The Geometric Canary: Predicting Steerability and Detecting Drift via Representational Stability
For practitioners deploying LLMs, this provides a unified geometric diagnostic for pre-deployment controllability assessment and post-deployment drift monitoring.
The paper shows that geometric stability of representations predicts linear steerability with near-perfect accuracy (ρ=0.89-0.97) across 35-69 models and detects drift with up to 5.23× greater sensitivity than CKA during post-training alignment, while providing earlier warnings and lower false alarm rates.
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 ($ρ= 0.89$-$0.97$) across 35-69 embedding models and three NLP tasks, capturing unique variance beyond class separability (partial $ρ= 0.62$-$0.76$). A critical dissociation emerges: unsupervised stability fails entirely for steering on real-world tasks ($ρ\approx 0.10$), revealing that task alignment is essential for controllability prediction. However, unsupervised stability excels at drift detection, measuring nearly $2\times$ greater geometric change than CKA during post-training alignment (up to $5.23\times$ in Llama) while providing earlier warning in 73\% of models and maintaining a $6\times$ 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.