Context Window 80
A shorter update this week: seven presentations and workshops in ten days have left me genuinely tired, but they’ve also been the best kind of primary research about how people are encountering AI day-to-day. One theme came up again and again: how to use AI in content and marketing without losing your audience’s trust. That throughline runs through a lot of what follows.
Anthropic has started giving subscribers access to its new Fable model, available as part of existing plans through 22 June, after which it will require separate usage credits. I get very cynical about AI companies claiming their latest model is groundbreaking, but this one is very, very good—the question is really whether it will be cost-effective for everyday use. Now is the window to test it.
Amazon’s Story So Far and Ask This Book AI features started rolling out to Kindle users in the US this week, nine months after they were first announced. That delay, and the fact that the features are advertised as being available for ‘thousands’ of titles out of a store of millions, hints at some of the rights discussions with publishers behind the scenes.
One of my favourite AI tools, Google’s NotebookLM, has been upgraded with the ability to run code for analysis, more customisation of outputs, and generation of PowerPoint and Excel files. This is not far off becoming the perfect all-round tool for research and analysis.
Perplexity published a new research piece on how AI agents shape knowledge work, based on use of its Perplexity Computer agent. It suggests that successful use needs more upfront work to frame scope and instructions, but that individuals can then accomplish what were previously team tasks.
New York passed a new law requiring disclosure when AI-generated humans are used in advertising. Expect similar actions in other markets around the world, not least enforcement of the EU AI Act’s transparency provisions from August this year.
Among other things, I’ve delivered a governance session for a major publisher, a conference presentation on authentic content, and an IPG training course on content creation in the last week, and based on the questions and discussion, this is becoming a sharp-end issue for marketers and communications people. Get in touch if your organisation needs help.
Something I missed in my coverage of the Competition and Markets Authority’s action against Google last week: Google’s own response—tonally different of course, but some interesting points of detail including that control of website content in AI results will be set through Search Console. This will be rolling out to UK site owners initially. Worth reading if you’re considering online content visibility as a result of the CMA ruling.
On this topic, a court in Germany found that Google is directly liable for the content of AI overviews because the output is its own content, not just a list of links to others. Google’s defence that users could check sources was rejected outright on the basis that users seldom click on citations.
Most publishers aren’t operating at Google’s scale, but there are two significant so-whats in this, particularly if the ruling sets a wider precedent. If you publish AI-summarised information, it’s not enough to use a disclaimer to push fact-checking on to your readers. And even 90%+ accuracy, the benchmark for Gemini in this case, means a lot of errors at scale.
For educational publishers, Oxford University Press published a major new report on students’ perspectives on the use of AI in education, showing a mix of uncertainty and optimism. Only 15% of learners said they had enough guidance, though three quarters would like teachers to use AI in lessons. That gap is a commissioning brief for publishers as much as a research finding.