Context Window 91
This week’s AI news include developments in Hollywood, the news media, and professional publishing—all of which pose questions for publishing as a whole—alongside practical new features from OpenAI, Google and Adobe that would be useful to a broad range of publishers.
There is a fascinating account in The Ankler of secret discussions in Hollywood to reach an industry consensus on AI, though the first conclusion to emerge from the process, “Hollywood needs to stop treating all AI as the same thing”, seems less than revelatory. On the one hand, I’d like to see similar consensus-building in publishing. However, trade publishing in particular has an unhappy track record for informal collaboration, and this kind of exercise naturally favours the largest players—it couldn’t be representative of hundreds of thousands of independent publishers and millions of authors.
The New York Times has been experimenting with AI for some time in non-news areas, but this week it started testing AI search and discovery features for news. The significance of better surfacing content to users will be obvious to any publisher. But the experiment has also set the Times against its editorial union, which opposes it. PW wrote this week about the unionisation trend in publishing, calling it a “hot labor summer”, and changes to employment law in the UK in October may give unions greater purchase in many workplaces. A debate that has often been framed as tech versus media may be more about management versus workers.
For any publishers thinking about search and discoverability, which is hopefully all of you, my friend Tom Critchlow has a great post this week with a structured framework for thinking about LLM visibility—or how you get from “How visible are we in AI search?” to “Can we prove that our interventions work?”
Sticking with big media, Thomson Reuters announced the release of its proprietary LLM, which claims leading performance at a fraction of the development cost of frontier models—though even leveraging open-source foundations and its own content, the company spent $40 million on development. They also published a good behind-the-scenes piece on the strategy and development of the model. Few other publishers are at sufficient scale to consider similar budgets, or to have the customer relationships to commercialise them, but the strategy of training on specialism rather than scale is a really interesting one.
More practically for many subscribers, OpenAI’s agent tool ChatGPT Work has a new feature, Cloud Browser, which allows it to carry out tasks online, including signing in to websites with a username, password and two-factor authentication. For publishers, this opens up a range of possibilities such as retrieving data from retailer portals like Vendor Central.
Also on a practical note, Adobe shipped an update for its Firefly model with support for audio creation, including music, sound effects and text-to-speech, all licensed for commercial use and useful for marketing materials, podcasts—and perhaps even elements of audiobooks?
For any publishers working with spreadsheet data and Google tools, Google has introduced a new Gemini feature called Sheets Canvas which automates the production of interactive visualisations from your data. I used the feature this week to build a marketing dashboard and to create a kanban board from a project plan, both in minutes, and it’s a game changer.
Thanks to subscriber Dan Doyon for sharing something someone else built: an AI fact-checking tool in Claude, which was set on a published history book and found hundreds of errors and contradictions. This mirrors the use of AI in scientific publishing to identify errors in older papers. A leading publisher claimed on BBC Radio this week that fact-checking “doesn’t really apply” to memoirs (and in some cases, perhaps to other classes of non-fiction), but homemade AI tools may take that decision out of publishers’ hands. I wonder how many surprises are sitting in backlists.
Incidentally, if you haven’t checked out Dan’s platform Readwise, which manages highlights from digital books and articles, you should definitely do so.
A slightly older link which I had missed first time around: The Authors Guild ran a head-to-head comparison of five leading AI checkers, which makes for sobering reading: the variance between them is extreme.
Of course, this is one test and results may vary, but I think the AG’s note of caution on the use of these tools is good advice for everyone reading this: “Publishers should be extremely cautious about using the tools and should never rely exclusively on them. At minimum, publishers should disclose their methodology, keep themselves updated on the ongoing accuracy of the tools they use, and create processes that give authors a fair and full opportunity to defend themselves. Publishers should never cancel a contract or pull a book on the basis of accusations or use of these tools alone; it would be a material contract breach to do so.”