Please fill out my AI usage survey if you work in publishing. Which models are you using? How often? What's delivering value? Takes under 5 minutes. Contributors get early access to results before I share them more widely towards the end of the month: ai-and-me.innovationideas.co.uk

A great example of the crap security in vibe-coded slop apps... www.wiz.io/blog/exposed...

Bluesky doesn’t really shine yet as a place to discover papers, especially at the intersection of biology and ML. To help a little, I’m going to start sharing papers I find interesting. To kick things off, here’s one from @petar-v.bsky.social and colleagues at DeepMind arxiv.org/abs/2601.22950

“If fabricated studies make their way into the evidence base, they can mislead real scientists and ultimately slow progress for patients." Published in The BMJ, a new AI tool developed by @aidybarnett.bsky.social and colleagues has exposed the scale of fake studies flooding cancer research.

world.hey.com/ian.mulvany/... / I’ve been blogging for 20 years this year.

This year we will launch a BMJ commission on the future of the doctor. We seek your support and your ideas for an inquiry built on the premise that if society seeks the best for patients it must also seek the best for its future doctors. @kamranabbasi.bsky.social www.bmj.com/content/392/...

Ireland getting nervous

I did my first escape room yesterday.

At BMJ we are doing some really interesting and important work with our partners at AWS to see how far we can push Agents in support of peer review. Very early days, but so interesting to see what these systems are capable of right now.

I have a “Riccola” hat. I got it over Xmas, I’m delighted with it. Riiiiiicooooolaaaaa !

Many thanks to our brilliant panel of thought leaders, including @journalologist.bsky.social, @clearskiesadam.bsky.social, @mdinat.bsky.social, @nataliejacobs.bsky.social, @ianmulvany.bsky.social, & @heatherstaines.bsky.social - for your expertise & willingness to share your thoughts

Moving from git branches to git work trees, and I feel like I can test more things at the same time.

Why, for the love of god, is the first link in a Teams Meeting calendar invite a link to the MS Teams help, and not the link to the meeting. Why?

I had a great chat yesterday with the folks behind pivigo.com - I wish this existed when I left academia.

I just finished reading “a memory called empire”. I don’t like the first 1/3 of the book but a friend who I respect a lot had read it so I kept going and was rewarded in the end. I liked it.

#FutureSciPub I do agree with the comment that a core question is the need to have the conversation about what research is for, and that this is a discussion that should happen at the national level.

#FutureSciPub in my opinion the call from the current panel to have funders underwrite loss making infrastructure, while at the same time wanting feature improvement, is unrealistic.

#FutureSciPub as Rebecca Lawrence mentioned, there are no shortage of “solutions” that have been developed. a hint - that the existence of these solutions have not solved the problems indicates that the problems lie elsewhere.

#FutureSciPub - prof jinghai li - call for AI translation between languages. This is such a good idea, and now cost effective to do. I feel there is a moral and commercial imperative to do this.

#FutureSciPub. Thought experiment. How would scielo work if it had 3.5 million submissions in a year? Could it continue to manage at a cost of $400 per article?

#FutureSciPub in case anyone missed it, post publication peer review through any system of engaging with the paper does not work, clear evidence ton Ijad from research gate, anyone holding that torch is misguided.

#FutureSciPub the underlying underdiscussed topics - for me - AI, scale, identity systems. I have now 20+ short pages of notes. Much comments to come.

#FutureSciPub Philip Campbell mentions that like laws of physics, systems change funeral by funeral. Luckily many of the people at this meeting are ready to support change!

#futurescipub It strikes me that the overlap between publishing/peer review for dissemination vs assessment is a bit like a thing sitting in a superposition of states. When we stare at is a bit the thing collapses in unpredictable ways.

I gave a talk yesterday at Publisherspeak in london, some notes about that talk, and links to the slides - https://world.hey.com/ian.mulvany/slides-from-my-talk-about-the-future-of-scholarly-publishing-c4f8c297

Thought - as we get older our photo memories get increasingly filled with dead people. I wonder if services could make a filter about this.

Test2

Test.

Web traffic from AI tech companies is overwhelming journal websites — my latest for @cenmag.bsky.social: cen.acs.org/policy/publi... @ianmulvany.bsky.social

With research funding for medicine and science weathering under Trump’s cuts, what can Americans who care about public health do? @ashishkjha.bsky.social talks to @kamranabbasi.bsky.social about cuts, censhorship, and the resurgence of anti-vaccine beliefs with RFK www.youtube.com/watch?v=ziM2...

Bloop bloop beep beep - bot problems.

The promise of AI in education is it will save educators time. That's rubbish. It's sapping all our time already in working groups, committees, workshops, and marking anxiety. It's the most time-sapping, exhausting and frankly boring thing I've ever encountered in HE - and it's in my research area.

First Saturday morning swim in a while. Right knee still a bit tweak from snowboarding injury in December.

I highly recommend reading this piece about AI and DOGE

Some thoughts about alt text.

In a world where political shifts disrupt the research landscape, ORCID's mission remains steadfast. Our commitment to supporting researchers and their work is unwavering. Read more on our latest blog: info.orcid.org/orcid-still-...

I kicked the tyres on Claude code for the first time last night, I just had a few minutes before going to bed. It was by an order the most powerful code assistant I've used. I was also deeply impressed with the cli UX. I think having strong tests helps provide confidence.

For those affected - these are good resources to know about. Apologies if you've already been sent this by me. europepmc.org covers all PubMed records and more. ebi.ac.uk for big biological data sets especially genomic data. These are not housed in USA.

https://www.tamingllms.com/markdown/toc.html# looks good on the topic of working productive with LLMs in code.

How do you finetune and train a base model? This paper provides a blueprint

https://scholarlykitchen.sspnet.org/2025/01/15/guest-post-an-editors-perspective-on-my-very-last-issue/ things are indeed changing, and we will lose things, so we must also work to gain other things too.

https://www.anthropic.com/research/building-effective-agents I'm a sucker for these kinds of workflow overviews, but aside from that this is great.

FYI not only is Pubmed Central a good replacement, @dimensions-ai.bsky.social has a free version with almost complete fulltext search (80%+) for 152 million pubs and datasets. It includes all of Pubmed. You can even filter by Pubmed.

Some interesting links from February.

https://world.hey.com/ian.mulvany/claude-3-7-is-impressive-how-do-we-start-to-think-about-what-it-can-do-aka-this-a-mike-drop-moment-386f91a2 - Claude and a research paper.

The Research to reader conference is coming up, I'm not going but it looks really good.

I'm at the Wallace collection today. Very good.

I think if we lived in a computer simulation there would be more Easter eggs.

OK folks, I have now created a google form that allows people to sign to show their support for my open letter to the President of the Royal Society. Please sign and share widely: forms.gle/miDciq35oxyw...