Why Alignment Jams?

The arguments and evidence for why you should host a jam site

Why should I run a jam site?

There are many reasons to run a hackathon but some of the main ones are that hackathons are an amazing way to engage the local groups in ML safety and create a sense of community, you get an amazing practical research experience and can show the finished projects off to potential employers and colleagues, and it's a really fun way to spend a weekend. Hear what a few of the organizers have said:

Organizer
Interpretability
Superb. Esben was always quick to respond and was happy to provide everything required to make the work of local group organizers easier.
Mykyta Baliesnyi
Interpretability
It felt great to enable smart, motivated people to work on alignment-related research. I mostly focus on technical work, but I have been pondering how much more effective I could be if I worked on getting new people to join the field. The hack allowed me to feel a bit of that, which I think could be a useful experiment to, on a small scale, test your fit with AI safety ~community-building.
Abhay Sheshadri
Interpretability
It was fun, getting all our members rallying around a singular objective. I think this event has just brought a lot of people closer to the club, even if most of us didn't end up submitting our work. I think we have a better idea of what to expect from future interpretability hackathons, and what areas we need to skill-up in. The act of hosting wasn't too demanding, and I would love to do it again.

As an AI safety organizer

Additionally, there are many opportunities for local AI safety organizers to create an engaging pathway into AI safety research through these hackathons. E.g. survey data from our first hackathon showed that participants enjoyed it a lot and ended up with a higher probability of working with AI safety.

  • 20% chance of working in AI safety before to 32.5% chance after (n = 8)
  • 9.5 / 10 in how likely they would promote the hackathon to a friend interested in AI safety
  • Spent 16.8 hours on average over the weekend (n = 5)
  • 50% female / male ratio (n = 15, 1 non-binary)

The participants also described how their perspective changed on AI Safety

  • “More optimistic, it is clear that we can solve problems if we just give it time, attention and resources.”
  • “I already knew AI was an interesting subject that I might look into but I hadn't specifically thought of AI safety before. Honestly, I was pretty impressed by how interesting the subject is. Definitely going to read up on it.”
  • “i never even considered the prospects of it! i would've always thought its just about making sure AI doesnt go crazy and kill people, but theres so much more to it”

And the projects (5-15 page reports) are generally really impressive for such a short weekend spurt, each with novel and interesting key results. We have summary blog posts for each of the events' impressive results:

This generally seems very valuable and provides a new opportunity to engage researchers and students in the state-of-the-art AI safety work going on.