EA Tech London x Apart Research: Deception Detection Hackathon
Are you interested in AI safety but not sure where to start? Perhaps you don't know what a deceptively aligned mesa optimizer is, but you think it sounds scary? Maybe you just want to spend a weekend trying to do something cool with language models, with expert mentorship and great company?
If your answer to any of these questions was yes, then you'll be excited to learn that EA Tech London are hosting a jam site for the next Apart Research hackathon, focussed on deception detection. This is a great chance to test your fit for AI safety work, as Apart provide everything you need to put together a successful project in a weekend, from access to relevant experts to Python templates. There are cash prizes available for the best submissions, as well as opportunities to meaningfully contribute to deception detection in only a short project.
Check out the Apart Research page for more information, and to sign up for further updates (remember to sign up on this page too!): https://www.apartresearch.com/event/deception
FAQ:
Q: Do I need to attend the whole event?
A: You don't need to attend the whole weekend, but it'd be great if you could. There will be a keynote speaker on Friday night that we'll all watch together, and a show & tell section on Sunday night to share our work. The goals of the hackathon are quite ambitious, and participants are encouraged to aim to have a full working demo and blog post or similar by the end of the weekend - this will take a lot of time, even if you're in a team.
Q: Do I need to find my own team, or can I participate on my own?
A: You're welcome to participate on your own, to form your own team, or we can also facilitate matching people into teams.
Q: What if I'm not technical?
A: You'll probably get the most out of the event if you have some basic familiarity with Python (although LLMs are pretty good at writing that for you). With that being said, most teams will aim to have a written component to their project. If you have a great idea for detecting deception in LLMs but aren't sure how to implement it, then we can help find someone technical for you to work with.
Q: Will there be food provided?
A: In order to keep the event free, we won't be providing food. There are plenty of places within walking distance of the LISA office that you can get food from. I (Jonny) will also probably do a takeaway order or similar for lunch and dinner on Saturday and Sunday, and probably dinner on Friday too, and you'll be welcome to chip in and join me.
Q: What time will the venue be open?
A: On Saturday and Sunday we'll be going from 10am to 10pm. On Friday, we'll be going from 5pm to 10pm.