AI Safety Research

I conducted several different research projects to gain familiarity with existing AI tools and research. These can be seen below.

  1. My BlueDot Impact Technical AI Safety Project focused 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?

    • Initial work started with the EU AI Act and GDPR laws. Tests ran within the UK AISI Inspect framework, a flexible framework designed to work with benchmarks that test AI models.

    • Tested fine-tuning and prompt-mitigation techniques to determine how robust and reliable models are

  2. My project idea for the BDI Technical AI Safety Course. This project was superseded by the one above.

  3. My first research project for the BDI AGI Strategy course extended ToxiGen research and benchmark.