Question

A financial services company wants to use a large language model to automatically calculate exact regulatory penalty amounts owed by clients based on structured account data. During testing, the compliance team notices that submitting the same account data multiple times occasionally produces slightly different penalty figures, and some figures are entirely fabricated. Which characteristic of generative AI makes it a poor fit for this specific task?

Accepted Answer

Generative models are nondeterministic and prone to hallucination, so they cannot guarantee the exact, repeatable numerical results required for a regulatory calculation.

Answer

Generative models require GPU acceleration, which introduces rounding errors into financial arithmetic.

Answer

Generative models can only process text and cannot read structured account data of any kind.

Answer

Generative models are limited by their context window and therefore cannot perform arithmetic at all.