Context Window 14
This edition covers the fractious end to the Paris AI Action Summit and Mistral’s new Le Chat, OpenAI’s plan to unify its product line into GPT-5, Adobe’s commercially-safe Firefly Video Model, a Thomson Reuters fair-use ruling against an AI competitor and the Authors Alliance critique of it, the rise of AI-enabled “personal software”, Matt Webb on agentic AI and Model Context Protocol, and a practical tip for getting ChatGPT and Claude to write in British English.
The Paris AI Action Summit ended with the US and UK refusing to sign the summit declaration. If it leaves you feeling stuck between US and Chinese AI models, you could try the newly released model from European AI company Mistral, Le Chat. My complaint last week about the proliferation of AI models and options was clearly timely: Sam Altman posted a roadmap update for OpenAI based around unifying the various choices into a single, integrated GPT-5, within months (“we hate the model picker as much as you do”). Adobe released its Firefly Video Model, which it touts as the only IP-friendly generative AI video model. It builds on the positioning of Firefly as a commercially-safe image generator, based on training data being fully licensed. The old corporate slogan was that nobody ever got fired for buying IBM—in a hyper-competitive marketplace, Adobe is adopting a similar approach. In legal news, Thomson Reuters won a court ruling that a competitor’s use of its Westlaw database for AI training was not fair use. This is being seen by many competitors as a watershed moment, though the Authors Alliance published a critical piece highlighting three flaws in the judgement. This is a great piece on the shift that AI is driving in software development, identifying the emergence of “personal software”, highly specific, niche applications. It resonated strongly as I ran a training course this week where a senior book editor used an LLM to develop a simple and very specific editorial tool that was simultaneously too niche to have been addressed in existing software, but too time consuming to manage manually. It’s interesting to think how many examples like this there might be across an average publisher… As the author puts it: “The real story here isn’t about AI replacing human developers… it’s about how AI tools are enabling a new category of software that simply couldn’t exist before.” Matt Webb writes one of my favourite blogs, and his latest post addresses some of the challenges and possibilities of agentic AI, why Anthropic’s Model Context Protocol is important, what this means for organisations, and why you’ll see changes to traditional web search by the end of this year. It’s a long read, but make yourself a coffee and settle in: “I think we underestimate how deeply the concept of ‘Google is the front door to the internet’ is embedded into companies after 20 years of digital… None of that will hold true any longer. So organisations will be in a situation where not only are they not present and showing up in AI chat but, for team and infrastructure reasons, they can’t adapt. For many this will be an existential problem.” Finally, on a very practical level, if you’re using AI on the same side of the Atlantic as me, this LinkedIn comment thread has a useful three-part prompt for getting ChatGPT to write in British English (similarly, if you’re using Claude, you could create a custom style to achieve the same effect).
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.