Context Window 3
This edition unpacks the HarperCollins–Microsoft AI licensing deal and its wider implications for trade publishing, OpenAI’s o1 models reaching the API, Gemini’s new memory feature, Ben Affleck on AI and craft, research on AI writing in Substack newsletters, and US dates for the IPG AI training course.
HarperCollins were the first trade publisher to announce a content licensing deal with Microsoft. You’ve probably seen the basics of this in the trade press, but as is often the case, it’s the second and third order implications that are most interesting. Most obviously, there’s now a publicly-reported benchmark of $5,000 per title for a three year licence, which will surely help other publishers value the offers they receive, even if they likely have to accept this will be at the upper end of the range (it’ll be interesting to see if the total deal revenue is broken out on the next News Corp 10-K as that will give some sense of how many titles were licensed). HC has clarified that the deal is subject to author approval is separately accounted, so they seem to have accepted that this wasn’t an existing contractual right and that proceeds won’t accrue to unearned advances. There’s a 50/50 split between publisher and author, though that’s less attractive for multi-author projects. All of which sets precedents for the publishers that follow this. Finally, although there are of course public naysayers, it’s likely that this will be lucrative for both HC and individual authors that opt in (reminder to LitHub and others that author opinion on this issue is not monolithic), and other major publishers may come under pressure from shareholders to deliver similar revenues. A couple of interesting technical developments this week. OpenAI has made its o1 Preview and o1-Mini available via API: for any publishers looking at scripting workflows in Python, the o1 models offer enhanced reasoning and problem-solving capabilities. Meanwhile Google Gemini is rolling out a memory feature, starting with users on the Gemini Advanced tier, allowing the model to remember interests and preferences. Setting custom, standing instructions for prompts is one of my favourite features in ChatGPT, and it’s good to see Gemini catching up on this. I didn’t expect to be writing about Ben Affleck’s views on AI, but his comments at the CNBC Delivering Alpha event attracted a lot of attention this week and are relevant to storytelling in other media: “Being a craftsman is knowing how to work. Art is knowing when to stop. And I think knowing when to stop is going to be a very difficult thing for AI to learn.” 7% of the most popular Substack newsletters “significantly rely” on AI writing according to new research. People who’ve done my training course will know my thoughts on using probabilistic models to find probabilistic outputs (or just false positives), and it’s not clear whether AI is being used to ideate, draft or refine. But it’s interesting to think issues around disclosure, consumer awareness and acceptance, and whether this pattern of use spills over into longer form writing. Finally, I’m delighted to announce that for the first time my award-winning AI training is coming to America next year, with the IPG offering sessions on US time in February and April.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.