Blog

AI verifiers are getting better and cheaper at finding errors in research papers
This is a lovely visualisation from Project APE at the University of Zurich’s Social Catalyst Lab, tracking how good AI verifiers are at finding deliberately planted errors in AI-generated research papers, and how much it costs to find them. Models released over the past 18 months went from finding about 4 in 10 errors to finding almost all of them, while the cost of finding roughly three-quarters of the errors fell about 1,000×, from $2.50 per paper for an August 2025 released model to $0.0025 with the latest models.

TL;DR: I took part in Apart Research’s Digital Minds sprint in August and learnt a lot. It was my first attempt at doing this kind of work. AI tools made it surprisingly easy to set up and run [poorly designed] experiments on how models describe their own preferences and experiences, but the model’s advice about

What happens to unpublished research once it’s typed into ChatGPT or Codex? A look at OpenAI’s public statements and terms and conditions to see what they cover, and what they leave unclear.

This issue is about AI writing: journal disclosure policies, authorship, watermarking, detection, a launch, some data on slop and hedging words, reflections on writing with AI and some longer reads.

I took a look at the preprint, Negligible participation in the scientific literature of AI unicorn startups, over the weekend and, whilst I think, who the hell am I to critique the work of John Ioannidis and team, I also think this is a weird preprint [1]. The study population is AI-related unicorns in 2025,

Here’s a simple exercise for testing Pangram on academic content. It takes around 15 minutes. You follow one passage through three stages: generated by AI, run through a humaniser, then copy-edited by another AI tool. Pangram 4 is harder to fool than earlier versions, but not that hard to fool.

I’ve put together a new datasheet about women and AI, collating study data about adoption, work, exposure to job change, attitudes, and who holds AI roles. Broadly speaking, a similar percentage of men and women are using AI, with Pew’s February 2026 survey reporting that 47% of women and 50% of men had used an

arXiv now hosts more than 3 million preprints, shown here against the total number of articles on four publishing platforms.

🚨 New issue of PubTech Radar is out: https://substack.com/@pubtechradar/p-202950318. Contents include: 📰 News: PLOS’s move to Janeway for submissions, Google letting publishers opt out of AI Overviews, Company of Biologists’ Fast & Fair trial cutting peer review decisions from 37.7 days to 5.5. BookStories (‘openscience.works), one of seven projects on the Dutch “Renew the Book 2026”

TAIS 2026 was a one-day conference held at Oxford’s Examination Schools on 14 May 2026, organised by the Oxford Martin AI Governance Initiative and Noeon Research. It brought together researchers working across technical AI safety, governance, evaluation, autonomous systems and international cooperation. The meeting covered how models reason, scheming evaluations, autonomous agents, formal verification, human

The recent AI Publishing Collective event at BCS London, hosted by BCSWomen, and BCS Publishing, brought together experts from across scholarly publishing to discuss how AI is shaping peer review and research integrity. AI + Preprint = The End of Journals? Chris Leonard opened the evening with a thought experiment: what would happen if we

These 30 people are navigating AI’s impact on academic publishing from the front lines. Whether they’re technologists experimenting with new workflows, librarians exploring new research assistants, or commentators tracking industry shifts, they’re openly documenting what works, what doesn’t, and what’s coming next. This living list captures people actively wrestling with these changes and writing about

David Worlock’s new article, AI and Publishing: Death, Revolution – or Oblivion? raises some interesting points. I’m no longer convinced that ‘disintermediation’ is the right framing for what’s happening in the ‘marketplace for information’. It feels more likely that we’re entering a new phase of intermediation, which is less visible but shaped by different actors.

Cambridge AI and the AI Publishing Collective hosted a joint meetup on ‘AI in Publishing’ at Fora Folgate Street on May 14. It was fantastic to see such a strong turnout from across the Publishing, AI, and Tech communities. 1. Ivy Cavendish (Springer Nature) Ivy Cavendish opened the evening by asking an important question: When