Erik Thiede

Excellent read for anyone trying to understand what is up with all the AI hype nowadays.

And here I thought it was if you try *non*-symplectic integrators you don't go back.... to the same energy value.

Posting a job ad on behalf of Mat Sikora, who is doing some fascinating work on cryo-EM for glycans. He's looking both for PhDs and postdocs.

science.phd.uj.edu.pl/join/biomedi...

Happy to see our first collaboration with the Yang group come through!

Junior/Assistant/Associate/Full Specialist Positions Available University of California, San Francisco is hiring. Apply now!

We are looking to hire (yes, even in this economy!) a jr. specialist to train in protein prep/structural biology related to our AVOID-ome work as part of openadmet.org.

A great position for someone who is looking to be a tech for a few years before grad or med school.

aprecruit.ucsf.edu/JPF05424

The Open Molecules 2025 dataset is out! With >100M gold-standard ωB97M-V/def2-TZVPD calcs of biomolecules, electrolytes, metal complexes, and small molecules, OMol is by far the largest, most diverse, and highest quality molecular DFT dataset for training MLIPs ever made 1/N

Yes, Bogdan is a smart guy who does great work: we met briefly at a conference two (?) years ago, and I've been following his work since.

No worries, I'll start by reading that code.

We are working on some techniques for cryo-EM that look mathematically similar to Lucy-Richardson, so I'm on the lookout for new tricks :)

Can we get a reference for the algorithm? (I saw your bioRxiv preprint from last week, but am not 100% if that is the right paper to refer to.)

Research at risk: Life-saving heart pumps for babies | Cornell Chronicle After receiving a stop-work order from the federal government, the future of a device to help children with heart defects is uncertain.

One of grants frozen at Cornell: $6.7M (4 yrs) to develop heart pumps for babies born w/ heart defects.

It took decades of work to get to point of prepping device for in-human clinical trials. No private company could do this.

The cruelty of the Trump administration is heartbreaking. So to speak.

Agreed: Universities are far too quick to ignore the words of the actual people in power, e.g. J.D. Vance saying, "Professors are the enemy" and "[Orban's seizure of universities is] the closest that conservatives have ever gotten to successfully dealing with leftwing domination of universities"

Welcome to upstate NY: it's great up here!

Our work also solves generalizability. As we know from a plethora of JCP, JPC, and JCTC articles, if something works on the alanine dipeptide it basically works on everything.

This makes the latent encoding immediately interpretable. (Although paradoxically, as anyone who has taught free energy in an undergrad p-chem class knows, this doesn't mean it is explainable).

The variational alanine encoder takes arbitrary data and transforms it into a latent random variable whose negative log-probability is given by the free energy surface for the alanine dipeptide in vacuum.

It is with great pleasure that I announce that the Thiede lab has solved the problem of interpretability in AI for Chemistry with our new neural network architecture: the Variational Alanine Encoder (VAE).

Wishing all the trans scientists out there who are continuing to do amazing work in adverse conditions a good trans day of visibility!

Ugh god yes. Drives me insane, and it actively stands in the way of scientific progress. I really wish I could go through arXiv and find/replaced every use of "AI" or "Machine Learning" with "A complicated statistical model".

On top of that, I really wish they had discussed WHY drug repurposing is so hot: if you are a pharma company and can find a new disease that your drug treats, you can keep the patent active, stop the drug from becoming generic, and continue charging crazy amounts of money *for its original use*.

So is there a methodological connection to the chatbot stuff? At times yes. But I don't think it obviates the very many and very valid criticisms of how these chatbots are being built, used, and disseminated.

Researcher in this area: Deep Learning is a type of Machine Learning, which is a type of AI. So it is AI and (probably) has a deep-learning component. Some models do try and predict performance, so there can be a language model component.

I think it has more to do with you might get locked up for absolutely no reason at all and then flown to El Salvador to die in prison even though a judge said you should be released

Our (SMBP + SMBP alumns) man doing us proud.

No matter what IT does here, they lose: following your tip (and being security-conscious) is really best for everyone then.

I think a lot of people are taking this to be IT's fault in the comments, when often they are just caught between competing demands. E.G.,
1. An instrument requires specific proprietary software that is ONLY supported on Windows XP.
2. The university mandates ALL computers be on actively updated OSs

Also: color commands work!

\displaystyle{\partial_t \rho = - \nabla_{W_2} E(\rho) = \color{blue}{\nabla} \cdot \left(\rho \color{blue}{\nabla} \frac{\color{green}{\delta} E}{\color{green}{\delta} \rho}\right)}

Just cite every paper you've every published every time you write. Bam, linear growth in h-index. Square root people need to step their game up.

Finally, remember that you can crop equations as images! This is an easy way to animate equations: make 5 copies of it, crop them into disjoint pieces, and have them appear one-by-one to emphasize terms. Cropping a large equation vs. rendering small ones ensures uniformity of style.