Did you miss the recent INMA Los Angeles Tech Innovation Study Tour? 

It was an intense five days of learning from news media and adjacent industries, a glimpse into the future that is coming at us, and an opportunity to reflect on how we can prepare ourselves for it — to say nothing of warm, insightful conversations with colleagues from around the world and a chance to enjoy the culture of golden southern California.

There was a lot to take in and digest. Here are a couple of GenAI-related topics I’m thinking about now that I am back in crisp, cool Canada.

Video: GenAI can change everything

We heard from many companies where the focus was visual content: KCRW, YouTube, Nota, Umi Games, Snap, Verizon, Nexstar, Runway, Channel 1, Media Monks, Aeon, and Fox, to name a few.

We learned about reaching younger audiences:

• The average screen time for Gen Z is seven hours a day. The top apps they use are TikTok, Instagram, and messaging. And TikTok is hot, but YouTube is huge, and it has a steady base of users because children are introduced to it while they are quite young and stay on the platform for life. 

• About 90% of users on TikTok do not follow a single news source, according to Nexstar.

• About 80% of the traffic on the Internet is video, according to Verizon. Nexstar found live stream is 80% of what users watch.

There were a couple of underlying AI threads that ran through most of our conversations:

Personalisation is the way forward.

For example:

• Verizon created a sports companion that chats with you about a game on TV and creates a highlights reel based on your favourite team and players.

• Nexstar’s connected TV product offers viewers content from local stations based on their location as well as video-on-demand clips that are algorithmically placed based on the user’s interests. 

• Using Channel 1’s tools, you can create a video stream that offers four versions of a story — you can pick the one that suits you best or you can ask the AI to pick one for you. You can chat with it as well. Over time, the model will learn from your engagement with the content. For example, if it sees that you often watch videos on a certain topic, it will learn to skip the background introduction at the beginning of the video.

GenAI makes video creation quick, easy, and cheap.

• You can create a lifelike video by simply describing in basic English what you would like to see, using a tool such as Runway’s or Nota’s, which will also create different versions for different platforms.

• Or you can provide a marketing brief, brand guidelines, and a feed of urls and ask a tool such as Aeon’s to create a video. It will write the script, find the best visuals from your asset library, and clip them. This is particularly useful for sites where inventory is constantly being updated or where there are thousands of pages of text on various topics (such as WebMD).

• Or you can use an agent, such as Channel 1’s, which takes a wire video, analyses footage from the video or a shot list, writes up every frame, then writes a story, uses a synthetic voice to read the script and then uses AI to edit the video. The only human interaction here, according to Channel 1, is the editor who casts an eye over the script and signs off on it. (The original wire story is also based on human journalism.)

• You can create a digital double of a human anchor or a purely GenAI anchor that speaks in several different languages around the clock to broadcast content.

• If you are producing content in a different format, such as text or audio, use a tool like Google’s NotebookLM to upload the show and understand its potential for video rather than listening to or reading every one of them.

Content will be everywhere.

How will our visual and audio offerings change when everything around us is a screen or a speaker? We’re talking about the change that will occur if products such as Ray-Ban Meta smartglasses become a household item, allowing people to record or livestream video to Instagram on the fly or join video calls through their glasses. We would no longer be creating content for a screen that sits on your desk or in the palm of your hand.

Date for the calendar: Wednesday, November 13

What have we learned about GenAI after an unprecedented year of discovery and best-practice sharing? Please join us for our free GenAI Town Hall to find out more.  

GenAI and Fox

We visited the Fox Studio Lot (home of The Simpsons!) and had the privilege of hearing from Melody Hildebrandt, chief technology officer at Fox Corp. Their AI use cases fall into three categories:

1. Create: Tell better stories. 

2. ⁠Curate: Meet people where they are, improve content packaging and discovery. 

3. ⁠Carry out: How can we improve execution of repetitive tasks?

Two key insights: Hildebrandt pointed out that for the last category, Fox found it most effective to use a bottom-up process, harnessing the energy and ideas of entrepreneurial employees, while the other two had been more top-down. 

The other distinction was that Fox decided to buy technology for the last category, while it preferred to build “highly custom” applications for the other two.

“Data is the new gold. We don’t want to undervalue it. The first two are highly strategic, so we are building,” she said.

Some of their internal tools include:

• Predictive graphics for live sports. These were created in-house. The production team focuses on optimising storytelling as they are fed real-time insights throughout the game as it airs. For example, if a particular pitcher is facing a particular batter during a baseball game, the tool provides them with an analysis of each player’s statistics and how this particular play could go. “It allows us to tell differentiated stories by putting data in the hands of non-technical users,” Hildebrandt said.

• Semantic search for Tubi’s 250,000 titles. Tubi, owned by Fox, is the most popular free CTV streaming service in the United States, and content discovery is a problem, as any of us who have tried to decide what to watch on a streaming service know. “How can you infer intent from users? LLMs are incredible at this,” Hildebrandt said.

• Clipping video to repackage it into different forms. Content is marked with different categories, using an in-house model. This tool uses computer vision on chyrons to understand the entities and topics that are relevant. “Creating video is really compute intensive,” she said. “Here, we are running AI across existing video.”

• Finding the right video clips. This tool has a semantic understanding of text articles and pulls appropriate clips based on how relevant they are. “The newsroom loves it,” Hildebrandt said. “It is a painful process otherwise, and they know it will lead to more engagement.”

• Streamlining product work. This tool ingests stakeholder interviews and pulls out use cases, pain points and opportunities, and creates a summary of discussion and action items and requirements for developers. 

• Data analysis. This tool streamlines tedious manual file comparisons between ratings services and log events, identifying errors in reporting, a daily task for the ratings team.

Worthwhile links

• GenAI and compensation: Britain’s prime minister says AI companies should pay news publishers for content.

• GenAI and transcription: Beware of rampant hallucinations in OpenAI’s Whisper.

• GenAI and data-driven news: The machine does not write as comprehensively as humans, it seems.

• GenAI and content: Over 47% of Medium’s posts are likely AI-generated, versus 7% on global news sites.

• GenAI and audio: Poland’s first experimental AI radio station launches — and shuts down. 

• GenAI and manipulation: Redditors are trying to mislead Google’s AI Overviews so their favourite restaurants are not swamped with tourists.

• GenAI and watermarking: Google releases an open-source tool.

• GenAI and search: OpenAI launches an ad-free real-time search engine. 

• GenAI and agents: We are still in the “super, super early stage,” says Gartner.

• GenAI and misinformation: Answers are more often wrong in Spanish versus English. 

• GenAI and marketplaces: Tollbit raises US$24 million, signs publisher deals

• GenAI and energy: KKR invests in data centres, power generation.

• An AI digression: The Computerized Confessional.

About this newsletter

Today’s newsletter is written by Sonali Verma, based in Toronto, and lead for the INMA Generative AI Initiative. Sonali will share research, case studies, and thought leadership on the topic of generative AI and how it relates to all areas of news media.

This newsletter is a public face of the Generative AI Initiative by INMA, outlined here. E-mail Sonali at sonali.verma@inma.org or connect with her on INMA’s Slack channel with thoughts, suggestions, and questions.

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