dmarchand90

What you do is you compute a lot of simulations with the expensive method. Then you train using neural neural networks (well any regression method you like).

Then you can use the trained method on new arbitrary structures. If you've done everything right you get good, or good enough results, but much much faster.

At a high level It's the same pipeline as in all ML. But some aspects are different, e.g. unlike image recognition you can generate training data on the fly by running more DFT simulations

To plug my own field a bit, in material science and chemistry there is a lot of excitement in using machine learning to get better simulations of atomic behavior. This can open up exciting areas in drug and alloy design, maybe find new CO2 capturing material's or better cladding for fusion reactors, to name just a few.

The idea is that to solve these problems you need to solve the schrodinger equation (1). But the schrodinger equation scales really badly with the number of electrons and can't get computed directly for more than a few sample cases. Even Density Functional Theory (DFT), the most popular approximation that still is reasonably accurate scales N^3 with the number of electrons, with a pretty big pre factor. A reasonable rule of thumb would be 12 hours on 12 nodes (each node being 160 cpu cores) for 256 atoms. You can play with settings and increase your budget to maybe get 2000 (and only for a few timesteps) but good luck beyond that.

Machine learning seems to be really useful here. In my own work on aluminium alloys I was able to get the same simulations that would have needed hours on the supercomputer to run in seconds on a laptop. Or, do simulations with tens of thousands of atoms for long periods of time on the supercomputer. The most famous application is probably alphafold from deep mind.

There are a lot of interesting questions people are still working on:

What are the best input features? We don't have any nice equivalent to CNNs that are universally applicable, though some have tried 3d convnets. One of the best methods right now involves taking spherical harmonic based approximates of the local environment in some complex way I've never fully understood, but is closer to the underlying physics.

Can we put physics into these models? Almost all these models fail in dumb ways sometimes. For example if I begin to squish two atoms together they should eventually repel each other and that repulsion force should scale really fast (ok maybe they fuse into a black hole or something but we're not dealing with that kind of esoteric physics here). But, all machine learning potentials will by default fail to learn this and will only learn the repulsion to the closest distance of any two atoms in their training set. Beyond that and the guess wildly. Some people are able to put this physics into the model directly but I don't think we have it totally solved yet.

How do we know which atomic environments to simulate? These models can really only interpolate they can't extrapolate. But while I can get an intuition of interpolation in low dimensions once your training set consists of many features over many atoms in 3d space this becomes a high dimensional problem. In my own experience, I can get really good energies for shearing behavior of strengthening precipitates in aluminum without directly putting the structures in. But was this extrapolated or interpolated from the other structures. Not always clear.

(1) sometimes also the relativistic Dirac equation. E.g. fast moving moving atoms in some of the heavier elements move at relativistic speeds.

Clean code talks about this very well. Like all things it's a matter of doing it write or wrong.

The checklist includes:

1 are you using comments to manage obfuscated code? I.e. the comment isn't the problem it's the spaghetti code it makes the person feel entitled to write after

2 can the code just be a function with a clear name

3. Is the comment a non comment ?

<some others I forget the book is a great read>

That said. Plenty of times you need a good comment.

This is the top-performing model for for humaneval code generation https://paperswithcode.com/sota/code-generation-on-humaneval

"Well, Steve, I think there's more than one way of looking at it. I think it's more like we both had this rich neighbor named Xerox and I broke into his house to steal the TV set and found out that you had already stolen it." Bill Gates

From my experience I use llms for writing tests because llms are much better at writing tests than the application code. I suspect this might be due to the fact that the code ends up being a very detailed and clear prompt to the llm.

Thanks! I saw the comment but your post gave me the nudge to actually click the link. I'm glad I did!

Less than .3% of the population getting sick from contaminated poultry seems very reasonable?

I think the article is very honest about this just being a fun exploration. They even show how you can get similar patterns with newton's algorithm which is a more "classical" take

In the most bizarre of coincidences I just installed anki now before seeing it on the front page of hackernews

I wanted to write this!

You can read old George Orwell essays about his experience with the media in the Spanish Civil War that sound like they could have been written yesterday. He talks about how he felt the media was entering some sort of post truth era even then from the blatant lies and total split in reality the newspapers would have amongst themselves

Ahaha yes, I discovered this after writing :) I think the main argument is still that Europeans don't get as much vitamins/minerals artificially added and 'may' (though not necessarily) be better poised to benefit.

I say may because these things are complicated. I also suspect people here give better respect to food and might be eating a more balanced diet naturally

Update: Experts are pushing back against the Spectator's reporting — including one source quoted by the magazine who says her remarks were taken out of context.

I wonder to what extent this is for Americans, whose bread and cereals are so loaded that they're effectively multivitamins anyway.

I live in Norway and just realized that they don't put iodine in their salt. (Actually a lot of minerals I would have normally gotten from enriched bread back home may or may not be added here). Locals eat a lot of white fish and get their iodine that way. But I don't and have started supplementing just in case.

There are two major frameworks managing DFT calculations: AiiDA and Fireworks. What's strange to me is Microsoft has put a lot of support into AiiDA (including hiring a few high profile developers). Yet in the paper they use Fireworks to run the simulation

Are they really "laws" my impression is its all just a bunch of empirical trends.

We cannot know truly how these parameters interact at large scale and also how they interact with each other.

Is it really the case that openai has data that Google doesn't?

I kinda wonder if maybe it's at least partially due to openai hitting a kind of hyperparameter lottery. When each experiment costs millions it might be that (aside from good/ unique data) they just have a good set of hyperparameters used in training and it's too expensive for a competitor to find equal or better settings

A fun fact is that even in humble fracture mechanics one gets singularities at the crack tip using standard elasticity theory. A bunch of funny mathematics is needed to get around this. And, of course, the reality is that at a certain point you hit the atomic level so the tip is always 'blunted' at least one atoms length.

For me it seems strange that nobody is mentioning the obvious: noise. I would like to see a study where someone adds "RANDOM_SEED N" (with being some random 7 digit number) to each message check the distribution in lengths and use that as a control.

There was a follow up paper that coupled this with an autonomous lab and was able to reproduce a lot of the structures.

Given the goal of:

vore.website is free for anyone to use, and will never be turned into a vehicle for profit.

It's actually a brilliant name, what a way to keep away mainstream corporate interests

I was hoping to be the first to notice this :D

It actually reminds me a lot of much older product: https://www.youtube.com/watch?v=vj24kNJEQJs (bonyt noticed this first)

I'm still not very clear why you can't just leave this as a personal moral decision rather than a legal one.

The only part I disagree with is the historical perspective. When was this time that people believed in personal growth? For most of history the nobels were nobels, peasants were peasants and that was that. If anything, the sense of having dynamic control of one's destiny throughout one's lifespan is a recent invention. (Well at least in the west)

Yes they used to be luxurious, because they all used to cost more than first class prices. You can still fly the 1960s, actually much better, if you want to pay like the 1960s https://brilliantio.com/how-much-were-plane-tickets-in-the-1...

All organism increase entropy

Now I'm curious how the results of study posted would be mitigated by the space barbell

I wonder if you could bypass these issues with strength training machines designed for space. E.g a squat machine that pushes down from above, cable tension based arm movements, etc. I bet if you're clever you can probably have a small room that can get the essential exercises in and takes up a similar amount of room as a toilet