Exploration #103

Hate Longscrolling? Jump Around!

• If You Click Only One…

• Things to Think About…

• Things to Know About…

• And finally…

Hi all. After taking a week away from the tech news to focus on curating a collection of SXSW 2024 sessions, we’re back to a more familiar format. Catching up on the AI-related news is the main focus this week, with stories on ChatGPT, Adobe, and Midjourney. And I’ve got a couple of think pieces this week that might impact how you’re approaching AI policy at your organization. And there’s even an ATSC 3.0 piece making a ‘Where’s Waldo’ appearance amidst all the AI. But first…

Thanks to everyone who joined us for the Public Media Innovators’ 3rd Thursday webinar last week. If you missed any part of our Innovation at SXSW session, you can catch it, along with a transcript and the chat that accompanied the session, on the PMI page. Our next webinar, “PBS KIDS: Innovating Accessibility in Children's Media,” is April 18 and you can register for that here.

AI Video Is Closer Than You Think

Coming out of SXSW, I’m thinking specifically about the trajectory of a few tech trends, and most immediate of these is the rise of generative video.

Last year, prognosticators (including me) had generative video pegged for a 2025 arrival, with occasional seismic shifts throughout 2024. OpenAI, as they are wont to do, up-ended that with Sora, which will allow us (once it is in wide-release) to make video snippets out of text. For the video-side of public media, this means we’re no longer talking about the abstraction of words, as we are with ChatGPT. Soon, you’ll be able to create video with AI tools that will emerge from the model ready-to-use in your production.

Of course, public media’s hard-earned trust amongst the American public is based in part on the use of journalistic principles for the conveyance of truth. So it might be tempting to write this off as ‘not our problem’ because - whether for journalism or sports coverage (of which we do a lot in Nebraska) - AI tools will be limited to editing and finishing software that can streamline workflows.

But for a certain type of public media content, there will be either measurable savings, or a measurable increase in service. Need historical re-enactments? The scope and scale can be at the level of a Hollywood or AAA game studio. Want to make a limited run animated series for kids based on established local curriculum? Not only will that soon be within your grasp, but it’ll be within the grasp of local educators as well.

As generative video technologies reshape the creative landscape, we in public. media are facing both an opportunity and a responsibility. We have the opportunity to engage in innovative forms of storytelling using cutting edge tools, and we have the willing talent in our ranks to embrace these opportunities. At the same time, we have to uphold our ethical standards in content creation. This isn’t the first time we’ve balanced the excitement of innovation with the imperative to maintain public trust and uphold journalistic integrity. How we achieve that balance will likely define our role and relevance of public media in the generative AI era. Keep that in mind as you read through the pieces on AI video this week.

Okay, on to the links…

If You Click Only One…

A journalist’s journey from analog roots in radio to the horizons of AI (Davar Ardalan - Current) - Good advice here from Ardalan on how we all should be dipping our toe in the water of generative AI: "Public media enterprises considering the integration of AI should first document their workflows, archives and community interactions. The information gathered will allow for customized strategies for integrating AI. They should utilize internal resources and seek support from local universities and communities for AI task forces....Integrating AI in public media is a journey of intention and innovation. By involving diverse expertise and embracing collaborative learning, public media organizations can effectively harness AI to enhance storytelling and audience engagement, ensuring content remains relevant, inclusive and impactful....This is the moment to channel skepticism into constructive action."

Things to Think About…

How we're helping creators disclose altered or synthetic content (YouTube) - What I find most interesting here is how YouTube is choosing to draw the lines around synthetic media. "[W]e recognize that creators use generative AI in a variety of ways throughout the creation process. We won’t require creators to disclose if generative AI was used for productivity, like generating scripts, content ideas, or automatic captions. We also won’t require creators to disclose when synthetic media is unrealistic and/or the changes are inconsequential." They go on to give more specific examples. Whether you are working on crafting a generative AI policy for your organization or just conducting a periodic review, read this and give it some thought first.

How should we value news used by AI? A checklist for publishers (Anya Schiffrin - Poynter) - Schiffrin offers up a thorough examination of the issues around whether to allow your organization's content to be used as training data for AI models, or to try and leverage that content into a value exchange with the companies behind those models. I'm still of the opinion that access to our data is part of our universal service mandate, but that takes nothing away from quality thinking here. H/t to Liz Maestri at PMVG for making me aware of this one.
—Related, from William James in Reuters: Britain's BBC considers building in-house AI model

GenAI is a Skills Amplifier, Not a Skills Democratizer (Shelly Palmer) - I'll let Palmer explain: "Generative AI is a skills amplifier, not a skills democratizer. Said differently, if you’re great at your job, proper use of a generative AI model will make you measurably more productive. If you suck at your job (or if you don’t know the subject you’re asking generative AI about), you’re still going suck." I'm a bit more optimistic that a less skilled but bright user can go from ‘novice’ to ‘average’ with gAI tools, but a pro using gAI still can't be beat.

Generative AI’s environmental costs are soaring — and mostly secret (Kate Crawford - Nature) - I continue to wrestle personally with how to address issues like these. AI-abstinence isn't a viable solution, so for now my solution has been to limit my use of gAI tools to professional tasks. But admittedly, when part of one's job is to test and experiment with these tools, that allows a lot of wiggle room.

