Man, when I was doing my PhD at CERN the most far out thing that we were generally allowed to use was boosted decision trees :) getting anything other than that approved for any Higgs-related analysis was… not a thing.

We surely came a long way since then. I recently finished my master thesis on self-supervised foundation models for jets :)

It was certainly dogma, but adventurous students were known to have activated the MLP models in TMVA ;)

Small nitpick: neural nets weren't new at all, IIRC. LEP experiments used them for b-tagging and Higgs searches in the 90s, and the Tevatron top analyses too. The dogma was more about trusting them for a discovery claim than the tools existing. BDTs just won on robustness.

Finally! As a SWE, I was getting worried scientists would be left out of the increasingly-worried-about-getting-replaced club that SWEs and mathematicians are currently enjoying being members of.

One of us! One of us!

Join us, in the pit of despair!

I've seen this in a lab setting, and the first useful thing was boring: we pointed a model at run logs to flag weird runs for a human. It was wrong plenty, but first-pass review went from hours to minutes. Someone still had to decide what "weird" meant, so nobody got replaced.

The thing a CERN is: You have some of the smartest and most educated people on the planet, and they are all super motivated to excel.

It’s one of the coolest environments one can be in. You can righteously wall yourself into a room and solve a problem.

I do hope that this is the start of something cool.

You also have the snowboarding weekend trips, the R1 beer garden, and above all for those that were around in the 2000s, L'Usine and Shakers. :)

There's something really cool about this but also something sad.

There was a time when CERN would have been establishing partnerships with US labs. The US is not a CERN member but has provided a lot of funding in the past despite it's observer status. These days funding is getting cut, the US is not as trusted as a partner anymore and a European partner is seen as important. This is sad.

That said I'm happy for Mistral and I'm sure they'll be a great partner for CERN.

This is great progress in fundamental research and I’m thrilled to see Mistral leading it.

Makes perfect sense and I’m glad that CERN is focusing on EU providers

It makes sense in a political way but I'm not sure if it's the best choice for scientific development seeing as Mistral has been pretty behind the curve it's entire existence.

It makes more sense if they're the only company who was willing to work on a more custom solution for the proposed LHC model but I don't that a bit as it would be great PR for the frontier labs.

This seems a lot more like an effort to bolster European AI by moving CERN money into Mistral.

I hope European university networks and their underfunded labs will join forces as soon as possible with other European national research centers.

I’d be surprised if that isn’t already the case with CERN itself already being a multi-national collaboration project and with things like GÉANT https://geant.org/ existing

Wonderful news to not see the duopoly accelerating real science.

Internet 2.0 here we come!

Great news, let them cook.

I'm not at all anti-AI, but I'm still holding my breath to see real scientific contributions coming from modern (LLM/CoT/agent) AI. Much of the premise behind "AI will make superabundance a reality" hinges on getting AI-driven science that affects everyday lives Anthropic's quasi-CRISPR-like discovery did not seem super compelling, and that is still the best I have seen. If anyone has more heartening stories that could convince me to be an optimist, please do share ;)

Is math or theoretical Computer Science not "real science"? Whatever one may think about the process, the results are real and are coming from LLMs.

It is.

What I think the parent post is missing is that breakthroughs in science don't immediately translate into breakthroughs in engineering and even less so into economically viable products.

The distinction for many people is that math and cs feel like very natural domains for LLMs (and importantly, domains that enable quick iteration and verification). Much less applicable with eg biology and other fields that require lots of physical experimentation.

(in)Conveniently those are also the fields where most ordinary people would experience the most effects from accelerated progress. Dario famously promises potential cures for cancer and dementia.

I think there's a pretty clear distinction between math and science. The former builds theoretical models of what certain axioms and assumptions imply. The latter builds actual models of what we see and measure in the real world.

Advances in math will probably result in advances in science, which in turn will result in advances in engineering (which is probably where the average person sees an actual benefit in their day to day life). But it's not automatic and a lot of the hand wringing by mathematicians about the AI math results include "is this (AI) method of finding proofs actually going to result in as many of those downstream advances as the old fashioned way?"

(I include theoretical CS as a subfield of math - and believe the S in it is a misnomer).

well they don't affect anyone's day-to-day lives, which is what the person you're replying to is talking about..

I have a feeling that many more fields are utilizing AI and having some success with it and we just don't know about it in our STEM bubble.

On the other hand, you could downplay the math findings a bit, since it didn't further our understanding of mathematics just yet, only provided proves to things we didn't have proves for.

AI has an incredibly long history of contributing to science and engineering. Some highlights:

Gerard t' Hoofd (under Martinus Veldman) proved in 1971 that Yang-Mills gauge theories are renormalizable, using a rule-based AI (LISP-styled but manually written in assembly) called "Schoonschip" (I'm guessing called that because there were several failed attempts that went before and failed). This program was an AI in the sense of a search algorithm searching symbolic proofs. It got the 1999 Nobel prize in Physics.

This was hugely important because first, it showed that it was possible to make the Standard Model of physics and it also helped beyond that by showing for a lot of (at that point considered) impossible theories exactly how to make them work (meaning not just a proof that it could be done, but a long list of "if you have problem X, do Y" recipes). One of these became first the "electroweak unification", then the standard model.

