I recently asked it to produce a regex for me, and it responded that as a large language model, it wasn’t allowed to produce hate speech… Not making this up.