
The conference attendees in the lovely Jardin Lecoq. Photo by Nicolas Cartade and Abdelhamid Haddad.
I’s been just over two weeks since we hosted the ATLAS “Exotics and HMBS” workshop in Clermont-Ferrand. Exotics and HMBS (=Higgs, Multi-boson and SUSY searches) are two physics analysis groups, which are communally in charge of almost the entirety of the searches for new particles with the ATLAS detector. Between them, they account for something like 40% of the papers published by the ATLAS collaboration (the other groups are specialised in measurements of known particles and rare Standard Model processes, including precision studies of the Higgs boson and top quark). The whole community numbers nearly a thousand scientists, of which we hosted 140 delegates for a week our capital of the empty diagonal, Clermont-Ferrand.
The purpose of this yearly get-together of folks searching for new particles, is to take time to reflect on our progress year on year, digest the developments and new information we have learned, and plan our next moves. In practice, it is also a chance for individuals to network, exchange ideas, and start new collaborations. Every now and then, when the conditions are right and the stars align, important decisions are made.
I have to confess that hosting a workshop like this in Clermont-Ferrand was not an stress-free task. On the scientific side, one wants to make sure that the right people are in the room, presenting material or chairing sessions for productive discussion to take place. That side was comparatively easy. The more difficult part is the logistics side: making sure the coach actually turns up to drive folks over from CERN, ensuring the caterers actually respect the (numerous and varied!) dietary restrictions of our guests, convincing the IT department that it is, in fact, possible to use Zoom instead of Microsoft teams in the auditorium… and of course, the curveballs no one can expect, such as mass demonstrations erupting all over France a week before our delegates were due to travel.
In the end, all went to plan, without any hiccups. In fact, it went positively smoothly, and I had many of our attendees complement the organisation. Moreover, we had some great discussions and even made some important decisions (more on that below). I’d love to take credit for the success, but in reality all the hard work was done by my post-doc, Marion, and our incredible administrator Florence. As anyone who has ever hosted a dinner party knows, when things go smoothly and guests are delighted, it seems a bit magic. But I know full well that there is no magic about it: the success of such an event rests on careful, meticulous and often thankless work. Florence was on the phone to the caterers several times a day to ensure that they respected the dietary restrictions. Marion had to harass the printing company to make sure our tote bags and branded notepads arrived on time. The two of them carefully curated the coffee-break snacks so that each day offered a different kind of French pastry (something for which we were complemented a lot). Beyond them, our PhD students, post-docs and colleagues in the group, worked around the clock putting up posters, designing interactive maps of bars and restaurants, testing the AV, shepherding people on the coach transfer, and organising an sold-out outreach event. I want to take a moment to thank all of the people: Adrien, Abdelhamid, Eva, Melvyn, Arthur, Andres, Nicolas, Sam and Bruna, I am so proud to have had such an invested and proactive Local Organising Committee!

Our workshop poster and logo, digitally painted by Melvyn Wirawan, PhD student in the Clermont-Ferrand ATLAS group. Here they are compared to the iconic Clermont-Ferrand cathedral which inspired them.
Of course, this was an ATLAS-internal workshop so I can’t discuss specifics, but there are a few general trends which I can highlight. One thing which really permeated the entire meeting, was how the advent of AI is upending the way we do physics, and how quickly the change is coming. When we met last year for the same workshop in Edinburgh in 2025, the use of LLMs was seen as a sort of niche topic. But this year, it is clear that it is already creating crises, making some old arguments obsolete, and opening up new opportunities. Probably this reflects the fact that LLMs have gotten, actually, rather good at coding. But the pace of chance is dizzying.
