Retrospective, Learnings, and Takeaways from the Heidelberg Laureate Forum 2026

Started Sep 27, 2026; Published Sep 27, 2026; Last updated Sep 28, 2026; personal note science GenAI

This year, I had the opportunity to attend the 13th Heidelberg Laureate Forum (HLF) as a young researcher. The HLF brings together 200 carefully selected young researchers from all around the world with the laureates of the most prestigious awards in computer science and mathematics (namely, the ACM A.M. Turing Award, the ACM Prize in Computing, the Fields Medal, the Abel Prize, and the IMU Abacus Medal). Through a mix of scientific and social program, it enables exchange across borders, both on the most recent research in computer science and mathematics and on the most pressing challenges these disciplines face. This year’s forum was dominated by discussions on how computer science and mathematics can adapt to the rapid advances in artificial intelligence (AI). Mathematics in particular is currently forced to reconsider its role and core principles, which was also the topic of a panel discussion (now gaining popularity on YouTube). I returned from Heidelberg on 19th September, but the impressions are still sinking in. Now that I have caught up with work, I will summarize the event from my perspective and document some things I learned about “How to HLF”.

How to HLF

There are several learnings and observations on how to make the most of the HLF that I want to pass on to others, and to my future self, who will hopefully attend the HLF again one day. I only list the ones that go beyond the typical advice, such as “Do not be afraid to talk to the laureates.” They are written from the perspective of a young researcher, but may also apply to other modes of attendance.

Socialize on arrival day. On your day of arrival, you are going to be tired. However, finding a small group of people to explore Heidelberg and the surrounding nature with on the first day is very enjoyable, and it is a great opportunity to make some first contacts that make it easier to join other groups later.
Talk to the “green lanyards”. Especially if you are looking for job opportunities. During the event, the different groups of attendees wear color-coded lanyards. The color code was announced during the opening ceremony and in the accompanying booklet. Green was the color for the scientific committee and members of supporting institutions. These are mostly professors, some from very well-known institutions from all around the world.
Explore vs. exploit when networking. You will meet a lot of new people during the first couple of days, especially at the reception after the opening ceremony and during the lunches. After a few days of this “explore” phase, having the same conversations and pitching your own research over and over can become very tiring. I think it is fair to accept that you cannot have a meaningful conversation with all attendees. At some point, you may transition to the “exploit” phase, where you mainly spend time with people you know and have deeper conversations. Typically, these are also the people on the same “wavelength” as you, so the chances that you stay in contact after the conference are higher. Of course, you should still go for some conversations on the side, especially with laureates.
Manage your energy. The HLF is very intense and tightly packed with a myriad of sessions and events. Even after the program, which may end as late as 10 PM, you will want to hang out with your fellow young researchers for a while, for example in one of the many bars nearby. Make the most of your time, but remember that the HLF is a long conference (Sunday to Friday). Plan for enough rest and sleep and do not miss the first session of each day. If you are like me and do not want to skip breakfast, also plan some time for that.
Front-load your work. If you need to get work done during the conference, you are going to miss out on something. Complete as much of the work falling into the HLF week as you possibly can before it, so you can enjoy the week more freely. Also, plan for ample time to catch up in the week afterwards.
It is OK to skip some side quests. There are going to be a lot of side quests, like a networking bingo or a photo challenge. Join them for fun or when they serve your goals at the conference, but give them the lowest priority. You do not want to interrupt a conversation with a laureate just to win some chocolate later.

Some takeaways

With these out of the way, here are my personal takeaways from the HLF. I will structure them into AI and non-AI, so you can skip ahead to the non-AI part if you want.

Takeaways regarding AI

The elephant in the room was, without any doubt, AI. This was further catalyzed by OpenAI revealing a solution to the Navier-Stokes Millennium Prize Problem that, as they claim, was produced autonomously by an AI agent just a week before. The conversation around the impact of AI on mathematics had already been going on for a while, at least since Levent Alpöge presented a counterexample to the Jacobian Conjecture in three dimensions, which he credited to Anthropic’s AI model Claude Fable 5. However, the continuing successes, and finally the solution of a Millennium Problem, made the issue urgent. Mathematics acts as a preview of what may come the way of other easily “verifiable” disciplines (such as computer science) as well. It is probably the first target because proofs can nowadays often be checked automatically. This means an AI agent can iterate rather quickly without ever leaving the digital space. At the HLF, the discussion was brought on stage in the aforementioned panel “AI in Mathematical Research”, but also in many other formats.

