The Geometric Canary: Predicting Steerability and Detecting Drift via Representational Stability

arXiv:2604.17698v1 Announce Type: cross Abstract: Reliable deployment of language models requires two capabilities that appear distinct but share a common geometric foundation: predicting whether a model will accept targeted behavioral control, and detecting when its internal structure degrades. We show that geometric stability, the consistency of a representation's pairwise distance structure, addresses both. Supervised Shesha variants that measure task-aligned geometric stability predict linear steerability with near-perfect accuracy (rho = 0.89-0.97) across 35-69 embedding models and three NLP tasks, capturing unique variance beyond class separability (partial rho = 0.62-0.76). A critical dissociation emerges: unsupervised stability fails entirely for steering on real-world tasks (rho approx 0.10), revealing that task alignment is essential for controllability prediction. However, unsupervised stability excels at drift detection, measuring nearly 2imes greater geometric change than CKA during post-training alignment (up to 5.23imes in Llama) while providing earlier warning in 73% of models and maintaining a 6imes lower false alarm rate than Procrustes. Together, supervised and unsupervised stability form complementary diagnostics for the LLM deployment lifecycle: one for pre-deployment controllability assessment, the other for post-deployment monitoring.

Source: arXiv cs.CL | 2026-04-21