Hao Wang proved that the logic of "Principia Mathematica" was actually correct using another symbolic proof generator. His efforts were also the start of many other such programs that contributed lots of stuff to mathematics that nobody sane was ever going to do, such as finding all the higher-order versions of the quadratic formula (there's a lot of them).

At Stanford university, tens of thousands of molecules were found in 1965 by having an AI, this one was what's called an "expert system", evaluate mass spectroscopy data and combine it with chemistry knowledge in the form of rules about what was possible or not possible and if only one option remained, spit it out. It almost doubled the sum total of our chemistry knowledge if counted by known molecular structures. It was later adapted to make mass spectroscopy useful in finding the composition of samples.

Then there's Alphafold, first a ConvNet (Resnet+Convnet to be exact), then switched to the transformer architecture, which gave use the folding of just about every protein we have ever found, and even more proteins that we've never even found. Nobel prize in Chemistry 2024.

If you are limiting yourself to modern whatever, you are not going to find that a lot of science that affects everyday lives. Between the first results and practical applications, it can take many years. There is a reason Nobel prize winners tend to be old.

Less modern AI (aka. machine learning) is used all the time in science. I don't know of any commercial breakthrough yet, but there are some promising developments, like in antibiotics. If it materializes, that would be a very big deal.

A lot of little things can add up to a lot but fly under the radar.

The antibiotics work is a decent example. Halicin and abaucin were both found through ML screening, but as far as I know neither has made it through clinical trials. Discovery was the fast part. Toxicity testing and trials are still slow and expensive, which is probably where the lag you mention comes from.

This post so perfectly captures the tragedy of how the term AI has lost its meaning to LLMs.

Because LLMs are the only thing that can be said to pass in some sense the Turing test, which is like the benchmark for AI.

I think maths is really the only solid case so far. I do wonder whether the big early contribution to experimental science will be in speeding up the scientists conducting it, at least until some non-negligible boundary in simulation fidelity is crossed.

There’s been a tendency since well before LLMs for AI demos to focus on “here’s a new problem we made up that it solves that you could imagine will soon transfer to an existing real problem” as opposed to “AI solves X”. This has been an ongoing challenge, it does cool stuff and looks so promising but has been very hard to harness for useful work. It’s still going to take time before the hype dies down and we see what’s left.

I don’t think it’s going away, LLMs and other ML stuff will no doubt remain part of people’s workflow, but I also see no evidence for superabundance or anything close to it.

LLMs are amazing for engineering work. They can automate design and testing of the entire spectrum.

Saw a post recently about how AI agents were used to check if papers had flaws in their conclusions and it found 15% did, that kind of auditing is already a massive massive advancement for science

this announcement is not regarding a large language model

with a different title, they would have achieved much higher engagement. e.g. CERN and Mistral evaluate if large latent models can advance physics

Wait, so we like AI now?

For me, the issue is the whole piracy and unauthorized use of intellectual property with no rewards, and the environmental damage intensifies it by 1000x. I know that this has been the case for many different situations in human history, but that doesn't make it any better.

AI != GPT

I think there is a large group of commenters who actually just hate silicon valley tech companies, and they're fine with AI as long as it comes from someone else.

Poetic considering the locale. But ya, modern tech is pretty gross, and I yearn for the open models.

1. Yes. 2. Maybe

Some people do, some people don't. Some it depends on what for and how it's used as well, as well as what kind.

Maturity is understanding there's an enormous range between "like" and "dislike", between "love" and "hate", between "good" and "bad", and its also possible to be neither at once or both at once.

AI has been beneficial in multiple research fields across maths and science. That is to say, it's actually benefited humanity in those particular use cases.

As opposed to enshitifying the lives of the working class in most other areas.

Yes, for their sense of nuance.

Using AI as a tool to investigate the mysteries of the universe is a huge win for humanity. Generating slop images to confuse people online is not. AI is tool, we like when tools are used for good.

I like this, and CERN might be closer to a potential European solution for us than you think, CERN was built on the premise that nuclear technology was too dangerous for anyone single nation to have. So they decided to collaborate.

Today they’re one of the biggest compute in the world, they have the strongest data interconnects, multi gigawatts of power and their deeply embedded in science thousands of the greatest scientist in the world work there including thousands on machine learning.

In my opinion, high energy particle physics is a dead end yet still they raise money like nobody in Europe for the next future circular collider they’re asking for close to €100 billion

Clearly, there is a path there where a fraction of that money, infra, and a fraction of the incredible geniouses that work there could be rerouted into a Opus/Mythos/Astra shape open / European system

> ... high energy particle physics is a dead end...

Why that? I'm not following all that too closely

> CERN was built on the premise that nuclear technology was too dangerous for anyone single nation to have

What? It’s not that kind of nuclear. The members collaborate because it’s insanely expensive to build these huge accelerators and not many countries have internal pools of scientists big enough to justify having one.

> Today they’re one of the biggest compute in the world

Not really. They have cool computers but the biggest are in the US and China.

> In my opinion, high energy particle physics is a dead end

Out of curiosity, what do you base this opinion on?

> Clearly, there is a path there where a fraction of that money, infra, and a fraction of the incredible geniouses that work there could be rerouted into a Opus/Mythos/Astra shape open / European system

No. People, even smart ones, are not interchangeable, universal geniuses. They end up in one position because they are good at something, it does not mean that they would be productive in a different position. And understanding our universe better is well worth the pennies we spend on CERN. Other people are already working on LLMs.

Besides, it’s not at all what is happening here.