PubTech Radar 第 45 期：Made by Humans、Xplore AI Research Suite、LeapSpace 写作教练等发布

Hello, and welcome back. It’s a gloriously sunny and warm autumn day in London, and it's hard to believe it’s October. This issue covers a lot of launches, plus publishing tech news, some stats, AI developments, opportunities and a selection of longer pieces. As always, thanks for reading.

🚀 Launches:

• Press Gazette and Flip-Pay have launched Made by Humans, a tool to check text, images, video and audio for AI or manipulation before publication (Pangram and identifAI under the hood). The tool grew out of Press Gazette’s Reality Wars investigations, which surfaced more than 1,000 published articles citing fake or AI-generated experts. For all the problems with research integrity in academic publishing, journalists have it far worse.

• XSci (a new combined preprint, AI, and paid review platform) is running its Open Review Competition 2026 until 30 November 2026. I think this LinkedIn description of how XSci works is clearer than the website.

• Various new ways to interact with the literature have been launched, including IEEE’s Xplore AI Research Suite; Elsevier has expanded LeapSpace with a new Writing Coach, Claim Radar, and Compare Tables; and NLM has launched Linked Discoveries: A New Way to Explore Connections in Biomedical Literature [Video | Announcement | Aaron Tay on how it works]

• Google has launched Expert Intelligence in Gemini Notebook, allowing you to query 100,000+ books from publishers including Bloomsbury, De Gruyter Brill, Johns Hopkins University Press, Macmillan, O’Reilly and Penguin Random House. See also Steven Johnson on the thinking behind the product.

• Pensoft has launched ORPHO, a new standalone “intelligent workspace for scientific writing” [Press release]

• River Valley Technologies has launched Zintzo Research Integrity, which brings multiple research integrity signals together, allowing editorial staff and research integrity teams to see whether a manuscript can proceed, requires closer review, or be escalated.

• Taylor & Francis has announced the introduction of Plain Language Reviews (PLRs) [note PLR not PLS]. These are “comprehensive research summaries designed to make technical information accessible to readers without a specialist background.” I really like this concept, but the implementation is a bit glitchy in places “Main ÿndings” ???

• Olive Turn has launched Razeta, a platform for putting trusted content to work in AI while keeping control over rights, access and attribution, and Scholar Core, a headless scholarly publishing system, including a WordPress plugin which, yes, can render JATS-encoded research articles directly on a WordPress site. [LinkedIn announcement]

• What AI Reads from Scite. Interesting to nose around the data on what AI is reading via their platform; perhaps unsurprisingly, “The top 10 publishers account for 57.2% of all AI reads (top 50: 76.6%).”

• MDPI has released a first version of an open scientific taxonomy: 114,000 Concepts as a fifth layer under the existing OpenAlex hierarchy (Domain → Field → Subfield → Topic → Concept). Explore the taxonomy here.

📰 News:

• Maverick Publishing Consultants has published Beyond Clicks and Citations: Rethinking Research Value in the Age of AI, about how AI is impacting traditional usage metrics and how different groups are considering their future in the world of AI bots.

• COUNTER has published a progress update to its Generative and Agentic AI Best Practice, which picks up on the work of the SPUR Coalition to standardise reporting requirements for AI content usage.

• Tony Alves helpfully summarises Peer Review Week (PRW) and the sessions he attended, commenting: “capacity must be created through better system design rather than greater extraction from already overburdened people; expertise should be applied more precisely and inclusively; meaningful feedback matters more than speed alone; AI must support rather than displace human judgment; reviewer contributions need to be recognized and carried forward; and trust depends on infrastructure that preserves provenance, context, and accountability.”

• Paul Litvak describes a new evaluation benchmark for empirical psychology papers in How well does AI peer review work?

• Digiday reports that Google has been testing “a Search Console-based program that pays publishers when their content “significantly” contributes to AI-generated responses across Gemini, AI Overviews and AI mode.”

