Question

A financial services firm wants to deploy a generative AI assistant that summarizes quarterly earnings for analysts. During testing, leadership notices that asking the model the same question multiple times sometimes produces summaries with slightly different figures and conclusions. Regulators require that the same input always yield an identical, verifiable output. Which characteristic of generative AI is the primary source of this concern, and what should the transformation leader communicate to stakeholders?

Accepted Answer

Generative AI models are inherently probabilistic and may produce varying outputs for identical prompts, so deterministic, verifiable results require additional controls or a different approach for regulated calculations.

Answer

The variation is caused entirely by poor prompt engineering, and rewriting the prompt will guarantee identical outputs every time.

Answer

The model needs more training data, after which all outputs will become fully deterministic and reproducible.

Answer

This behavior indicates a security breach in which unauthorized users are altering the model's responses.