National Trust for Local News), Mike Rispoli and Richard Young (via When the People Decide), Sarabeth Berman (American Journalism Project), Rabbi Hillel Goldberg and Shana Goldberg (Intermountain Jewish News), Lyndsay C. Green (via The Journalism Salute), Rashad Mahmood and Mark Glaser (New Mexico Local News Fund), Christian Vanek and Barbara Hardt (The Mountain-Ear), Dan Grech (BizHack), Zack Richner (Easy Tax Credits), Tracie Powell (Pivot Fund), Dan Oshinsky (Inbox Collective), Linda Shapley (via What Works), Yehong Zhu and Jake Seaton (Zette, Column), Charity Huff (January Spring), Joaquin Alvarado and Dave Perry (Aurora Sentinel), Steve Waldman (Rebuild Local News), Maritza Félix (Conecta Arizona), Michael Bolden (American Press Institute), Jeff Roberts and Corey Hutchins (CFOIC, Colorado College), Eve Pearlman and Erica Anderson (Spaceship Media), Jennifer Brandel (Hearken, Democracy SOS), Corey Hutchins with Bay Edwards, Todd Chamberlain and Raleigh Burleigh (Sopris Sun).