• Slightly old news, but OpenAI is putting their “frontier models and tools in the hands of 100,000 scientists, mathematicians, and engineers—at no cost.” Sensible marketing strategy, but I can see a point where institutional access/rationing of these tools becomes a massive equity headache. The more prestigious/well funded your institution, the more tokens and better models you get? See also Editorial Lead, OpenAI Student Collective: “The Student Community team is building OpenAI Student Collective, a global program that equips students with community, resources, and support to help their peers build AI capability, confidence, and opportunity.”

• Newish BMJ Open paper on remuneration for peer review in The BMJ (£50 payment or a 12-month online subscription to a selection of BMJ journals on top of our usual rewards.) Results summary: “While many see payment as an important step toward fairness, equity, and valuing lived experience, others raised concerns, such as potential tax headaches, negative impacts on disability benefits, or shifts in reviewer motivation. Journal subscriptions were a popular alternative for some reviewers. Choice and flexibility of incentives matter. Offering optional incentives respects individual needs and values.”

📉 Data and interesting stats:

• James Butcher notes that Dimensions has added a new article-type filter for retractions. 71,800 articles were retracted between 2011 and 2025, of which 38,000 (53%) had an author from China. Of those, around 20,000 were published as hybrid or gold open access, with NSFC funding for about 4,300 of them. If you assume an average APC of $3,003, he estimates the cost could be up to $60 million, before waivers and diamond open access.

• Frontiers told MedPage Today that its Research Integrity team had rejected more than 12,000[!] submissions in the previous year based on simple queries of public datasets.

💻AI:

• Stuart Leitch from Silverchair walks through how they are breaking down a complex prompt and farming the various subtasks out to sub-agents using their Discovery Bridge technology. If you still think of AI as a chatbot, this is a fairly accessible explanation of how the technology is evolving applied to search.

• Great notes from Daniel Acuna from the AI for Research Integrity Montreal meeting summarising tools, institutions, and initiatives.

• Back in issue 38 (October 2025), I noted the launch of Paper2Agent, which has been formally written up as a Nature article. I think it’s hard to disagree with Magdalena Skipper’s comments: “My own colleagues who make the Nature weekly podcast asked me if I thought this approach could spell the end of a paper as we know it. I told them I didn’t think so but I do see a great potential for us finally moving beyond the static pdf we’ve all lived with for decades. Over the 25 years that I’ve been an editor, conversations about the future of publishing haven’t changed much to be honest. Although Paper2agent hasn’t come out of the blue - it’s a sophisticated next iteration of the idea of an executable paper that’s been around for a while - it might just finally shift us into a new space of knowledge sharing. But for it to really take off, we as the community need to figure out who’d be building these agents and whether there will be actual incentives for agentifying papers.”

• Anna Abalkina on Disappearing GenAI disclosure statements… Not too hard to guess at the many reasons why this might have happened, but still, it shouldn’t have happened.

• FT Strategies explore 3 ways of governing AI within an organisation: the centralised Centre of Excellence, decentralised or embedded teams, and the Centre of Enablement, which blends central oversight with local autonomy.

🔑 Opportunities:

• Digital Science has announced their new Catalyst grant theme, Agentic workflows you can trust for early-stage ideas. Deadline 5 October

• The NIH has launched the SPARK PubMed/PMC Challenge, offering $250,000 in prizes for AI systems using open-weight models that can answer complex biomedical research questions using evidence from PubMed and PubMed Central. There are two tracks: one for finding specific papers from natural-language descriptions, and another for open-ended research questions requiring synthesis across multiple sources.