An interesting point made by Geordie Williamson and Peter Scholze during the panel (at around 1:19:30 in the video) is that progress right now is so rapid that it precludes any kind of prediction. As during the COVID pandemic, the world has become less predictable, and the probability distribution over future states has become very diffuse. In Williamson’s opinion, doing what you really want, rather than being strategic, is the best, and I would say rational, option you have right now.

Over the week, I heard many positions, ranging across the whole possible spectrum. Some noted AI could become an existential threat to humanity (see for example the Spark Talk by Jacob Tsimerman). Some questioned what the role of mathematicians would be once finding proofs is automated. It was also prominently argued that “Everything verifiable will be automated” (Torsten Hoefler, Spark Talk). Along these lines, Jeffrey Ullman argued that more automation could allow us to reduce our working hours and have more recreational and family time (in the same Spark Session). And finally, some also rejected any use of AI.

My personal conclusion is the following. The main issue, as outlined above, is that we currently have no idea how much better the capabilities of current AI systems will become. Everything else depends on that. Given the sheer workforce and dedication the major players put behind the topic today, I think it is fair to assume that current trends will continue. If they do, AI may surpass human abilities to come up with good hypotheses on a broad range of topics within several years. Thus, I will formulate my conclusions under this premise. I believe only some fundamental roadblock (or, less likely, worldwide legislation) could slow this trend down. In that case, what I describe below would apply in a weaker form (which I would not be opposed to).

First, AI will be part of our future as researchers, no matter what we do now. We can choose to ignore it, but I would personally rather face it with open eyes, because it is likely that it cannot be ignored if one wants to stay “in business”. Apart from that, in my experience so far, it can greatly extend one’s abilities and allows us to do research that would otherwise be out of scope. However, the role of humans will change drastically over the next decade or so. Right now, a possible future becomes apparent, in which humans will not be in the scientific loop anymore like they used to be; they will just give directions on what is worth pursuing. In mathematics, this is already starting to happen: people took important, unsolved problems and generated solutions with AI. Once AI can do so more reliably and cheaply than humans, it will be difficult to find funding for the “traditional” way of research. Directions on which problems are of practical relevance, values, and taste will thus become important human input. Maybe some humans get the chance to work on the AI workflows themselves. Other disciplines that require real-world validation will follow, but on a much longer timescale, as real-world validation is a major bottleneck. Until we have adequate simulation models or actuators, it is not possible to easily test thousands of hypotheses, for example when developing a new drug.

To someone like me, who is driven by curiosity and a desire to develop deep knowledge about the world, this sounds devastating (so please contact me with any counterarguments to what I’m saying or post a comment below). After all, knowledge implies a subject that forms understanding. AI could very well become this subject itself one day, but this would leave us humans as passive, clueless consumers, which would defeat my personal motivation for doing science. Or, as Efim Zelmanov put it in his laureate lecture (not yet released on YouTube), the results of complete automation would be “just like Coca-Cola: everybody drinks it, but nobody knows what’s in it.” And to a certain extent this will be the case in the scenario above. However, I don’t think this has to be the end of human understanding. It may just move to a different level, manipulating and working with much higher-level concepts than we have right now. Then again, the premise behind these conclusions may not fully hold, leaving some hope that the change will be less drastic for now. Also, even if the future is with fewer human elements in the loop, humans can still pursue the same level of understanding as a recreational activity. This would tie in nicely with what Jeffrey Ullman suggested. Just getting paid for it will be much more difficult.

Personal takeaways

Beyond likely volatile conclusions on AI, I took home some things that will stay with me for a long time to come.

Science across borders works. The HLF deliberately builds an audience that is diverse in nationality, gender, career stage, and institution. Still, our discussion culture and common goal of advancing what we know let us connect and communicate across any possible personal or cultural differences. This is one of the many things I like about academia and would wish were more widespread in the world.

