Welcome to the first episode of AImpactful, where we delve into the evolving world of journalism in the age of artificial intelligence. In this enlightening discussion, we are joined by Joe Amditis, an expert in integrating AI into journalistic practices.

Joe shares his valuable insights on how journalists can effectively utilize AI tools like ChatGPT to enhance their work. He emphasizes the importance of experimenting with these tools, understanding their capabilities and limitations, and using them to streamline workflow and focus on impactful storytelling.

From offering practical advice on setting goals with AI tools to discussing the potential risks and ethical considerations, Joe provides a comprehensive overview for journalists stepping into the AI arena. He also shares his experiences with creating “The Beginner’s Prompt Handbook: ChatGPT for Local News Publishers,” revealing the motivations and aspirations behind this valuable resource.

This episode is not just for journalists but for anyone interested in the intersection of technology and media. Join us as we explore the exciting possibilities and challenges AI presents in modern journalism, and gain insights to navigate this new frontier more effectively.

Transcript of the AImpactful Vodcast

Branislava Lovre: Welcome to our very first episode of AImpactful. Today we’re focusing on the correlation between journalism and artificial intelligence. Our first guest is Joe Amditis.

We are hearing a lot about ChatGPT and AI tools these days. However, not all journalists have begun to use them. For those who are new to AI, what simple advice would you offer? And what should be their first step?

Joe Amditis: I think the most important first step is to just start playing around with it, you know, just look at it. Make a free account or play around with one of the ones that doesn’t have to be ChatGPT necessarily. Start playing around with it. Just start, start talking to it. Start trying to trick it, see what it can do. See what it can’t do, start to push the boundaries and see what it’s good at. And then you’ll start to understand some of the things that it might be good at helping you with. A lot of the folks who I talk with and I’ve worked with on this kind of stuff, start really small for instance. They take a story that they have written or a piece of writing that they’ve produced, and they put it into the bot and they ask for a one-paragraph summary and then they, because they’ve written that material, they know it intimately and they can go look through and make sure that it’s right. And then from there, you know, make me a list of five additional questions that readers might have after reading this story. You know, little things like that, just to sort of test the waters and see sort of how it performs with different tasks. I think that’s the best way. The best way is to just get started.

It is very early still. And so, you know, in all likelihood, within six months the tools will be different, they’ll be bigger, better, more evolved, more efficient. But the core aspect, the core relationship between that information and the tool that you’re using to edit it, I think is going to relatively remain the same. And so just like when people started to learn how to Google things, it’s kind of funny because back then in the early 2000s, everybody was learning how to Google and you had to frequently remind people, no, you don’t put a whole sentence in there. You don’t ask a whole question. You just use keywords. And now sort of we’ve wrapped around and now people have to learn like, no, you don’t just put the keywords in there, you just ask a question, just talk to it, you know, ask it what it needs from you to accomplish a task. Ask it for suggestions on things and just start to get a feel for how these back-and-forth works and what it might look like. You’ll immediately start to see possibilities for your own workflow and your own operations to see how you can maybe expand on those questions and prompts.

Branislava Lovre: For which purposes can we use AI tools?

Joe Amditis: Ask it to reformat your text to make it easier to read. You can do a variety of little like basic tasks, a little back-office stuff that you tend to find like most time-consuming or here and there, the sort of annoying stuff. I’m a lazy person at heart. I’m a lazy person who doesn’t want to, in the sense that I hate doing things that I know are bullshit and I know are boring. Maybe they might be necessary, but I’m always looking for ways to streamline those tasks, or make them, you know, put them over to the side so I can focus on doing the important work, the stuff that I get excited about, serving the community, making things, creating. And I see these tools and I have from the beginning as a way for my lazy mind to spend less time on the boring and repetitive stuff. Obviously assuming and ensuring that they’re accurately and successfully completed to just spend less time on doing that and more time doing the things that I love and that I wake up and go to work for every day.

Branislava Lovre: How can we set goals and discover which tools are most beneficial for us?

