Context Window 56
A slightly shorter newsletter than normal this week, as I have been balancing four days on the road for a conference and board meetings, and trying to close off various projects before the holidays. I’m planning to do one final newsletter for the year next Friday, 19 December, then take a week and resume publication in the New Year.
OpenAI released ChatGPT 5.2 this week, and the reaction from expert reviewers like Ethan Mollick and Florent Daudens has focused on its performance on GDPval, a benchmark for knowledge work which is probably the most relevant of various benchmarks for publishing-related tasks. For significant tasks which would have taken a human 4-8 hours of work, 5.2 was judged to produce equal or better results than human workers 71% of the time—and that is, as judged by other humans. If you’ve found earlier versions wanting, it’s definitely worth trying 5.2.
Anthropic published new research using an AI tool to conduct qualitative interviews on AI use with 1,250 professionals, creatives and scientists. The findings are interesting for two reasons. For publishers, it’s interesting to consider the viewpoint of creatives who are balancing productivity increases with peer disapproval of their AI use. More broadly, the research points to the potential for qualitative research to be carried out at a scale previously impossible thanks to resource constraints.
Congratulations to Bloomsbury, which announced a major AI partnership with Google, including access to Gemini, Vertex and NotebookLM models. Alongside previous announcements from Bertelsmann and John Wiley, it’s one of the biggest AI infrastructure announcements I’ve seen a publisher make.
Possibly the biggest AI and media news of 2025: Disney announced a sweeping partnership with OpenAI, including Disney brands and characters being supported in the Sora video model, distribution of the best fan videos through Disney+, deploying OpenAI products within the corporation and making a $1 billion investment in OpenAI. Of course, The Mouse is also suing competitor AI model Midjourney, which doesn’t hurt OpenAI either.
Meanwhile, another OpenAI partner, the New York Times announced a new lawsuit against Perplexity (coincidentally, the Chicago Tribune filed similar suit in the same week). Perplexity followed its usual playbook of positioning the action as anti-consumer and anti-innovation, but it’s racking up lawsuits at a time when OpenAI’s partnerships are effectively inoculating it against litigation from major media players like Disney.
On the subject of the news business, Nieman Lab has published its annual predictions, which are full of interesting perspectives on AI: for publishers, I particularly recommend Davey Alba on AI outwriting humans, Rubina Madan Fillion on AI adding value to investigative writing, Parker Molloy on bias, and Felix Simon on AI bubbles versus consumer AI use. But really, you could make a coffee, set aside an hour to read all of them, and come away informed and challenged.
I’ve commented before on the news media’s willingness to discuss AI strategy and the relative lack of communication from some major book publishers (Wiley excepted). I have no doubt that AI startups would have plenty to say, but a challenge to the publishers who read this every week: would anyone like to share a short, personal view on how AI in publishing will develop in 2026, to be published after Christmas? Get in touch!
Finally, I recently published my favourite books of the year on LinkedIn: publishers focused on AI might be particularly interested in Naomi Alderman’s excellent new book Don’t Burn Anyone at the Stake Today, which puts the current online world in the context of previous information revolutions such as the emergence of written language and the development of the printing press. Highly recommended.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.