Context Window 29
Maybe there will be a quiet week for AI and publishing. If so, we haven’t seen it yet. Across a range of stories this week, the clear theme is that while AI might automate the output, the value lies in the inputs: good content, sound metadata, thoughtful contracts and sound human judgement. OpenAI has published a guide to identifying and scaling AI use cases. It’s a fantastically practical document that I’m surprised isn’t being promoted more, and the underlying methodology works regardless of what AI model you’re using. It’s not publishing specific and needs a layer of industry experience for full utility—just as well for consultants like me—but I’ll be using it as a conversation starter with all my clients. Accessibility expert Simon Mellins, whose podcast I recently appeared on, shared this excellent case study on using AI to write alt text for images, and which models do it best—something every publisher should be thinking about with EU Accessibility Act deadline next month. Amazon’s latest AI innovation is short audio summaries of product features and customer reviews. These are rolling out now on selected US store listings, presumably headed for wider use over time. There seems to be little opportunity to directly shape those summaries, but publishers can influence the inputs to them through good product metadata and encouraging helpful customer reviews (e.g. through Vine)—hopefully areas you are already active in. Bridging ecommerce and news, the New York Times announced a major content licensing deal with Amazon AI platforms. This includes news coverage and model training, but for Amazon a lot of the value will be in the Times’s content from specialist verticals such as Cookery and product review site Wirecutter. This year the Pulitzer Prize required entrants to disclose any use of AI, and Nieman Lab has an interesting review of how newsrooms used it, across one winner and three finalists. It’s interesting that the focus here is very much on more traditional data science applications, though there are some good generative AI use cases as well. Oxford University Press is the latest publisher to use AI for manuscript assessment and editorial workflow, signing a deal with Hum for its Alchemist Review product. Like similar trade-focused tools such as my friends at Storywise, the emphasis is on productivity and decision-support, not taking decisions away from editors. Nature has a fun interactive feature on AI ethics and how it should be used for academic writing and peer review: take the quiz and find out how your views compare with the panel of 5,000 researchers that Nature asked about the issue. AI platform Poe has published some useful trend data on which models its customers are selecting across key categories such as reasoning, text, image and video generation—useful if you’re currently considering a choice of model, as the data indicate where each major platform is strongest. The latest legal skirmish between publishers and AI developers saw a California federal judge opine that Anthropic’s initial copying of books for training data was a violation of copyright law, but that subsequent use of that material was Fair Use. This underlines the importance for rights holders of successfully demonstrating harm from copyright infringement, something that courts have not been completely convinced by (see previous coverage of Universal Music v. Anthropic and Raw Story Media v. OpenAI). “AI first puts humans first”: digital publishing OG Tim O’Reilly has a thoughtful essay repudiating the view he sees from many in Silicon Valley of AI as an opportunity to put people out of work. His argument is that companies that use AI to cut costs will be outcompeted by those that use it to expand human capabilities (unstated, but the implication is that companies that don’t use AI at all are in danger of being outcompeted by both camps). It’s a valuable point of view given his long experience and the fact that he relates the real opportunity to examples from his publishing business. These are zero-to-one use cases such as translation and ancillary content (it’s also really interesting that he notes that O’Reilly pays authors royalties on AI derived products such as quizzes, summaries and audio).
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.