Joe Amditis: The goal is not to just use these things everywhere just because they exist or for the sake of using them because they’re neat. The goal is to find ways to make journalists more effective, more efficient, and to allow them more time to be out there reporting and serving their communities. The goal should not be to offload all of the important stuff so that you have more free time to do whatever you want. It should be to find ways to eliminate useless, extra extraneous busywork that you can relegate to these tools. And you know, something like Zapier, which as audit is an automation tool. There are tools out there that have sort of been doing that. There are companies and resources out there that have allowed journalists to do that for ages. This is just another one in that stack that you can find ways to incorporate into your workflows. And it’s pretty easy to integrate, especially with something like Zapier, which I use all the time, which basically just allows you to set up if and then responses. So, if something happens in this Google sheet, then do something like this. If something happens in this Airtable base, then perform this action. If I get a new email that I flag under this label, then apply this action and so when you start to see, we start to build out, you see, okay, it’s not just that I can ask these questions of these bots, but if I can get these responses tailored to that, they’re consistent and I can automate them or make them happen based on certain triggers or inciting events, then I can start to see where the productivity really comes in. And I maybe I don’t have to. Maybe I take a police blotter or, you know, police press releases and I have it give me bullet points of what the important things are and use that to develop more robust reporting instead of just simply running the press release from the police. There’s a range of opportunities there to use these tools to put the relevant information in front of you at the forefront so that you don’t have to spend time sifting through that to get to the nuggets, especially if, you know, after doing this so many times and after working with these materials for so long where the nuggets are, but it just you just happen to have to open six or seven windows, for instance, in order to get to those points. Those are opportunities. And, you know, areas where you can so you can set up these types of automation and bot responses. If you’re confident in their accuracy, to shave off those couple of minutes that you would have spent. And over time that’ll build up and that’ll, that’ll add up.

Branislava Lovre: One day, I was on LinkedIn and noticed that your guidebook had been widely shared. I believe, by now, most people have read it. Those who haven’t definitely should, as it’s full of useful tips. Could we highlight a few of those for ChatGPT prompts?

Joe Amditis: One of my favorites that I think is most useful and we found really, really interesting is when you publish a story, do a blog post, write about something, or publish a report. It often seems self-evident to those who worked on it or the journalists who produced it, that this is something people should care about. However, it may not be as self-evident to someone who isn’t as involved in that process. So, we like to have it, we’ll write a blog post or a medium post or whatever, and we’ll put the text into the bot automatically through Slack. It’ll come back, and I’ve asked it to tell me, give me five bullet points and explanations for why people should care about this. Make the case for why this community would care about this. Give me five possible questions that were not addressed in the original copy. That sort of augmenting and amplifying capability to sort of find blindspots that you might not have realized is really useful. It can help contextualize your work and put it in a light that you might not have initially understood or recognized off the bat. A lot of times it’s useless, a lot of times I already thought of this, or we’ve already thought of that in the process. But every once in a while, it’s like, “Oh wow, actually, we totally missed that. Let’s get back in there and add something so that we missed that.” You know, it’s stuff like that. It’s acting as a sort of limited coworker or assistant at a level that you would never expect it to be accurate or right the entire time.

I think that one of the most annoying things that I see is that when journalists or public speakers or whatever media critics, they talk about this and they equate it with a search engine. And I kind of did that earlier talking about how to Google things back in the day versus how you would ask these questions. But it’s so frustrating seeing people treat this or assume or talk about it as if this is some kind of factual database, knowledge database. A lot of the stuff that’s scraped in this database, it’s like Reddit comments and yes, Reddit comments are very useful when you know that it’s correct. But the majority of Reddit is just trash. It’s trained on how people speak to each other and it has information about the relationship between different words and concepts and how they relate to other words and concepts. Nothing in there is… It says that any of this stuff is going to be factual. It’s just most likely the next word or set of words. And so when you start to see it like that and you get out of this mindset of like, “Oh, I searched for something and it gave me the wrong answer.” It’s like, no sh*t. Of course it’s going to give you the wrong answer. It’s not a factual database. It is a words and language database. And people say a lot of false stuff all the time that doesn’t mean that it’s not valuable. It just means that that’s not why you go for your facts. And so I think if you can get out of that mindset and get more in a, hey, let me see what this might do. Let me see if there’s something here maybe I missed then you’re going to find a lot more uses for it. That would be my answer.

Branislava Lovre: As we mentioned, these tips can be found in your handbook. What led you to write “The Beginner’s Prompt Handbook: ChatGPT for Local News Publishers”?

Joe Amditis: So, I have been using ChatGPT for a while and had resisted actually paying for anything. There are great resources in repositories of information that people are putting together in good faith out there, like The AI Exchange by Rachel Woods, a fantastic resource with a community built around sharing information, tips, and guides. I didn’t want to pay for anything other than that, though I paid for that for a month just to see what it was like. But then I came across a prompt handbook that cost $2 for a PDF, and I thought, why not? As soon as I opened it, it was obvious it was written by one of the bots. It was lazy, worthless, and it might as well have been lorem ipsum text. So I realized I got scammed. I had a lot of stuff floating around in my head about this, and I wanted to learn Notion, so I decided to use this as a project to create my own prompt handbook. I focused on the core aspects of the language used and the relationship between you and these tools. I started dumping everything in my brain onto the screen, and it turned into this handbook, which keeps getting longer because I’m still adding to it. I try to keep it super beginner, but the most valuable aspect, honestly, is the appendix with all the additional links and resources. It’s just getting longer with real current and good stuff in there. That’s how it came about: I got scammed, got mad, and didn’t like it, so I decided to make an actually good one.

