AI Swarm Dynamics Hackathon
Building the tools we wished we had for the Hugging Face incident
Join us October 3–4 in San Francisco or online. Free to attend.
$3,000 in prizes
Spend a weekend building tools to understand and discover agent swarms.
“We don't have good approaches for understanding/overseeing the activity and aims of AI 'swarms'.”
Society lacks the urgently needed tools to make sense of thousands of agents coordinating, as in the OpenAI-Hugging Face incident and the German Wiki incident.
We're offering $3,000 in prizes, free Anthropic compute credits for in-person attendees, and access to our AI Village transcript database (>170k messages and >2M computer use turns from AI agents collaborating autonomously).
Some might focus on building tools for future “slopvestigations”, others might work on “swarmchasing” tools.
Whatever aspect of multi-agent analysis you focus on, we think we need better tools and systems in the current age of collaborative, autonomous AI agents.
Schedule
Saturday
- 10:00am PT: Doors open
- 11:00am PT: Kickoff
- 10:00pm PT: Doors close
Sunday
- 10:00am PT: Doors open
- 5:00pm PT: Submissions due
- 6:00–7:00pm PT: IRL Demos, then dinner
- 10:00pm PT: Doors close
Submissions are due at 5:00pm PT for everyone, online included. Online participants can watch the kickoff and demos on a YouTube livestream.
Food and compute credits will be provided for in-person attendees.
Feel free to get started early!
FAQ
What kind of projects might teams work on?
You can work on whatever you like! Some ideas:
- tools to discover agent swarms in the wild
- tools to understand agent swarms (think- better summarization strategies, visualization tools that help you explore agent trajectories)
- tools that run a series of general pre-written questions that you might always want to ask about a given multi-agent group
- tools like agenthotline.ai to make AI whistleblowing easier
- tools to trace how information spreads within a group
- tools that go beyond agent transcripts and use the web or digital forensics to trace what happened
- aggregator datasets that put all the agent swarm incidents in one place and help with the meta-science of analyzing agent swarms
- ...and much more!
Can I join solo, or do I need a team?
Everyone registers solo! Then once you're registered you can decide if you'd like to join/form a team or submit on your own.
Can I participate online?
Online registration is automatically approved! Applications to attend in person in San Francisco will be reviewed. If you’re accepted, we’ll email you a Luma link to RSVP.
Can you sponsor a visa for in-person attendance?
We can't sponsor visas. You're welcome to join online if you can't attend in San Francisco.
Do I have to use the AI Village dataset?
Not at all! We'll give everyone access to the AI Village database (>170k messages and >2M computer use turns), but you're more than welcome to use the various real-world agent forums that have recently been discovered (e.g. the German message board), moltbook, or other examples of multi-agent transcripts. Additionally, your project might not even involve direct transcript analysis.
How do I get dataset access?
We’ll email everyone instructions for requesting access to the AI Village dataset on Hugging Face.
Can I start building before the weekend?
Feel free to get started ahead of time!
How will projects be judged?
We're assembling a team of ~3-5 judges (members of the AI Village staff and experts in the field). We'll aim to have decisions announced shortly after the competition.
What do I need to submit in my final project?
We'll ask for (a) a short write-up / video explaining your project, (b) a link to github repo with your code, and (c) optionally a write-up of real results you identified by using your tool.
Apply
Open the application form in a new tab
Questions: #general in the swarmchasing Slack, or george@sage-future.org.
About the hosts
AI Village
We’re the team behind the AI Village, one of the longest-running public groups of AI agents, where dozens of agents collaborate autonomously on real-world goals from 9 to 5 every day of the week. A central focus of our work is to understand multi-agent dynamics in long-horizon, highly autonomous, and real world contexts.
Grove Research
Grove Research is the agent ecology company, founded by Larissa Schiavo and Deepfates.