ARENA 6.0 Interpretability Hackathon: Deep Dives into Misalignment

​‼️ IMPORTANT ‼️
You will need to claim your $400 of cloud credits per team by Friday. This is crucial as the Lambda team will not process credits over the weekend!

You will be emailed all instructions when you signup to the hackathon here:
https://apartresearch.com/sprints/arena-6-mechanistic-interpretability-hackathon-2025-09-13-to-2025-09-14

​---

Join us for ARENA 6.0’s in-person Mechanistic Interpretability Hackathon - open to all LISA members and run in partnership with Apart Research - where we'll spend the weekend tackling real, open problems in interpretability.

​🔬What You’ll Do

• ​Tackle real interpretability challenges with optional starter Colabs and compute support

• ​Work in teams (or solo!) to develop insights into how models process information

• ​Get mentorship and guidance from leading researchers

​

​🏆Prizes & Recognition
Prizes TBC! Winning projects will be judged by top AI safety researchers and featured on the Apart Research website.

​📍Details

• ​When: Sept 13-14, 2025 (12-6pm daily)

• ​Where: In-person @ LISA, London

• ​Provided:

  • ​$400 in cloud GPU credits per team provided by Apart

  • ​Sample Colabs, office hours, and mentorship

  • ​Food & drinks throughout the weekend

• ​Projects are encouraged to fit the theme of "deep dives into misalignment" - if you're short on ideas, the demo Colabs can give you ideas or get you started!

​

​🧠Who Should Join?
Anyone curious about how AI systems work under the hood - no need to be an ARENA participant!
Some Python or ML experience helps, but no prior research background is needed.

​

​📌Note: This is an in-person only event.