I think that we physicists have actually been extremely cautious in our adoption of AI: it already seems clear that in industry, in data science, in software development companies, there has been a push for mass adoption. But we pride ourselves in collider physics of having fine control of the details of our analyses. Could we really “let go” ? I think that in some places yes: for example, a long-standing argument in analysis preservation (for the nerds: whether to preserve our analysis selections in the Rivet or SimpleAnalysis software), which was still raging even last year in Edinburgh, is now obsolete: someone showed that Claude or similar, can basically seamlessly convert one format to the other. Moreover, someone else showed that if a paper is well written, once can just give a frontier LLM an arXiv link, and a pretty good Rivet routine will pop out the other side. Still on the topic of analysis preservation, I made the case that we can actually completely fill the backlog of analyses whose results are not yet digitised on the HEPData platform, with the help of coding agents. As an example, at the conference dinner, I had discussed an analysis which was missing from our HEPData portfolio with one of the conveners of the relevant group. It was deemed a hopeless case because only the raw, unsorted inputs remained, and the student who knew how to reproduce the final plots we needed to digitise had graduated and left the field. I told the convnenver to send me the inputs. I said I would have a draft of the HEPData record by the next evening: instead, I had it for them by lunchtime. I tell this anecdote because it shows that LLMs can be used as accelerators of this kind of task, which is critically important but extremely tedious, and hence often left aside. Now, in the space of a year, we can see hope that previously unreachable goals (for example of having 100% of all ATLAS search papers on HEPData with a matching Rivet routine), suddenly seem totally within reach, thanks to progress in AI.
On the other, hand, a lot of the discussion was also worrying about the ravages LLMs could do in our community. While AI can be an accelerator for time-poor academics who have zillions of ideas but are rusty on their coding skills, it can be an existential threat for younger colleagues whose jobs may be replaced by a Claude subscription. More deeply, it also calls into question what the purpose of scientific research actually is. Is society interested, like us, in the answer to some esoteric questions about the nature of the universe, or is society actually interested in the new ideas, methods and breakthroughs which come as a by-product? Particle physicists care about how the electroweak symmetry-breaking mechanisms works and whether it could have led to a first-order phase transition in the early universe, leading to an overabundance of matter over antimatter. I’m willing to bet that few people reading this blog give a damn about that. But to study that question, we had to invent ever more complex colliders, data-sharing techniques, and power transfer methods. Some of the outcomes of that are microwave ovens, proton therapy treatments for cancer, and the word wide web. But if, thirty years ago, we had only to type our question into a prompt to satisfy our curiosity… would all those technological leaps have been made? Probably not. What is the point of knowing the answer to a question if you can’t follow the reasoning? In the Hitchhker’s Guide to the Galaxy, scientists invent a computer to answer fundamental questions. “What is the meaning of life, the universe and everything ?” They ask. The answer is an unhelpful “42”. So it is possible to know the answer to a question, but not learn anything in the process. I think I will need to make an entire blog post to address this, but suffice to say this discussion was a common thread all through the workshop.
On a more human aspect, I was very proud that in parallel to our workshops we hosted some awesome outreach activities. It’s not every day that a big important physics meeting comes to little old Clermont-Ferrand, so we wanted to make sure the local population benefited. Adrien, one of the recent post-docs in our group, took charge of organising an evening in a cinema, where the general public were shown short clips of Sci-Fi movies, and a couple of our french-speaking delegates from the conference were there to answer their questions and clear misconceptions. The event sold out the 70-seat venue, and we had to move it to a bigger auditorium to accommodate the demand ! I also asked our local masters students to be microphone runners, and to attend out poster sessions. Even though they had important exams the next week, a dozen of our students volunteered. It was very satisfying for me to see them absorbing the discussion, and seeing up close how scientific discourse works. As the next ATLAS outreach coordinator, I felt very happy to have used the meeting as a way to also engage and excite the public as well as our future colleagues.
This post is already too long, so I will just finish by saying, that I thought that we also had one of the best visual identities of any recent particle physics meeting. We made a deliberate choice to have the workshop poster be designed not by AI, but by a human. By chance, Melvyn, one of our local PhD students, is a very talented digital artist. He designed our poster as well as the logo, based around the iconic Clermont-Ferrand cathedral. We used the logo and poster on our tote bags, note books, name cards, and stickers given out as goodies. I lost count of the number of people who complemented our visual identity. Multiple people asked to bring home one of our glossy-print posters to hang up at their institutes, and I now often see folks with our workshop logo nestled among other stickers on their laptop cases. Some speakers even used our poster in their talks to illustrate physics points. (see below). I think the reason people responded positively, is that we made a deliberate choice to avoid making a generic AI slop poster: they recognised the care and effort of a human being who invested hours of their life to make a vibrant, human poster. I think there is something in that, to remember and hold on to, as the AI tsunami washes over us all.

A screen shot from a talk by Thomas Critchley (University of Geneva) about alternative data collection workflows, balancing resolution against collection rate.
All in all, this workshop was a wonderful way to hand over the leadership of the ATLAS Exotics search programme to the next generation of conveners. My next challenge will be to tackle ATLAS Outreach activities: I start that new mandate in March.