Branislava Lovre: I always find examples incredibly helpful. Could we walk through a specific example of how journalists can implement ChatGPT in their work?

Joe Amditis: I would start with really low-stakes examples and tasks, something that you would never put your name on, something that would never be published for a general consumer audience. For example, take messy notes or a transcript from a recent meeting and just clean them up. That’s where I started with it. I had transcripts or meeting minutes, and I’d like to just clean them up and make them look nice, reformat the text, give them some headings, use some markdown formatting. This tool can convert natural language into markdown formatting, which is very useful. Start messing around with some of the formatting and see what it can do. You’d be surprised how much of a benefit sending a well-formatted email to your colleagues can have. Just start, mess around with it, ask basic stuff, try to trick it, see what it can and can’t do, and go from there. Start with very low-risk environments where you don’t have to be perfect and then build out to things that are more complicated and maybe more consequential.

Branislava Lovre: Could you share one of the most successful or interesting ways ChatGPT has been used in a local newsroom?

Joe Amditis: I don’t know if it’ll be interesting to everyone, but for me, I really like its ability to go through a transcript of a town council meeting and pull-out quotes that it has identified as emotionally resonant or interesting or engaging, and then to be able to explain the news judgment that went into those choices. Using it as sort of a red team for your stories or for your coverage can be really useful. For example, give it the text of your story and say, “Tear this apart. Be very critical.” Using it alongside human editors can save some time and be really useful. It’s nice to see a different perspective, one where you don’t have to feel ashamed if it criticizes you, as it’s not your coworker telling you it’s bad but a bot. It might help with grant writing for local news organizations, but those are very high stakes and carry possible risks. Just like any text you get as a publisher or editor; you need to edit it and vet it to make sure it’s okay to put out under the name of your organization. You would do the same thing with a random freelancer that you hired. You wouldn’t just publish it because it’s from a human; you would edit it. In the same vein, you need to be even more critical with these bots. There’s only so much trust that can go into what this thing puts out, and you should have no trust in what it puts out. Always verify, even when it seems right. The last thing you want is to be one of the first or most public examples of a journalist who got lazy. Publishing bot-generated content that ends up being wrong or, God forbid, harmful, goes against every journalistic ethics principle that’s ever been written.

Branislava Lovre: What do you see as the biggest potential risks or dangers associated with the rise of AI in journalism and news production?

Joe Amditis: A couple of things concern me. These models that are trained now are proprietary, but there are also open-source ones. As soon as a group of right-wingers or fascists get their hands on a model they can train without the social and cultural guardrails that have been put on these models, it’s going to open the door for a lot of targeted, customized harassment of marginalized communities, immigrants, trans people, people of color, and others. The work and effort required to produce these kinds of materials is drastically reduced. Voice cloning is another concern. It’s already easy to fool people with scam calls, and voice cloning will make these scams more convincing. We’ll see an increase in these scams, and journalists need to understand how these tools can be misused to equip the public to protect themselves.

Branislava Lovre: It seems that it could be quite easy to be misled or manipulated in the world of AI. How can we overcome these challenges?

Joe Amditis: I don’t think we can overcome them, but we can prepare ourselves and our community members. Journalists need to understand how these tools are used and misused. They need to be tapped into the reality of the situation with this technology, not just the optimistic, tech-minded, utopian version. We need to understand how the laziest, the scummiest, and the most intelligent but nefariously inclined people might use them. For example, I was able to create a voice clone of myself and fool my mom. Scammers might use voice cloning to make more convincing scams, especially when language barriers are a factor. Journalists need to understand this and preemptively educate the public.

Branislava Lovre: What does the future hold for us?

Joe Amditis: Small newsrooms are best positioned to take advantage of AI from a human resources standpoint. My biggest fear with large news organizations like The New York Times is not the tool itself, but the profit motive driving their business and newsroom decisions. With layoffs and job losses, it’s capitalism that’s driving these, not AI. Smaller teams or independent journalists are less likely to lay themselves off in favor of bots, especially since bots cannot actually do the job of a journalist. We need to be wary of executives and CEOs who think these bots can replace humans. We don’t want journalism to become the self-checkout version of media. In small settings, AI can be great – it can make certain tasks faster and more efficient, allowing journalists to spend more time serving and engaging with their community.

Branislava Lovre: Thank you, Joe. I appreciate it. It was great talking to you.

Joe Amditis: Thank you so much. I appreciate it. It was great talking to you.

Branislava Lovre:  Thank you, everyone. I hope that we will meet in the next episode of AImpactful.