Context Window 50
Greetings from New York City, where I’ve been on a working vacation for the last week. It’s my first time back since Covid, and besides the inevitable changes to the city (I could date the last fifteen years of photos by the maturity of the foliage on the High Line) the thing that’s struck me is the ubiquity of technology. Every billboard in Midtown is for an enterprise AI tool, and while this was primarily family time, I had some good conversations with publishers about their use of AI.
The flawed MIT/Project Nanda study claiming 95% of AI pilots fail continues to get a lot of airtime, including from senior publishing leaders at Frankfurt. So it was interesting to read a counterpoint in a large-scale Wharton School study released this week: 75% of companies saw positive ROI with only 5% reporting negative results, 82% of business leaders used generative AI weekly and 88% intend to increase spending. Another new study focused specifically on the marketing industry, based on responses from over 600 creative professionals. I liked this one because its conclusion are immediately useful reminders for publishers:
- There’s no single optimal AI model: it depends on task, culture and context.
- Unmediated AI delivers average ideas (but sometimes the average is what we’re looking for).
- AI is good at volume, humans are good at selection.
The Authors Alliance has a particular ideological position, but continues to publish some of the most interesting analysis on the tension between Open Access publishing and AI training. Its most recent piece argues that OA content that restricts text and data mining is a contradiction. This strikes me as reductive: restrictions are not merely a failure of openness, but reflect legitimate concerns about authorship, access and reuse. But this is a debate all OA publishers should be engaging in.
Questions of copyright and website access aren’t going away. Reddit has filed a new suit against Perplexity and three of the AI company’s partners alleging illegal scraping of content. And a judge in New York rejected an attempt by OpenAI to dismiss a copyright infringement class action brought by authors.
More practically, it’s been a bumper few weeks for new tools and features:
Anthropic has released a new Claude for Excel add-in, which provides a new sidebar within spreadsheets and a range of finance-oriented data integrations. Potentially interesting for finance and corporate development teams within larger publishers, though as ever, accuracy is the key question.
For travel and reference publishers, Google introduced a new Maps tool within the Gemini API, allowing developers to integrate trusted geospatial information within applications and outputs.
Adobe’s Max event saw a range of new AI features for Creative Suite, including AI assistants within apps and the ability to choose from a range of AI models, including OpenAI, Google and ElevenLabs. I thought the most interesting feature was Firefly Custom Models, which allow a designer or illustrator to upload existing assets and train a custom model to create new images—potentially interesting for brands/licensed characters.
Incidentally, I’ve just updated my IPG AI training to include more material on image creation, and I found this Adobe guide to text-to-image prompting a really useful resource.
From ebooks to fugazi books (sorry). Rolling Stone looks at the growing problem of AI-generated slop in the Kindle Store. The piece includes the remarkable statistic of more than 90,000 TikTok videos on how to run an ebook publishing side hustle. Relating this to the story I featured recently on hyperscale podcast production, it does seem that any platform optimised for catalogue breadth has a growing signal-to-noise problem.
I have to say, I didn’t have this problem in McNally Jackson this week, and I wonder if there’s a positive second order effect for print and bookstores (though I worry about how many potential readers don’t make it that far). I was also interested to read Bain’s new research on the book industry, which shows that across all media formats, consumers have the lowest tolerance for AI content in books. For all the noise, it’s a reminder that readers still seek authenticity—and that’s something the book trade can own.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.