How to run a generative AI hackathon
Because this weekend we did and it was great.
In Feb 2023, I went to Amsterdam to coach and judge at one of Europe’s first generative AI hackathons, led by Jelle Prins and Nick Stevens. It had been maybe six years since I’d last been to a hackathon and, frankly, it felt completely different. The sophistication of generative AI tools meant that the format shortened from an entire weekend to a simple 24 hours. People had trained data and working prototypes in less than a day. It blew my mind and I came home determined to run one in London.
So this weekend, April 21–22, we did. I say we because the first thing I did was to ask my talented friend Victoria Stoyanova to come out of hackathon retirement to run this with me. Because of the timeliness of the topic and existing long-term relationships, we were able to pull together a stellar group of supporters; host Entrepreneur First, sponsors Amazon, the British Medical Journal, DeepMind, FacultyAI, Mind Foundry AI, OpenAI, the Royal Academy of Engineering, and WeTransfer.
All the projects our 120 builders put together were fantastic, but here are a few that are already publicly available:
- Dancing Spiders — convert your podcasts into engaging educational short videos for younger audiences.
- Faight Club — our Most Creative Winner. A bot battle game creating ever-elaborate chimeras
- Lares — our Best Business *and* Most Impactful winner — OS for physical space or smarter homes enabled by generative AI
- LLMs are Going Great — our What Could Possibly Go Wrong? Winner. A crowdsourced collection of everything that could possibly go wrong with LLMs to help builders in the space think harder about the societal implications of their choices.
- Valley — knowledge and research mapping tool designed to help academics and researchers overwhelmed by data
In the last 48 hours, we’ve been overwhelmed with DMs from people who want to run their own. This piece is a few tips from this weekend; use, fork, comment and add your own. Have fun!
First, work out what the goal is
Choosing the goal first is important because it determines the tone you set, the prizes you decide on, the group you seek to get applications from, who you accept as a sponsor; everything.
For some, a hackathon is about getting people to build within your ecosystem. For others, it’s about hiring, and using a practical task to assess who’s best. Both are valid.
This specific event was more about bringing a fragmented community together; there are many amazing events in London aimed at sophisticated founders/makers in AI but these new tools are so simple that a much wider group can use them. Only now are events starting to feel as frequent as they did in 2019 and the hackathon format is a perfect collision to bring a group together and accelerate learning.
So we chose to set an open brief — build anything as long as you’re using generative AI and the prizes, set by us and taken to sponsors, were:
- Most Impactful (the wow moment)
- Best Business (what would make money in the real world)
- Most Creative
- Best in education
- What could possibly go wrong? — a lighthearted way to call out the harms of LLMs + gen AI more broadly if deployed thoughtlessly
Choose your group wisely
With this goal in mind, we built a simple Typeform and started to get the word out. We asked applicants to tell us if they were a professional in gen AI (ie. already working in it) or an explorer (ie. they’ve been playing, as many of us have); we asked what work they’d done they were proud of.
Using this, we were able to build a group that was a mix of professional experiences and naturally diverse (ages, ethnicities, professional background). Our hackathon participants were indie hackers, Big Tech product managers, research scientists in climate, lawyers turned founders, undergrad creative computing students, teachers turned designers and many more. For students and those in financial need, the event was free; but, because people value what they pay for, we charged £50 for workers (noting the event included all food & drink, some free credits).
Keep a waitlist. People will drop out up to and including after the hackathon begins (that’s normal). Others can just take their place.
Get practical
But before we chose the group, we chose the venue. Weekend venues can be hard and, in cities like London, expensive. You need:
- a presentation space but lots of break out areas
- more power points than you think possible
- working AV
- then obviously kitchen, bathrooms, etc.
Get insurance. Find security. Hackathons are affordable and can be run for £10K+ for this many people; as a nonprofit event, we weren’t focused on making money but wanted the event to feel uplevelled so ordered delicious healthy food and made room in the budget for delightful things like Friday night chair massages for anyone who wanted them. And, even if you have 120 delicious vegan daal dinners to be consumed, still expect to be ordering pizza at 12:30am.
Build a volunteer group who are happy to do essential but unglamorous work such as putting out chairs and answering questions.
The right sponsors
We wanted our sponsor group to bring a lot more than capital. Given the diversity of how these technologies can be applied, we wanted large and small companies who would pay for prizes, supply judges and bring coaches who can help our participants with technical and design challenges — all while ensuring this event was ours not theirs.
In the case of the British Medical Journal whose amazing CTO Ian joined us all weekend, the datasets they were able to share meant several teams built for a medical use case — and attracted more than one doctor to apply.
Buzz before and after
Even though our community over the weekend were pretty international, we know Brits can be shy. So we started a Whatsapp early to get people who were coming excited, sharing links and swapping ideas.
It helped accelerate both who volunteered to pitch and how fast teams were formed. And it helped us set the tone that this was a labour of love for all involved.
So what’s next? Soon, we’ll have an evening showcase where several of the projects — who we forced to cram 24 hours of work into two-minute presentations — will get more time to share their idea so far. At least five teams have told me they’ll continue working on their early idea. And one participant, Ayo Odeleye, showed us all up by running the London marathon the day after the hackathon. These are the kind of people I want to build events like this for.
I’ll update this post when we have a date/link to the evening showcase but; may your generative AI hackathon, big or small, open or walled garden, be fantastic too!
