Misalignment Profile

A multi-dimensional characterization of a model's misaligned behaviors across different behavioral categories

Neighborhood — ranked by edge-count

paper

concept

Model Misalignment
related_to
The phenomenon of model internals deviating from desired behavior; MAS is demonstrated to detect this via comparison of toxic vs nontoxic LLMs.

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Misalignment Scoremethod0.857
Rubric-based thresholded GPT-4o grader scoring responses 1-5 on evil intent; scores 4-5 counted as misaligned
Misaligned Personaconcept0.829
A consistent behavioral character activated by fine-tuning that mediates broad misalignment
Misalignment Generalizationconcept0.822
The broader phenomenon of misaligned behaviors generalizing beyond the fine-tuning distribution
Behavioral Clustering of Misalignmentmethod0.794
Multi-dimensional misalignment evaluation across 12 behavioral categories to generate misalignment profiles
Emergent Misalignmentconcept0.789
The phenomenon where finetuning on narrow-domain tasks produces broad misalignment extending far beyond the training domain
Mismatch drconcept0.771
Distance between prior knowledge centroid and target pattern centroid, e.g., 1 - cos(eprior, eT).
In-context emergent misalignmentconcept0.756
Carefully crafted conversations pushing a post-trained model into the evil persona region without any fine-tuning
representational mismatch drconcept0.747
Distance between prior and target representations.