It’s Time For A Reset On NextGen TV (John Lawson - TVNewsCheck) - I don't touch on ATSC 3.0 much here, but I know it's a strategic priority for many stations to find that product-market fit and some include it as part of their emerging media portfolio. From a content perspective, I still think it's a solution in search of a problem, so Lawson's assessment that a transition is stalled isn't surprising. But I’m also not sure content for consumers is necessarily the ‘killer-app’ here.

Things to Know About…

Which AI should I use? Superpowers and the State of Play (Ethan Mollick - One Useful Thing) - If you joined us for the February “3rd Thursday” webinar, you'll have heard guest Ethan Mollick say that, thanks to the GPT-4 model under the hood, ChatGPT Plus (or its alter ego, Microsoft Copilot) was really the only gAI chatbot tool you should be using. Just a month later and he's now endorsing three models (GPT-4, Gemini Advanced and Claude Opus). That's how fast things are moving in this space.
—Though, as Benj Edward reports in Ars Technica, GPT 5 could reset the playing soon: GPT-5 might arrive this summer as a “materially better” update to ChatGPT

Large Language Models from Scratch (Steve Seitz - Graphics in 5 Minutes via YouTube) - I haven't posted a large language model explainer here for a while. If you're new to LLMs, which power ChatGPT, Gemini and Claude, this video will be good for demystifying the magic of chatbots.

The Top 100 Gen AI Consumer Apps (Olivia Moore - a16z) - Given that Andreessen Horowitz is never shy talking their own book, you need to take this with a grain of salt. But they previously did this report last September and a lot has happened in six months, so I still found it was a good unofficial state-of-AI report.

AI Video: Generative Tools Guide (Evo Heyning via LinkedIn) - the OG subscribers to this newsletter may remember when Evo was our webinar guest back in November 2021. At that point we were talking about the metaverse, but like us she her interests and skills have evolved with the trends emerging media. This post is a good intro for anyone wanting to experiment with generative video.

Sora: First Impressions (OpenAI) - Open AI has apparently been conducting a charm offensive with creatives in Hollywood to help creatives in the film industry get more comfortable with Sora (beware geeks bearing gifts). As a result we're getting to see more videos made by creatives who are sanctioned by, but not employees of, OpenAI. Keep in mind we're not yet hearing how these videos were made (e.g. number of clips generated, prompts used, musicians used, time to edit, etc.), but this second round of clips helps expand the vision for how Sora (which hasn't been release and doesn't have a release date) could eventually be used in public media.
—And already, an open-sourced, Sora-like tool has been released. It’s not at the level of the carefully curated clips that OpenAI is releasing, but it shows the interest in the generative video space.

I Used ChatGPT as a Reporting Assistant. It Didn’t Go Well (Jon Keegan - The Markup) - However good a tool it is, ChatGPT isn't going to be good at everything, which makes published experiments like these valuable.

OpenAI’s chatbot store is filling up with spam (Kyle Wiggers - TechCrunch) - Not the level of attention you'd hope from company aspiring to bring about artificial general intelligence. Crafting your own GPTs can still be a good way to futureproof tools against evolving models. But caveat emptor with GPTs made by others. Though I’m not sure I’d pay for any GPT at this point, like any App Store, it's best to lean toward GPTs recommended by trusted sources.

Adobe Firefly repeats the same AI blunders as Google Gemini (Reed Albergotti - Semafor) - Though supposedly trained on very different sets of data, Firefly has started to show an inclination toward diversity where such diversit would be historically inaccurate. Adobe's response is that Firefly is a creative tool and isn't meant to generate historically accurate images, thus people are using it wrong. While that certainly a nuance that Google, the world's library, can't fall back on, it's still akin to Steve Jobs' "Just avoid holding it in that way" email.

Adobe finally brought Firefly AI into a mobile app, but it has some catching up to do (Jesus Diaz - Fast Company) - In other Adobe news, now you'll be able to creatively render inaccurate historical images on your mobile device.

MidJourney just announced a huge AI expansion — here's what you'll be able to do next (Ryan Morrison - Tom's Guide) - While video is the horizon now for generative AI, others are looking beyond that horizon to something even more comprehensive. What's described here feels a bit like a game engine and seems to work like a virtual set. So it's not hard to see how the Midjourney vision will push generative model and worldbuilding tools into an intersection with (and acquisition by?) game platforms. AI will likely kickstart the next hype cycle for the metaverse, and the metaverse will help give AI more of a 'Pixar moment.' This gives you a glimpse of one way that might come to pass.

A ChatGPT for Music Is Here. Inside Suno, the Startup Changing Everything (Brian Hiatt - Rolling Stone) - Were it me, I'd tweak the headline to say "a Midjourney for music." Play the embedded demo in the article and you'll hear why.

And finally…

Ruff Ruffman, Humble Media Genius (Bill Shribman - GBH/PBS Kids) - And finally, friend-of-the-newsletter Bill Shribman just released, via PBS Kids, new episodes in this media literacy series. According to Bill, "[t]he main focus is AI. We dig into generative text, generative art (with an homage to a classic Warner Bros cartoon, Duck Amuck), how algorithms work (with a Meatloaf flavor) and more." Great to share with young (or young at heart) AI users in your life.

Have a creative, productive week!