Context Window 11

This edition covers the $500 billion Stargate AI infrastructure consortium, impending litigation between the News Media Alliance and an unnamed AI company, OpenAI’s new Operator browser agent in the US, Google’s bundling of Gemini into Workspace and the resulting privacy questions, Goldman Sachs research on small business attitudes to AI, a Nature piece on LLM hallucination rates, an AI book-recommendation app surfaced by Storythings, and personal news on an Arvon masterclass for authors and a non-executive director role at Burleigh Dodds.

It’s been quite a week, one where the old line about weeks where decades happen seems particularly relevant. With a new administration in place in Washington, a consortium including OpenAI, Softbank, Oracle, Nvidia and Microsoft announced a new $500 billion company, Stargate, to build AI infrastructure (power generation and data centres). In a remarkable display of political disunity, this attracted public criticism from Elon Musk, and a quick retort from OpenAI’s Sam Altman: “what is great for the country isn’t always what’s optimal for your companies.” ​ Also in Washington, Semafor writes about impending litigation between the News Media Alliance and an unnamed AI company. It is described as one of the largest actions by publishers, and one that is giving a new lease of life to the NMA. For publishers, license-vs-litigate continues to be the key question. ​ OpenAI users in the US can now try Operator, a browser-based AI agent that can carry out more sophisticated tasks than the Tasks functionality I wrote about last week. I can imagine all kinds of publishing and marketing workflows being built with this. ​ Google made changes to integrate its Gemini AI tool more closely with Google Workspace with a single, increased price (it was previously a separate, paid-for option). Several people I spoke to this week asked about privacy and security implications: if your publisher runs on Workspace it’s worth reviewing this support page for details. ​ Much of the research on business sentiment towards AI focuses on large corporations, so I was interested to see Goldman Sachs publishing research on small business owners and AI: 60% expect time saving, while only 17% expect to see a reduction in jobs. This data tracks closely with what I’ve seen in SME publishers using AI to scale operations without proportional increases in overhead. ​ There is a good piece in Nature on accuracy in LLMs, and how to mitigate hallucinations. While it acknowledges the inevitability of hallucination, I was interested in the data it includes on accuracy rates: the three best-performing models have an average 1.33% hallucination rate. This absolutely does not obviate the need for human review of outputs, but it would be interesting to compare with error rates by human workers. For publishers, such a comparison could inform decisions about where to deploy AI tools in editorial workflows—though even low hallucination rates introduce reputational risk as with Fable last week, and serious consequences in high-stakes publishing contexts like medical or legal content. ​ I’ve subscribed to Storythings longer than any other email newsletter, and it is consistently excellent. This week they highlighted an AI book recommendation app. I’d love to know who is behind it and how it has been trained (the speculation is Goodreads data), but it’s quite fun and I’ve already bought two books that I didn’t know I needed. ​ Two bits of personal news. Most of my work on AI has focused on impacts for publishers, but in April I am running a session on AI for authors for the Arvon Foundation. Quite rightly there is a broad range of opinions on AI among creative people—many strongly held. This session will present a balanced and pragmatic view and help authors decide what is right for them. Please do share with any authors you know. ​ Finally, back on the publishing side, I’m delighted to be joining the board of independent publisher Burleigh Dodds as a non-executive director.

This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.

Written on January 24, 2025