One benefit of this will likely be an acceleration of understanding in how LLM actually reach their conclusions with a clear breadcrumb trail of the decision making to allow better compliance with the directive.<p>Alternatively they may just run the result through some sanitizer but that's definitely a worse experience, likely akin to Bing's Chat reluctance to do anything you ask that's even a bit difficult, with blanket "I'm sorry, choose a different topic" refusals, which will be suspicious and infuriating all its own. Especially for queries that have to do with culture and history.