# Luiza Corpaci

I study and build certificate-backed methods for detecting semantic drift in AI-generated specifications, proofs, and other artifacts. I currently mentor research at MARS V and SPS Apart, participate in CAISH’s Hardware Assurance Program, and am completing an MSc in Data Science at TU Wien.

## Start here
- /blog/limits-of-formalization: what formal systems can and cannot reach, and what that means for honest AI
- /limits/: interactive walkthrough of the self-reference limits (Liar, Tarski, Gödel, alignment)
- /map/: interactive map of 24 self-reference instances, from quines to Gödel to ELK

## Research threads
- /formalization-limits/
- /interpretability-grounding/
- /ai-safety/
- /collective-intelligence/

## Core topics
- Semantic faithfulness across representations: Equational Theories Project as external ground truth, probe transfer gaps across textual renderings
- Limits to formalization: syntax-semantics interface, incompleteness, what formalisms cannot reach
- Interpretability and grounding: what measurement reveals and conceals about neural networks (Grounding Relative Understanding)
- Social bias measurement: geometric approaches, midpoint geometry, SONAR embeddings
- AI safety: feature invariance, reliable interpretability, formal tools for deployment guarantees

## Projects
- /projects/sair-embedding-geometry/
- /projects/bias-measurement/
- /projects/ictss-playground/
- /collatz-chattie/

## Writing
- /blog/limits-of-formalization: The Limits of Formalization for AI Safety
- /blog/good-formalisms-for-good: What formalisms are good for defining good?
- /blog/good-feels-like: How Does Good Feel Like?
- /blog/verification-validation-boundary: The Verification/Validation Boundary

## Academic profiles
- https://scholar.google.com/citations?user=OhRQsPAAAAAJ
- https://orcid.org/0009-0004-4432-6843
- https://github.com/corpaci