I also took home some advice from the laureates. Here is one little anecdote: Dennis Sullivan reminded me to always make my problem statement say what I mean. This was after he had given me the long answer, with ample historical context, to a seemingly simple mathematical question on polynomials that turned out to be neither simple nor the right question. I also got some of the laureates to sign my copy of “Masters of Abstraction”, which contains photos and a short text on all laureates, and now also some encouraging words and signatures by them, for example from Jack Dongarra, who attended with his wife Sue (see picture below; posted with their permission).

Jack Dongarra, Turing Award winner of 2021, his wife Sue Dongarra, and me on the boat trip at the HLF 2026.

Lecture hall during Master Class by Nobel laureate Brian P. Schmidt.

The scientific program was very broad, and I also enjoyed looking beyond my own research, for example in a Master Class by Nobel laureate Brian P. Schmidt on “The Creation of the Elements in the Universe” (see above).

Finally, I want to mention that I had the opportunity to talk about the interesting topic of alternative computational paradigms. Naturally, Gilles Brassard and Charles H. Bennett were talking about quantum computing, but I also had discussions on using biological systems for performing calculations (Charles Bennett seemed interested in this as well), new neuromorphic hardware, and in-memory computation. It seems that what I cautiously put in the category of science fiction in a previous blog post is not that far from reality anymore.

Of course, I also made many connections to people from all around the world, whom I will visit when I’m in their corner of the earth.

Our own un-conference

This year’s HLF debuted the “un-conference”, which took place on the final conference day in two blocks of six sessions each. Young researchers pitched the topics they wanted to see discussed on a wall, and every young researcher could vote for up to five of them. The organizers then consolidated the topics with the most votes into the un-conference schedule.

I pitched a topic on publishing, metrics, and incentives in the academic system. A fellow young researcher, Oliver Gracewood, pitched one on publishing practices in the age of AI. As both topics are deeply intertwined, they were merged into “The Academic System in the Age of AI – Rethinking Publishing, Incentives, and Research Culture”. HLF alumnus Michael Bonfert joined us, and together we organized a 60-minute discussion session.

Wall with pitched ideas for un-conference sessions.

Wall with selected ideas for un-conference sessions.

It turned out to be a personal highlight of my week. I had already posted some initial thoughts on this topic in February, and this gave me the opportunity to dive deeper and even hear the opinions of laureates, alongside those of a diverse audience from all around the world. The topic attracted such a large audience that we had to split into smaller groups (which we had anticipated and planned for). After an introductory discussion in the plenum from which several subtopics emerged, the groups discussed “review system”, “metrics”, and “quality vs. quantity”, before we reconvened and each group presented its results.

We then had just a few minutes to condense everything into four bullet points, which Oliver presented on stage to the entire HLF audience in 90 seconds. Here is a slightly longer version of what we summarized:

Text has lost its value as proof of effort.
A review is not only about correctness, but also about whether the right questions were asked and whether a paper adds to the body of knowledge. Current LLMs still struggle with both.
The flood of submissions predates LLMs but is drastically intensified by them. It calls for bottlenecks: submission limits, reputation systems, or fees (possibly refunded on acceptance). This creates tension, as underfunded institutions may be left at a disadvantage.
Many are already experimenting with such measures. As long as there is no consensus, everyone organizing conferences, journals, or review processes can and should look at what others are trying and adopt what works in their institution.

As this does not live up to the breadth of arguments presented, I will cover the discussion in more detail in a separate write-up (coming soon).

Me talking at the un-conference session.

Conclusion

The HLF was the most exciting conference I have attended so far. During this week, I was surrounded by a group of peers from all around the world and immersed in discussions on the broad (research) topics currently relevant to them. My network is now enriched with many new contacts that may in the future facilitate collaboration and the next steps of my career. I took home several learnings, for example on how to deal with the rapid changes induced by AI. The encouraging words from the laureates in the Spark Sessions and personal conversations will motivate me for years to come. I wholeheartedly recommend applying for this unique experience to any young researcher.

I want to thank the Heidelberg Laureate Forum Foundation, in particular the young researcher relations team, for the outstanding organization, and SAP for sponsoring my attendance. The photos of the master class and of me at the un-conference are © Heidelberg Laureate Forum Foundation.

The sunset over Heidelberg as viewed from the castle.

Comments

If you'd like to reply, discuss, or ask something, you can join the thread on Mastodon or send me an Email.