Context Window 19
This edition covers Meta’s attempt to block Sarah Wynn Williams’s memoir Careless People and The Atlantic’s forensic exposé of Meta’s training data, the billion-download milestone for Llama, attitudes to AI among Ibero-American publishers from a trip to Madrid’s Parix IA conference, the Thaler v. Perlmutter appellate ruling reaffirming that copyright requires human authorship, Y Combinator’s report on AI-generated startup code, Baidu’s aggressively priced Ernie models, Anthropic adding web search to Claude, OpenAI’s updated voice models, Zoom’s new AI agents, and details of a forthcoming Publishing Scotland webinar.
What a week it’s been for Meta. It started with their efforts to prevent promotion of Sarah Wynn Williams’s memoir Careless People, suggesting they had never heard of the Streisand Effect. Yesterday it culminated in a superb piece of analytical journalism by The Atlantic on Meta’s use of copyrighted content for training data. This is a must-read for all publishers and authors. While the basic facts are familiar to regular readers, the scale of Meta’s data harvesting—laid out in forensic detail by The Atlantic—is staggering. Any publisher of scale will likely find something of theirs in the dataset. Meanwhile, Meta announced that its open source Llama models have now been downloaded more than a billion times. Meta’s aggressive open-source push is making Llama models ubiquitous—while at the same time facing justifiable backlash over how they were trained. I was in Madrid this week to speak at Parix IA, the excellent conference organised by the training school for the Spanish book trade. By coincidence, as I was arriving in Madrid, PW published new research on attitudes to AI in Spanish-speaking publishing: the stand-out was that 54% of industry professionals regarded AI as positive, with only 20% foreseeing negative impacts. It would be interesting to compare that with similar research in the UK and US, where sentiment often seems less positive. But it was very much in line with the pragmatic tone of the conference, which emphasised practical applications as much as big questions around copyright, authenticity and sustainability. Speaking of copyright, the latest round of Thaler v. Perlmutter, litigation that I’ve often referred to in my training courses, ended with a unanimous decision by an appellate panel that copyright law requires human authorship. The ruling further cements the idea that AI is a tool rather than an independent creator, a principle that will shape future cases on AI-assisted works in publishing. I think it might be the first time that I’ve seen a judge referring to Star Trek in a decision. For anyone interested in deeper analysis, there’s a good piece from the Authors Alliance here. Y Combinator reported that nearly a quarter of its current startup cohort has more than 95% of its code written by AI, enabling smaller companies to achieve revenues that would previously have required large teams. Baidu released two new AI models, Ernie 4.5/X1, which stand out for their aggressive pricing: 4.5 outstrips GPT-4o for about 1% of the price, and X1, a reasoning model, matches DeepSeek at half the price. Cost has been a key barrier to use at scale, and this makes AI more accessible to some smaller business users while also challenging the (let’s be honest, quite speculative) profitability of bigger AI companies. Anthropic has added real time web search capabilities to Claude, finally addressing a feature gap with major competitors such as OpenAI and Google. OpenAI updated its text-to-speech and speech-to-text models, with a new demo platform to showcase some of the options. AI narration continues to evolve, but ElevenLabs is still ahead on quality. For publishers considering AI-driven audiobooks, model selection will be key to maintaining production values. Zoom announced a family of AI agents to increase productivity. For publishing teams, AI agents in productivity tools could streamline processes: automating meeting notes, tracking the critical path for publication, and reducing admin overhead. When they can come up with an agent to deal with people talking while they’re on mute, I’ll be really interested. Finally, a personal update: I’m going to be doing a short webinar for Publishing Scotland members on 8 April—register here if you’re interested.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.