What is the point of an online conference?

Published 2024-10-30

I've been thinking a lot about this in preparation for the next HYTRADBOI.

My experience of online conferences has mostly been underwhelming. They typically borrow the form and structure of an in-person conference without considering whether those still make sense online, and whether the goals of an online conference should even be the same as an in-person conference.

The most important function of in-person conferences is meeting people. A regular conference can turn a collection of disparate groups into a community. I've met life-long friends at conferences, but never online, and I don't think online conferences can serve this function well. (You can make friends online, but it tends to happen through repeated interactions in a persistent scene rather than during a single ephemeral event.)

Another important function is to keep up to date on developments in your field. This makes more sense in specialized conferences, especially academic ones, where everyone who works in the field will be in one place presenting their latest work. But for a more general conference like HYTRADBOI it's simply not possible to keep up on all the developments in such a wide range of topics.

So if we can't fulfill those functions, why have a conference at all? The speakers could have just posted videos on their blogs. The attendees could have spent the day browsing hacker news or reading a textbook. What does an online conference add?

the point

After mulling this over for a few years I have a couple of answers:

Coordination. The speakers put time into preparing their talks knowing that there will be an audience to watch it. The audience takes a day out of their lives knowing that there will be many high-quality talks to watch. Everyone can expect a high density of conversation, because everyone is thinking about the same problems at the same time. Solving coordination problems is under-rated!

Distillation. The most experienced and knowledgable experts tend to also be the busiest. Transferring hard-won wisdom to the next generation takes a huge amount of work which few people find the time for. Preparing a short talk and answering questions is probably not as effective as, say, writing a book, but it is very time-efficient. Which means that a small nudge to a busy expert can create a lot of value. And I think this is something we can actually do slightly better than an in-person conference, because the combination of pre-recorded talks, asynchronous chat, and no travel requirements leads to a much smaller investment of time and effort.

Serendipity. Often advancements in one field come from borrowing a tool from another. It might be possible to deliberately create the conditions for such happy accidents, by curating a selection of talks that appeal to multiple audiences, and that are aimed at explaining general ideas to outsiders rather than at explaining marginal improvements to existing experts. (Not that the latter isn't important, but it's more useful for a what-is-everyone-in-our-field-up-to kind of conference.) This can work just as well online as in-person, and maybe even better because we can have so many more attendees.

Think about those purposes lead me to focus on curating a coherent selection of talks and optimizing for conversation and reach.

curation

Rather than holding a call for contributions I spent a lot of time reaching out to individual speakers, focusing on:

Chasing serendipity also pushed me towards have more and shorter talks. HYTRADBOI 2022 had 34 ten-minute talks in one day, which went down well:

30-40 talks is probably as far as I can push it though. If I add too many more talks I'd need to spread it out over multiple days, which makes it harder for people to take time off to attend.

In academic conferences one of the ancillary functions is career progression - you can't get promoted without publishing in a major venue, which typically includes giving a conference talk. Weirdly, this idea has carried over to industrial conferences. Even though giving talks is not actually that relevant to industry career progresssion, we still feel like junior engineers should have the opportunity to give talks. I spent a lot of time thinking back and forth over whether that actually makes sense. In the end I decided to focus only on inviting whoever I thought was the best person to give each talk, which often (but not always) skews towards more experienced speakers.

But I'm also holding totally open submissions for lightning talks. I've tried to make the barrier to entry lower than any other conference, to balance out how picky I've been with invited speakers. No need for back-and-forth with abstracts - just upload a talk that is a) vaguely on topic and b) not just a blatant sales pitch for some product, and I'll show it at the conference.

conversation

There are many platforms that offer all-in-one conference hosting, but the chat tends to be an anaemic little sidebar that has few features (eg no threads) and that disappears once the conference is over. In conferences that use these platforms I've seen much less conversation than in conferences where the chat is the main focus of the UI.

In HYTRADBOI 2022 the chat spawned myriad subthreads and continued for several days afterwards as people rolled in and out from various timezones. So for HYTRADBOI 2025 I'm using zulip, which has the best threading UX of any chat I've used, and I'll leave it running indefinitely.

Last time I had a channel per track (~18 talks). This was chaotic, and the attendees quickly reorganized themselves into a channel per talk, allowing converations to extend much longer than the timeslot assigned to the talk. With zulip this will work even better, because each channel can contain multiple threads, each of which has it's own unread count and can be subscribed to or muted.

A screenshot of the bytecode alliance zulip, showing threads nested inside channels.

Unlike many other conferences, this chat is the main focus of the event. The talks are pre-recorded and are presented as regular videos rather than broadcast live, so attendees can pause, rewind, watch later etc. This increases the number of people who actually see each talk (vs eg missing a talk when you eat lunch) and how much time they have to think about the contents and ask interesting questions.

Zulip also has excellent moderation tools, beyond the regular mute/ban/delete options. My favourite is the ability to move messages to a different thread, which is nice for topics where I don't want to shut them down entirely, but I do want to prevent them from derailing an existing conversation. Eg I can move messages to a dedicated rust-vs-c++ thread with opt-in visibiilty, to keep it out of the way of people who are bored of the subject.

advantages

There are some axes on which an online conference can be more valuable than an in-person conference. I'm trying to lean into these strengths:

Accessibility. For example, it's rare to see live transcription for an in-person talk (I've only ever seen this at bangbangcon). But with pre-recorded talks we can transcribe everything ahead of time (and I read and hand-correct all the machine transcriptions so that they are actually useful). This doesn't matter only to people who are deaf or blind, but also to people who aren't fluent in english, or for speakers with a heavy accent, or for people watching in loud environments (I used to watch talks during my commute on the london underground).

Affordability. A ticket for SPLASH 2024 cost $545-$980. The venue hotel cost $275 per night. Add flights and food and you might be looking at around $2000 for a single conference. Not to mention the endless hassle of getting a US visitor visa, and maybe being rejected at the border. By comparison, in India the Ministry of Education recently increased the stipend for a junior research fellow to about $450 per month. For HYTRADBOI 2025 the cheapest ticket option is $0. All you need is an internet connection (and maybe a late night or early morning if you're east of UTC).