Ongoing Projects

  1. My BlueDot Impact Technical AI Safety Project focuses on legal compliance and knowledge in AI models.

    • Do AI models know the law? Can we reliably improve their knowledge? Can we use AI to help us police AI?

    • These are some of the questions I’m asking and trying to answer in this project. My initial work starts with the EU AI Act and GDPR laws. To run tests, I’ll be using the UK AISI Inspect framework which is a flexible framework designed to work with many benchmarks that test AI models.

    • I also want to try varying fine-tuning techniques and prompt-mitigation techniques to determine how we can effectively ensure that our models are robust and reliable.

    • Long term, my goal is to expand into other legal jurisdictions.

  2. My second project for the BDI Technical AI Safety Course. The course ended, I got pneumonia and then the Technical AI Safety Project immediately began so this project isn’t even started yet. Now that I’m working on the above Technical AI project, I may tweak this a bit and create a visible dashboard to show the results. My primary goal here is really create something technical that could benefit people with law, legal or governance expertise.

  3. My first research project for the BDI AGI Strategy course. This project focused on extending an existing paper and learning how to use the ToxiGen benchmark. This project is currently on pause while I focus on the Technical AI Safety Project.