Research on LLM Output Drift in Financial Workflows: Quantifying Consistency Across Model Sizes
Financial institutions deploy Large Language Models (LLMs) for reconciliations, regulatory reporting, and client communications, but nondeterministic outputs (output drift) undermine auditability and trust. We quantify drift across five model architecture
arxiv.org8mo ago
