Nik Reeves-McLaren

Nice work! Can we skip a year for great new books at some point please, my TBR pile will be the end of me already

When at Cosener’s… getting ready for my talk at the Winter Crystallography Meeting tomorrow where I’ll be talking about vibe coding and how we trust anything in the age of AI

New @jmaterchem.rsc.org paper on how we can maintain and build trust in scientific data in the age of genAI now spotted in the wild!

pubs.rsc.org/en/content/a...

Really chuffed that our paper on Data Integrity in Materials Science in the Era of AI: Balancing Accelerated Discovery with Responsible Science and Innovationhas been accepted for publication in @jmaterchem.rsc.org!

Local field next to my house looks well. Think Herbie approves

6/6: Next Steps
Ultimately, checklists &manual review won't scale. We need a new class of community-developed tools to vet datasets for statistical anomalies and physically implausible results before they are published or used to train future generations of AI models.

#AI #DataScience #OpenSource

5/6: The Real Fix: Mandate Raw Data
The single most important step? Journals must require raw data files (.raw, .xrdml etc), not just processed plots. Raw data is harder to fake (though not impossible). This one change could be transformative.

#FAIRdata #OpenData

4/6: A Solution: Checklists
We propose a practical checklist for reviewers and authors alike. It provides clear, technique-specific validation steps to make it much harder for poor or fraudulent data to get published.

#PeerReview #OpenScience

3/6: Example 2: Batteries
My field is a key target. GenAI could easily "smooth" noisy battery cycling data, hide capacity fade, or invent a few extra cycles to make a material look more stable than it is. It's a potential nightmare.

#Batteries #Electrochemistry

2/6: It's Deeper Than Just Images
XRD isn't safe. Published Rietveld refinements are often rife with errors: impossible stats (χ² < 1.0), missing R-factors, and physically absurd parameters like negative ADPs.

#Crystallography #Diffraction

1/6: The "Garbage In, Garbage Out" Problem
AI models learn from the published literature. But studies show our data is already riddled with errors - like a review finding over 40% of XPS papers had significant errors. AI will just learn and amplify these flaws.

A short thread on the specifics, because this is a challenge for the here and now. 🧵

I think it's a problem that needs serious attention. AI will revolutionise materials science, but also makes it dangerously easy to fabricate or manipulate data. The traditional peer review process is not equipped to handle this threat.

#MaterialsScience #AI #AcademicIntegrity #PublishOrPerish

Just resubmitted this to the Journal of Materials Chemistry A as a possible Perspectives piece:

TITLE: Data Integrity in Materials Science in the Era of AI: Balancing Accelerated Discovery with Responsible Science and Innovation
AUTHORS: Reeves-McLaren, Nik; Christensen, Sarah

From aerospace engineering studies to restoring Italian grinding machines and bringing vintage scissor patterns back to life - this is exactly the kind of hands-on, real-world problem solving we love to see from our students and researchers.
Sheffield Made. Sheffield Innovation. 🔴⚪

This project beautifully demonstrates how our facilities and talent support local industry. The metal prototypes from the Royce Centre allowed Ernest Wright to iterate their designs efficiently while maintaining their traditional hot drop forging process.

What makes this extra special: Elliott did his UG degree here; the Royce Centre's 3D printing technician George did his PhD with us (In-Situ Monitoring & Control during Laser Powder Bed Fusion of Nickel Superalloys) & continues to work in our School of Chemical, Materials and Biological Engineering.

Elliott Nurcombe helped Ernest Wright & Co redesign their vintage Nurseryman Scissors using metal 3D printing from our Royce Centre.

Pruning in 3D | Stories | Ernest Wright Ltd. We explore the irreplaceable joy of making things by hand – and how this enjoyment leads some craftspeople to undervalue their products.

Delighted to share this wonderful story about @Sheffield engineering connecting traditional craftsmanship with cutting-edge tech 🔧✨

www.ernestwright.co.uk/pruning-in-3d/

It must be soul-destroying for those who don’t.

Hope it gets sorted.

It’s a disgrace and the form is unfit for purpose. I seem to remember we wrote ‘see attached statement’ in most of the text boxes and then attached a lengthy statement that states the case in full in the way the case needed to be told.

It’s just a nightmare. Worth the fuss if you get support but…

This is a really great idea

Apparently there’s a course I can do…

A friends daughter just asked me if I would marry her and her fiance.

Apparently I’m the most sensible man she knows.

When her brother was 4 I turned up to his birthday party with what turned out to be meningitis. I think we need some calibration on where the sensibility bar is

Decide to give it another go. First word: FEET.

Not in the list again. Sigh

Testing last night’s creation, hedgerow jelly. Featuring foraged elderberries, sloes and blackberries. It’s even better than the cherry plum chutney!

I’d been playing daily from the start - this was the breaking point!

The nature need report is paywalled, but the article itself isn't. Read it here: www.nature.com/articles/s41...
this is the wildest thing I've read in a while...

Also: MACE, MADE… but I could have had Cadi or Calid. And Laic, of course

Think I’m done with the @theguardian.com word wheel daily puzzles when today’s word list doesn’t include LICE or MICE.