RSFI: Riemannian System Fidelity Index. Метод динамического контроля семантического дрейфа и сикофантии LLM на основе неевклидовой геометрии (Zero-shot, Black-Box, ZCA Whitening).
python nlp machine-learning jailbreak cybersecurity zero-shot ai-safety riemannian-geometry llm-security sycophancy jailbreak-defense semantic-drift sycophancy-detection black-box-api black-box-security
-
Updated
Aug 26, 2026 - Python