Context Window 42
The AI and content ecosystem is maturing fast. The big news this week is a settlement in Bartz v. Anthropic with implications for other litigation. But besides the big picture, the newsletter highlights new tools and practical things you can try. Have a good weekend.
The news of the week is the announcement of a settlement in Bartz v. Anthropic, the class action lawsuit by authors against the AI developer for copyright infringement. This is momentous but unsurprising: had it gone to trial in November, statutory damages of up to $150,000 per work would have bankrupted Anthropic. As a reminder, the judge in the case had found that Anthropic’s use of copyrighted materials was Fair Use, but the acquisition of those materials (from piracy/shadow libraries) was infringement. This is the first major settlement of AI litigation as it applies to publishing: it’s going to be interesting to see what effect it has on pending cases against Meta and Midjourney, where similar infringement is alleged. As Chris Buccafusco of Duke University put it: “Given their willingness to settle, you have to imagine the dollar signs are flashing in the eyes of plaintiffs’ lawyers around the country.”
AI startup Perplexity is launching a new revenue sharing programme with news publishers, offering a share of a $40+ million revenue pool when it uses publisher content to answer queries. The new business model comes days after the company was sued for copyright infringement in Japan by publishers including Nikkei. Expect more litigation—and more attempts to pre-empt it with monetisation models—following the Anthropic settlement. And while the initial focus is on news publishers, the business logic applies equally to publishers of books and journals.
For anyone trying to get their head around the various copyright issues, the Authors Alliance has published an updated FAQ. Caveats: this is clearly US-focused, and the Authors Alliance has a particular, pragmatic perspective compared to other rights holder groups such as the Authors Guild or Society of Authors. But it’s a useful starting point for research.
Thanks to the American Society of Indexing’s AI Committee for sharing their new white paper on AI and indexing. This unsurprisingly concludes that AI cannot compete with human indexing. In terms of constructive criticism, the methodology of the research is very clear, but it used a range of now-deprecated LLMs (GPT 4.5, Claude 3) and different test texts for each model to accommodate varying context windows. The research gives a good sense of the state of the art with last year’s kit, but it would be interesting to see how tech capability evolves over time.
Noah Brier has an entertaining terse blog post with 29 quick, highly opinionated views about AI: it’s full of sharp insights and useful tips, and easier for you to read than for me to summarise, so give it a read.
Brier’s views on AI functionality and creativity are blunt: if it isn’t working, try harder. That’s easily said in a polemical blog post, but what does that look like in practice? A couple of longer form pieces this week provide useful examples and structure.
Data scientist Tobias Zwingmann outlines a smart seven-point checklist to assess whether an AI prototype is ready to scale—essential reading for publishers trialling tools across areas such as editorial, metadata or workflow automation. For example, does your AI tool actually save editorial time, or just look impressive in demos? Are real users asking for it—or avoiding it? Crucially, Zwingmann stresses the importance of leadership buy-in and early compliance checks, both often overlooked in creative sectors. His core message is clear: successful pilots transition cleanly from prototype to production. If yours can’t, the best next step might be to stop.
In terms of creativity, Katie Parrott discusses building an editorial AI tool for content platform Every: not so much in terms of tech (that’s covered in a separate post), but more about how to preserve style, taste and tone of voice. That’s often a key question when I talk to publishers’ marketing teams about content creation, and this is the case study I’ll be referring people to in future.
I missed this first time round, so thanks to Helen King’s excellent newsletter for highlighting this cool Microsoft demo: the company’s Edge browser can now use open browser tabs as RAG sources for Copilot. Technically brilliant, especially for anyone with too many tabs open, but interesting to think through the control and IP implications for the publishers of those tabs. It also illustrates the growing trend of AI tools leaning on user-curated contexts, blurring the lines between proprietary, public, and licensed content.
Finally, in other news you can use, Google’s NotebookLM now offers video overviews in eighty languages, making research and content outputs more accessible globally.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.