• Melissa Cantrell and Lauren Collister (editors) have a Call for Proposals for an upcoming edited volume, Reimagining Scholarly Communication: Frameworks and Trajectories. “This edited volume addresses how we understand the persistent crises of academic publishing and knowledge sharing, using interdisciplinary theoretical frameworks to interrogate the seemingly intractable problems of scholarly communication, including exploitative publishing models, flawed metrics systems, infrastructure sustainability, and access inequities ”

• The OpenAI Foundation has launched Public Data for Health, a new programme “funding the creation and preservation of high-quality scientific datasets made broadly available to researchers.” It is starting with more than $125m in grants to nonprofits and universities, covering areas such as drug development, regulatory data and personalised cancer vaccines. The Foundation says it is interested in future proposals, particularly around connected, scarce and direct biological/health data.

• BMJ is looking for a Technical Product Manager (applications close 11 October 2026).

📚 Longer reads:

• 🎬I enjoyed this presentation about ResearchHub’s experiments (and problems) with peer review, building a journal, and a demo of AI “Expert Finder” and an automated proposal-drafting tool to turn a posted research question into a pre-registered, fundable proposal built from a scientist’s own publication history. Interesting because the mindset behind these services is very different from traditional publishing.

• Don Wallace et al. from Google DeepMind on agentic science and what could be coming our way. Conjecture Machines: AI agents and the new validation bottleneck in science. “The scientists using AI agents are not going back. The technology will continue to improve. The institutions that govern science — and the physical and social infrastructure it runs on — were built for a research enterprise that is now evolving faster than they are. We need to upgrade them for the agent era.”

• This report by Signals at Scale is over 80 pages and more about news publishing than scholarly publishing but still interesting. Value in the emerging AI-mediated information ecosystem. “The report anticipates a transition from an ecosystem organized around fixed content artifacts — articles, videos, podcasts and websites — to one organized around continuous, machine-readable flows of "liquid" information. AI agents will likely gather information at extraordinary scale, manage it as referenceable units, attach provenance and rights, verify and enrich it, combine it with personal and collective context, and assemble moment-specific experiences for humans or for other machines. Distribution may become continuous negotiation among agents over relevance, quality, access, price, permitted use, and action. Information flows may become bidirectional, with signals flowing back from consumption to producers and infrastructure providers.”

• Anita S. Coleman on a quite spectacular book/metadata blunder.

• Andrea Chiarelli asks Whose infrastructure is it, anyway? And closes with “The (daunting) task ahead is to widen what we mean by openness. Access was about reaching the research. Agency is about whether the people and institutions who produce it stay competitive and legitimate as machines get better at doing what researchers do.” On a similar theme, Who Should Innovate for Science? Prophy Predicts [Webinar Recap] “If you really care about creating infrastructure, you have to move into some kind of non-academic sector.”

• 🎧The Year AI Came For Us: Teaching Entrepreneurship Will Never Be The Same. Interesting for anyone working in Product Development. AI gets you to a working prototype really quickly, but then people find it much harder to abandon the prototype and move on.

• Microsoft AI has a new quarterly publication, The Humanist Review of AI, which explores artificial intelligence through philosophy, culture, and society

• 🎧I am working my way through the “Metascience 101” podcast series. “A nine-episode set of interviews that doubles as a crash course in the debates [US/policy focus], issues, and ideas driving the modern metascience movement” by Caleb Watney, Matthew Esche, and Tim Hwang.

🔚 And finally…

Neil Blair Christensen has made Scholarly Tradeoffs, a fun and clever promo: “You run a scholarly journal, and the job arrives as email. Every reply pleases one group and angers another — you can’t keep them all onside. Answer wisely, survive as long as you can, and see how your tenure stacks up.” Should you be facing any of these issues in real life, you can meet Neil at the Frankfurt Book Fair to discuss.

Thanks, as always, to everyone who takes the time to get in touch with links and feedback.

Helen

End notes:

• If you found this useful and feel so inclined, you can always buy me a coffee.

• If you need consulting help navigating any of this, find me at Maverick or here.

Thanks for reading PubTech Radar Scan! Subscribe for free to receive new posts.

注意转载授权 · 阅读原文