ECONOMIC VIABILITY ANALYSIS OF LARGE LANGUAGE MODELS FOR HANDWRITTEN SIGNATURE VERIFICATION TASKS
Pavel I. Kolykhalov, Evgeny L. Finogeev
National Research Nuclear University, Moscow, Russian Federation
Keywords: Handwritten Signature Verification (Forensic), Large Multimodal Language Models (LLMs), AI Economic Viability / AI Cost-Effectiveness, Forensic Signature Analysis, Total Cost of Ownership (TCO), Data Localization (Compliance)
Abstract
The article proves that large language models (LLMs) are economically unviable for signature verification. The apparent low cost of token-based pricing is negated by LLMs’ catastrophically low accuracy (59 % error rate), leading to unjustifiably high expenses. Specialized CNN solutions are significantly more efficient and cheaper. LLMs are fundamentally incapable of analyzing details critical for biometrics. The economic paradox of LLMs turns their accessibility into a financial trap. The study proposes a Total Cost of Ownership (TCO) model for Russia, accounting for sanctions, and recommends using LLMs only as a pre-filter in hybrid systems.
|
