On Massive Academic Conferences

By Radosław Miernik · Published on 31 January 2026

Table of contents

• Intro
• Planning
• First impression
• Main track
• Opening reception
• Closing thoughts

Intro

Last week, I had the pleasure of attending the 40th Annual AAAI Conference on Artificial Intelligence (AAAI 26) – one of the largest academic conferences in the world, and most likely the largest AI one. It reached a new record of 9562 registrations (attendees), up over 50% from last year¹.

Now let me remind you that usually (if not always) at least one author of each paper is required to attend, as otherwise the paper won’t get published. They received nearly 31000 abstracts with almost 29000 submissions to the Main Technical Track, out of which roughly 23000 papers went under review (the rest were rejected for formal reasons, e.g., were not anonymized), and only 4167 got accepted, resulting in a 17.6% acceptance rate².

This time I co-authored one of these 4167 papers, one of us had to attend the conference. And since it was a good reason to finally visit Asia (Singapore), I decided to do so. My expectations were low, but oh my, oh my, what was that…

Planning

Up to this point, I both submitted to and attended only more narrowly focused conferences, like the IEEE Conference on Games (COG) or the IEEE Congress on Evolutionary Computation (CEC). There were some exceptions, but let’s say I was used to 1-3 tracks (parallel presentations), and an overall feeling that I can talk to everyone and find some common ground shortly.

When I got the detailed tracks for AAAI, I got scared. There were over 20 (!) parallel events multiple times per day, for three days straight (sure, there were some demos and a job fair, but mostly presentations). On top of that, there were two whole days before and two whole days after the conference with workshops and other activities. I really wanted to make the best out of the conference, and it took me only three full hours to decide what to attend.

I also bought an extra ticket for my wife for the opening reception, as these have been one-of-a-kind experiences for me so far, like an evening mingle at Queen Mary University of London’s Octagon. (More on that later.)

First impression

I decided to skip the tutorials (they were paid extra and didn’t seem worth it), but the “Doctoral Consortium” sounded like a good place to start. I also wanted to see the venue, get the badge, meet new people, yada yada yada…

It was held in Singapore EXPO, both in the expo halls (or hangars, if I may) and lecture rooms. Much easier to get around than your average university built a few hundred years ago, but at the same time, there was nothing nice about it.

Anyway, I got my badge, the usual dose of merch (how many tote bags does one need?), grabbed a cup of tea (at least twice as strong as my usual, but that’s a local thing), and headed to the first session. As it turned out, I ended up in a room full of PhD students whose only goal was to present their hard work and network ignore everyone else, so I left during the first poster break.

Main track

“But these are only students; the main track will be better!” I thought. Sure, I was also just a student on the main track, but that’s usually not the case. The first presentation was the opening talk in a huge expo hall. If we exclude the dragon dance (Lunar New Year is big here), there was nothing worth mentioning.

Soon after, the first main track sessions began. I chose “Machine Learning 1” first and “ML: Reinforcement Learning 2” right after it; both seemed like the closest to what I’m interested in and working on³. Here’s what was wrong:

• Technical issues. I had to create four new accounts: OpenReview for the submission and review process, some internal registration tool, Whova for on-site organization, and Underline for recordings. I wanted to list all that, since the last provided on-site help for the presenters.

  But that wasn’t enough to provide us with a “world-class experience” – I had to watch someone’s PowerPoint in edit mode (including notes), via a shared screen over Zoom, including a thick Windows dock and a two-part Zoom overlay at the same time.

  I won’t shame the presenter here, but I just had to take pictures of it, as my friends wouldn’t believe me otherwise. And don’t even get me started on random microphone noises because someone was doing test setup in the other room and forgot to mute themselves…

• English level. It’s not my native language either, I get it. But your paper got accepted here, and it had to be written in English; so were the reviews, which you also had to write. And before the conference, I knew at least some of the papers – as well as the reviews – were automatically translated. It wasn’t a problem back then, as out of the papers I reviewed, exactly zero out of four got accepted⁴.

  I also don’t mind when people agree with someone they know to ask them questions they agreed on upfront. For many of the presenters, it’s the first time they ever presented in front of a relatively big audience⁵.

  But when they can’t understand a question they were asked by someone they know, another friend of theirs rushes to the podium to translate the question to their language, then they try to answer in that language but the session chair disallows that, the presentation is wrapped with “Thank you, let’s talk on the break.” and a rushed exit of all three… That’s something I wouldn’t believe if I wasn’t there.

• Lack of novelty. It may be just me and a few people who I talked to on the conference, but… Since a lot of papers can be summarized as “We took the current state of the art and improved it by 1% on a single benchmark.”, it feels like they could be easily generated⁶.

  I’m not saying they were! I definitely didn’t spend enough time to confirm that either, but again, a handful of people I talked to shared this feeling. It could be a matter of machine translations that gave this feeling.

However! I don’t know why, but all these problems were present only in the sessions on the most popular topics, like Computer Vision, Machine Learning, or Reinforcement Learning. On the last day I spent more time in Games Theory and Economic Paradigms, both during the poster and presentation sessions… And it was a whole new level – zero technical issues, excellent presentations, and a bunch of small-scale breakthroughs.

I have a theory supporting the above observations, but I need at least one or two years to validate them⁷. Let’s say there was something special about this year’s conference that won’t repeat for at least the next two years…

Opening reception

As I mentioned, I got an extra ticket for it for my wife. Until the morning of that day, there was no information about what it looked like, except for the date and time. (The latter was changed, too, but that’s not a big deal.)

Finally, less than eight hours before, we got more information: River Wonders of Mandai Wildlife Reserve it is. Luckily we didn’t go there before, but at least a few people did, so… Tough luck I guess. Was it at least interesting, though…? Let’s say that when we looked for reviews, most of them said “Do it on a rainy day cause it has a covered track.” And that’s during the day; we went there in the evening, so some animals were already asleep.

Was it worth it? It was included in the conference fee, so I wouldn’t mind. But the extra ticket costed S$125. That included a S$45 entry, one can of Coke, one can of Sprite, and a plate of mediocre food worth at most S$10. And no, there was nothing else – we were let in, got handed out the drinks and food at some point of the track, and then we just left.

I understand it’s hard to organize something for nearly ten thousand people, I really do! But if you can’t do any better, maybe you should limit the number of attendees? The extra opening reception tickets were limited, so it is possible.

Closing thoughts

I’m happy our paper got accepted to such a prestigious conference. I’m also happy I attended… Only because I know I don’t want to do it again⁸, but well, it was worth it. To some extent, it was the same as reading Uncle Bob’s books: just to tell others, it may not be worth it.

I’d love to only attend (and publish at!) small conferences, where I could meet people who are interested in similar things as me. If I publish something myself (unlikely) or together with a student (likely), I’ll try that. But if someone employed as a researcher (not as an assistant) joins me, they will aim higher. Not necessarily because they want to, but because they are evaluated on the conference rank. (And that may impact their salary, etc.)

Yet another rant is over.