Context Window 71
I don’t set out with a particular theme in mind for each week’s newsletter: sometimes the things I’ve found in my research cluster serendipitously around a particular area, and more often there’s a range of themes. As it happens, this has turned out to be Use Case Week: practical ideas from Anthropic, the BMJ, consultants Fathm, and Paul Ford. If you can’t find something practical to try as a result of this week’s newsletter, you can have your money back.** Starting with something super practical, Anthropic recently updated its use cases for Claude, each one a simple tutorial for using the model for a particular task. For example, this week, I found this really useful for creating branding guidelines and assets. Most of the prompts given are transferable to other LLMs with a little experimentation, and I would be surprised if you didn’t find at least one that is immediately useful.
Anthropic also released a new managed service for developing agents, being used by enterprise customers such as Notion and Kobo parent company Rakuten. It’s priced at $0.08 per hour over normal token costs. It will be most relevant to other large corporations, but that includes quite a lot of you. And with this pace of activity and enterprise-friendly features, one can understand why OpenAI is worried about Anthropic taking share of corporate customers.
British Medical Journal CTO Ian Mulvany has a good piece on AI capabilities in the latest issue of InPublishing: I particularly liked his observation that AI can help to create things, but it can’t create time—and each new AI development requires ongoing attention.
I’ve linked to Paul Ford’s writing before as he’s one of the most consistently wise and funny people in tech. Case in point: the perfect, practical what-to-do-first-with-AI:
Media consulting firm Fathm published a useful list of the ten areas where they see AI driving real impact for news publishers. With the exception of interviews and transcription, which are obviously more foundational for journalism, every other use case has a clear parallel in academic or trade book publishing.
The Wikimedia Foundation published an update on how it is managing traffic to Wikipedia from AI bots: it is now blocking two billion requests from bots every day. Hopefully relatively few publishers will be affected on that scale, but the foundation’s strategy is a useful early example of how publishers may need to manage bot access, rate limits, and infrastructure strain.
Publisher-focused advertising platform Ozone launched a new R&D platform, Ozone Labs, which has developed a simulator to help publishers understand how their content appears within LLM outputs. The tool isn’t publicly accessible yet, but there is some preliminary data on their site.
The European Investment Bank published new research on job impacts from AI, which shows that AI increases labour productivity rather than reducing jobs. AI is still at a relatively early stage, and the long run impact on employment is uncertain, but this is a useful counterpoint to more pessimistic narratives.
A second really interesting piece of research, this time from the FT, which reviewed political statements from LLMs and found that whilst social media tends to elevate politically polarised content, AI outputs tended to be more balanced, centrist and expert led.
That may sound more contextual than directly relevant to publishing, but stay with me for a moment. Ofcom’s latest Media Use and Attitudes report was published last week, and includes some complementary data: people are using social media less, and AI much more. Some of that may be novelty, but, per the FT, LLM outputs may also seem more reasonable than some social channels.
If this thesis holds, then over time, we should expect to see social becoming less important as a discovery channel for books and chatbots becoming more important, and a move from algorithmic feeds to conversational interfaces, something publishers should be actively preparing for now. How far along are you?
The New York Times ran a credulous piece on a two-person, $1.8 billion company enabled by AI, which promptly got taken to pieces by critics. As ever, the most lucrative software sector is BSaaS.
Finally we have just a few spaces left for the 23 April run of my fully-updated, award-winning training course on AI and Publishing, delivered through the Independent Publishers Guild: if you’ve been thinking about taking it, find out more here.
*The newsletter remains free of charge. 😀
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.