> especially with AI.

While AI models still hallucinate commonly, this might not be the best use of it.

The downsides of the occasional hallucinated answer for building standards seems like it could be pretty bad. Seems much worse than the upsides?