Hi there.

I’m writing this newsletter on my way back to Los Angeles from Silicon Valley, where I have spent the week hosting our a deep dive on AI by means of an INMA Silicon Valley Study Tour. I need time for the vast amount of information to settle — to pull together all the threads of what we have just learnt and its potential impact for media, to give some context we met with people at Google, Microsoft, Open AI, Stanford, AI startups, and prominent media. 

It was a lot (in a good way). 

Today, I want to share two things at either end of the spectrum of GenAI: practical tools we can use now and a consumer thinking trend that we should be cautious about. In my next newsletter, I’ll try to give a high-level view of the state of media and AI, what some of the big questions are, and where I would focus if I was running a news organisation.

If you have questions, or answers, you can always reach me at Jodie.hopperton@INMA.org.

Thanks, Jodie

Using GenAI in the storytelling process

When people think of GenAI in the newsroom, often the immediate response is, “We don’t want AI writing our articles!” Neither does the audience. Because if everyone does that, it’s a race to the bottom. But this is an exceptionally simplistic view of AI. You’ve gone from A to Z, missing the other 24 letters of the alphabet.

GenAI can, and should, be an assistant to journalists and the newsroom. Let’s give you some examples:

• Starting at A, what are you writing your article about? Maybe you are stuck. Use GenAI to inspire you. It’s excellent at idea generation and brainstorming. 

• Now you have a great idea. Before you get stuck into reporting, you probably want to check if it has already been covered. You can use search, or you can use GenAI. The advantage of GenAI here is that you can use follow-up prompts and questions to go further rather than opening up a bunch of links and reading through vast amounts of text., which is what currently happens through search. 

• Now, who knows about this topic? What sources could you use? I know an assistant who can make some suggestions to check out. See note above referencing search vs GenAI.

• You start the story and there are a number of data sources, all in different formats. It’s going to take a massive amount of time to go through and normalise. Unless you use AI to sort it for you (Google’s free Pinpoint is excellent at that, even for PDFs, which, as we heard, is where “data would otherwise go to die”). 

• Great story. Could it be better? Maybe it’s a bit depressing and you want to add some levity. Maybe you have 2,000 words where you need 800. Guess what you can ask? Yes, GenAI. Is it cheating? No. Would I suggest you ask GenAI and then publish? Absolutely not. Review, change, publish, It’s simply rearranging your words into a better format. An assistant editor if you will. 

• Excellent article, nice work. Now, what about a headline? Maybe you have some ideas. Maybe you don’t. Brainstorm it with your news assistant? 

• You have a great headline, maybe even two. You can use AI to A/B test which is better. Don’t be too upset if your human version doesn’t work. But feel free to be smug if you win.  

• It’s published, woo hoo. Where else can we distribute it? You probably spend a bit of time reformatting for different channels. Oh no wait, you don’t have to do that. Guess what can help you reformat one piece of contact into multiple channels? OK, you get no prizes now for guessing that GenAI can help with this.  

• Great article. Shame some people don’t have time to read it as they drive to work. I have an idea: Use a synthetic voice to turn it into an audio article. Better yet, ask the voice AI to adapt the text to be more conversational to make it easier to listen to. 

Congratulations. You are now at Z. Was it exhausting getting here? Or did the AI make you quicker and smarter? Please let me know. I’m at Jodie.hopeprton@inma.org.  

Reading this and wondering why it’s part of the Product Initiative? Because you lucky folk who get to figure out how this becomes part of the workflow. I’ve been told that the API integrations are fantastic. Is the phrase “it’s a simple API integration” the new “one line of code?” Again, please e-mail me at jodie.hopperton@inma.org to set me straight (or complain).

Date for the diary: See you in London! 

We’ve recently announced the line up for our World Congress in London next April. It’s sensational. As with all INMA events, it will be a mix of strategy and practical insights that can be used today. We’ll have some excellent product specific sessions, and I’ll organise some off-schedule drinks for smart media folk. I truly hope you can join us. Sign up here.

LLMs do data and information: We do knowledge

One of my aha moments in last week’s study tour was seeing one of the participants looking slightly shell shocked after a meeting with a Big Tech company early in the week. When I asked her what was wrong, she said: “They don’t care about news. They literally don’t even think about it.” 

She’s right. Guess what? We all live in our bubbles. Ours is news. We need to get out of that just as much as the tech companies do. 

Is it tech’s job to think about reliable information? Absolutely. What percentage of information that is generated/queried is news? Very little. Just check out this article on how search for news is decreasing. So Big Tech is unlikely to spend a lot of time thinking about one of the smaller slices of the pie.

As one participant told me, “I am surprised by the naivety, or maybe ignorance, of the platforms when it comes to news and the potential impact on democracy.” A Silicon Valley resident and AI expert was surprised by the surprise. And therein we have our bubbles. 

Let’s break this down a little. 

By nature, LLMs (large language models that train generative AI) need huge data sets. And by definition, news is not a large set. It’s recent information. What we do, or rather where we excel, is explaining and giving context to nuanced situations. In their words, they do data, we do knowledge.

So as GenAI develops, what does news look like? 

ChatGPT, the most prominent of the GenAI tools, doesn’t cover up-to-date information and is clear about this when it responds, often citing “as of 2022 in my last update in XXX.” Other platforms, as far as I am aware, are a mix of GenAI and search, therefore using citations and linking out to sources. 

Where consumers may be unclear is that the words returned are data. LLMs understand the likely sequencing of events and return the words it thinks best. 

In the same vein, GenAI doesn’t hallucinate. It’s not human. It’s code and the sequencing. It makes errors in its sequencing. We would do well not to anthropomorphize AI in our vocabulary — both internally and in our writing. It perpetuates the problem and doesn’t help people understand what GenAI actually is.

The big question we need to help the tech platforms figure out, or convince of the importance to figure out, is how to make sure reliable sources — not just the quickest sources — are surfaced. If I wanted to be dramatic, I would say that if speed is rewarded over accuracy, democracy is at stake.

This is not doom and gloom. It’s something we need to figure out. Ideally together with the technology companies that are at the forefront of this revolution. 

Lastly, to end on a little levity, I thought I’d ask ChatGPT about myself. If chat is the new search, I want to know what’s out there. Excellent response for the most part. And also a good demonstration that this is data led, not human led. 

About this newsletter 

Today’s newsletter is written by Jodie Hopperton, based in Los Angeles and lead for the INMA Product Initiative. Jodie will share research, case studies, and thought leadership on the topic of global news media product.

This newsletter is a public face of the Product Initiative by INMA, outlined here. E-mail Jodie at jodie.hopperton@inma.org with thoughts, suggestions, and questions. Sign up to our Slack channel.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .