A team preparing a production rollout of a generative AI assistant is unsure whether factsheets apply to it. What is the correct position?
Select an answer to reveal the explanation.
Short Explanation
You don't get to skip factsheets just because generative AI isn't deterministic. If anything, that makes purpose, data, and evaluation evidence even more important. You document generative AI assets the same way you document traditional predictive models.
Full Explanation
Factsheets and evidence records apply to generative AI assets just as they do to traditional predictive models. Purpose, data provenance, evaluation, and decision documentation are as necessary for generative assets as for any other model type.
The claim that factsheets should be avoided because outputs are not deterministic is wrong: non-determinism is a property of outputs, not a reason to skip documenting purpose, data, and evaluation.
The claim that factsheets are optional and can be replaced with user reviews is wrong: user reviews are not structured evidence records.
The claim that factsheets apply only to externally sold generative AI assets is wrong: internal assets carry the same documentation need.
For the exam, apply factsheet requirements uniformly to generative and traditional AI assets alike.