For UKRI grant reviews, we have to tick a box promising we haven't used 'AI' in our review (even to edit text);

YET, you are fully permitted to use it for writing grants, AND you cannot write in your review that you suspect the grant was written by a GPT or use that as a factor in your scoring.

I have not-irregularly come across PIs during the last ten years who accept an excess of peer reviews, and then farm this out to their students, I suppose to curry favour with the journal / editor, to make their own papers pass more smoothly.

So not really surprised at all?

What a fun little letter! Thanks for sharing.

I'm intrigued!

Hard hat suggests its some kind of large scale facility, but the rats-nest of BNC cables suggests it wouldn't be a large particle accelerator experiment.

Synchrotron somewhere? Late 90s early 2000s ? And whats on the DSO? Ring current?

Is the crazing of the acrylic on the outside? You could probably polish it back up if so. You'd need some opticians rouge (and a lapping cloth) and a lot of time.
(In a pinch, I've used toothpaste + a microfibre cloth to bring CDs back to life.)

Would I be able to use this to suggest the nature of a boundary?

i.e. start a supercell with molA at one end and molB at the other, and have it try and guess how these should mesh (so not the co-crystal, but the interface / boundary)

(evem -> even in your preprint, p10 top)

Just slap a GNN on it; they are shockingly effective.

Oh nice! All unbiased MD?

Just south of the 'Faculty' building, there are double-decker stands which are accessible to the public, which is the best for the core of the campus.

(The rest of the bike parking is behind swipe-card only access.)

Sorry I didn't see this in time!
But you found my favourite spot: just in front of the Flowers building.

There are much more extensive Sheffield stands at the East entrance of college (in front of the RSC Chemistry), and a few council stands on Prince Consort Road (the North entrance to campus).

Ah, did it take down the email server though?

www.bbc.co.uk/news/blogs-t...

www.buzzfeed.com/patricksmith...

Interesting talk!

But if the impacts are due to LLMs being (~free, instant) research / writeup assistants, wouldn't exactly the same economic drivers exist for having research staff? i.e. a reduction in overall quality, while still increasing marginal benefit?

Found these ones in a previous code test,

GBGA.peptide_fitness(["DEADTEEF","DEADTEETH","DEAD","TEETH"])
population: ["DEADTEEF", "DEADTEETH", "DEAD", "TEETH"]
4-element Vector{Float64}:
-5.0
-4.9
-3.0
-1.9

(っ◕‿◕)っ

I still burn mix CDs for my kids! No wifi required, no 'internet of things', can play them in the car...
'Tonies' / 'Yoto' can get in the bin.

And as ACE is a linear model, it's super fast to run, and 'training' the model is well understood and reliable linear algebra.

Many thanks to Christoph Ortner for explaining the mathematics, and helping to get it to actually, you know, work...

We've only applied it to a few basic molecular systems so far (ethene, thiophene, napthalene) but it seems to be massively more data efficient than just slapping a neural network on the problem.

Our expectation was that ACE is a natural representation for transfer integrals, as it looks (and feels!) like a many-body extension of the Linear Combination of Atomic Orbitals (LCAO) which underlies the quantum chemistry calculations.