Hi Lisa,
I noticed your interest in verification and technical AI governance and wanted to share a related research direction.
I am Rafik Zeraoulia, a mathematics researcher and part-time lecturer in Algeria. My work includes resolving open conjectures with AI assistance, with responsibility for checking the proofs and conclusions remaining mine.
I am developing a project on the reliability of AI-generated mathematical proofs: identifying plausible but invalid arguments, constructing checkable counterexamples, and evaluating when independent verification improves error detection. A particular question interests me: how can we ensure that a verified proof establishes the intended claim, rather than a subtly altered statement?
The project is currently at the proposal stage. Its intended outputs include expert-checked examples, reproducible evaluation materials and a public report on the limits of model self-checking. Although the scope is mathematical, I hope these examples could inform broader discussions about verification and trustworthy evidence.
My research record is available here:
https://orcid.org/0000-0002-5436-3320
If this overlaps with your interests, I would welcome a brief methodological comment or a pointer to relevant work.
Best regards,
Rafik Zeraoulia