Are you thinking about using GenAI for audio? 

According to an impromptu recent straw poll on a recent INMA GenAI Webinar, almost everyone is either trying it already or contemplating it.

The reasons are many. Podcasts can be monetised well, and they typically help news brands reach younger, richer audiences, as Svenska Dagbladet’s head of product and UX Ebba Linde, pointed out during the Webinar.

The Schibsted-owned brand started experimenting with GenAI audio recently after its Norwegian sister company, Aftenposten, found articles read in its synthetic voice were effective at engaging readers.

Svenska Dagbladet tried using a cloned voice, but it did not work well in Swedish. Undeterred, it turned to manually recording a few articles a day to see if its readers would appreciate them.

“The results blew us away,” Linde said. Older and younger consumers alike provided positive feedback. The only negative feedback was they could not find where they could access more audio. In addition, some users are not keen to keep pressing the “play” button, since they may be using their hands for another activity (such as driving or cooking). 

So, Svenska Dagbladet built an AI-driven playlist that automatically plays another audio article. The technology taps into user-intent data as well and is based on Schibsted’s in-house personalisation recommendation engine for text articles. (For comparison, here is how Amazon Music and Spotify do it.)

Svenska Dagbladet now uses the synthetic voice for some shorter articles, such as the morning briefing, and human voices for longer reads because it has found users generally still prefer the human voice to the AI clone sometimes, “and listeners are super happy.”

Here’s a look at another experiment: A radio station in Switzerland, Couleur 3, let AI run its programming for 13 hours. The voices, the scripts, and even the music were created by AI at this radio station, which is part of the Swiss Public Broadcasting Corporation.

“We wanted to have this as the starting point of discussion with our audience to understand how it feels like to listen to radio that is made by a computer,” said station head Antoine Multone.

“We now use AI for a lot of tasks. It helps us with live transcription and with tagging content. We want to use it to distribute our content in a better way… . For any of these tasks, a human will check everything that was made by an AI, so we never leave AI by itself. We use it to go faster, and we use it to do things that we don’t have time to do.”

These applications are in addition to previous audio initiatives we have written about at INMA. Based on AI companies’ investment in voice technology, we should expect them to only get better, according to Wharton Professor Ethan Mollick. 

“I think that voice capabilities like GPT-4o’s are going to change how most people interact with AI systems. Voice and visual interactions are more natural than text and will have broader appeal to a wider audience,” Mollick said. 

“The future will involve talking to AI.”

Date for the calendar: September 4

Wednesday, September 4, 2024: Please join us for a Webinar on insights and inspirations for building GenAI chat products. The members-only Webinar will feature two prominent names in the business: The Washington Post’s Phoebe Connolly and Schibsted’s Martin Schori.

GenAI and money 

We’ve seen a few intriguing new developments over the past few days regarding GenAI and money.  

News publishers have signed up for two different sorts of revenue-sharing programmes run by tech companies.

The first is run by Perplexity AI, a generative AI company that first received rave reviews and then faced accusations of plagiarism. It uses conversational AI to synthesise and summarise articles from the Web in real time rather than offering a list of links to those sites. It plans to launch a new advertising feature by October that lets advertisers recommend follow-up questions along with each answer displayed.

Brands such as Der Spiegel, WordPress, and the not-for-profit Texas Tribune have signed up for a “double-digit percentage” share of revenue that is generated every time a user asks Perplexity a question and it cites one of the publishers in the answer. They get paid per article cited.

The company plans to have 30 publishers enrolled by the end of the year, and it is looking to partner with some of the publishers’ ad sales teams. The news organisations will have access to data on which queries are surfacing their content and driving revenue.

(This is different from the agreements that have been signed previously with Google and OpenAI because this is a revenue-sharing deal rather than a licensing deal that covers archival content for training purposes and does not focus simply on big publishers.)

The second agreement is between some of the biggest names in the business — Financial Times, Axel Springer, The Atlantic, to name a few — and an AI start-up that most people have never heard of, Prorata.ai, which is launching a subscription AI chat product over the next couple of months.

“ProRata’s technology enables generative AI platforms — for the first time — to accurately attribute and share revenues on a per-use basis with content owners. The company’s attribution technology will protect and reward creators while preventing unreliable content from driving AI answers,” according to the company.

The company promises publishers and authors a 50/50 revenue-share and a chance to monitor how AI systems are using their work. It uses multi-dimensional attribution algorithms to determine the fractional contribution of content used to generate a response, and covers generative text, images, music, and video. This could eventually be licensed to bigger AI firms, such as OpenAI, and to marketplaces such as Tollbit that bridge the gap between AI firms and publishers.

These two agreements come as many investors raise concerns about whether all the money pouring into GenAI at tech companies will ever be recouped — and which companies will still be serving clients a few years from now. 

You’ve read about it here, at INMA, before. Now, the Information reports OpenAI is set to burn US$5 billion this year. Goldman Sachs points out AI capex will cross US$1 trillion in coming years, with few tangible results.

“AI technology is exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn’t designed to do,” said Goldman’s Jim Covello. He is skeptical AI’s costs will ever decline enough to make automating a large share of tasks affordable.

“In the 18 months since ChatGPT kicked off an AI arms race, tech giants have promised that the technology is poised to revolutionise every industry and used it as justification for spending tens of billions of dollars on data centers and semiconductors needed to run large AI models. Compared to that vision, the products they’ve rolled out so far feel somewhat trivial — chatbots with no clear path to monetisation, cost saving measures like AI coding and customer service, and AI-enabled search that sometimes makes things up,” writes Claire Duffy at CNN Business. 

“In short, investors’ fears can be boiled down to: Is all of this actually worth anything? Or is it just another shiny object the industry is chasing to bring back its dreams of endless growth before it abandons it and moves onto the next big thing?”

Worthwhile links 

• GenAI and search engines: Looks like people use generative search engines primarily for knowledge-work tasks: “They are not simply looking information up but instead doing tasks that involve processing that information, for example, by analysing or evaluating it,” according to the report.
• GenAI and scraping: AI companies change the names of their bots to circumvent crawler blockers.
• GenAI and scraping II: Nvidia scraped videos from YouTube and several other sources to compile training data for its AI products.
• GenAI and lawsuits: OpenAI suggests reporters are essentially learning from public information and repurposing it for their own work, just as it is.
• GenAI and the one that got away: Intel passed on an opportunity to invest in OpenAI.
• GenAI tools: Nine to try.
• GenAI and jobs: They may be meaningless jobs, and they may be replaced by jobs with even less meaning.
• GenAI and the productivity obsession: Whenever new technology is introduced in business contexts, the holy grail always tends to be “efficiency.”
• GenAI and misinformation: Elon Musk hears about it from former top U.S. officials.

A non-AI diversion

Talk about being opportunistic: Someone cybersquatted on the HarrisWalz.com domain for four years and just sold it for US$15,000. He’s now looking at the future as well. 

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.

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