Context Window 31
It’s Friday 13th. Unlucky for AI image platform Midjourney, which is about to find out why “don’t mess with The Mouse” (or an earthier paraphrase) is a popular aphorism in media law. Luckier for publishers, with some powerful new tools this week: in particular, Google’s Deep Research could be a game changer for production of ancillary content. James Butcher’s always-excellent Journalogy newsletter looks at the documents from the recent Springer Nature AGM, including the section on executive compensation. The company’s Chief Publishing Officer achieved 150% of target for “market performance and usage of AI in publishing processes”; the Chief Operating Officer also hit their goal for “editor satisfaction and implementation of new AI tools.” James makes the point that “what gets measured gets done.” The corollary is that what isn’t measured and incentivised will struggle to get traction. In my experience, too many businesses expect AI—and digital innovation more broadly—to be an extra on top of the day job. But unless it’s built into performance targets and compensation—measured against real business outcomes, as at Springer—it risks being just another talking shop. Are you building AI experiments into your goals and incentives? Bad news for AI image generator Midjourney, which was hit this week by a lawsuit from Disney and Universal. It’s hard to imagine litigants with deeper pockets or greater determination to protect their IP, and the filing highlights some pretty hubristic actions by Midjourney (including failing to respond to prior legal correspondence). On the other hand, if a Fair Use defence prevails in this case, it’s likely to hold more broadly. Google released its own Deep Research tool in Gemini, which is able to research online information and files uploaded by users. The most useful aspect is the ability to create new media based on the research: with a single click, it produces infographics, audio overviews and quizzes. I’ve been really impressed by the results. For educational and non-fiction publishers, it could streamline creation of ancillary content from manuscripts. Thanks to subscriber Alex Boden for recommending Prompt Maker: a custom GPT that takes basic prompts and develops them into more sophisticated, structured instructions. Worth trying if you’re having issues with a particular task or are looking to level up your prompting. OpenAI released its latest model, o3-pro, which it recommends for uses where reliability matters more than speed. The benchmark performance is hugely impressive, as is an 80% reduction in cost for the existing o3 model, suggesting deeper economies of scale than many expected. In parallel with the new release, Sam Altman published a new essay on the trajectory for AI. The general tone is big picture boosterism: it’s crying out for footnotes and data sources. But it contains one specific claim that caught my eye: that a single ChatGPT query uses only ~0.34 watt-hours of energy—that is, roughly ten times more efficient than Andy Masley’s recent, already revisionist estimate. On the face of it, this is really positive news, and may explain why significant price reductions are possible, though if the picture is truly better than expected, there’s no reason AI companies shouldn’t publish detailed environmental reports. If you’re looking for an alternative to o3, French AI company Mistral released its first reasoning model, Magistral. In a competitive market, its positioning is interesting, highlighting traceability and audibility of its reasoning as making it particularly suitable for regulated industries. These compliance features could make it particularly attractive for scientific, medical, and legal publishers working with sensitive content. Publisher Keith Riegert did a superb presentation at the US Book Show last week on how he is using AI in his business. He shares some highly practical advice on what to try right now, and predicts a New York Times bestseller from an agentic AI publisher by 2030. My bet’s sooner—2027. What’s your over/under? Finally, the first step to figuring out where we’ll be in two years is establishing where we are right now. My friends at BISG are running research on that topic: if you are a North American publisher, please take ten minutes to complete their survey.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.