“"Show me the reasoning behind this determination" goes from unanswerable to a database query.”
The problem, and we’ve already seen it demonstrated, is that the person typing the query has no control of the query that actually runs. It’s a trivial exercise to set a flag on the data that says “unless the person running the query has this access, don’t even show the data exists.” It’s also a standard database feature called “query rewrite” where the query is transformed for any number of reasons.
Not “there are records you can’t see”, but “no record exists”. The FBI used it systematically over RussiaGate. So there are already controls over who can see the data.
What bothers me most about reading AI-generated text is that it asks more time and energy of the reader than the writer. Even AI "layered over" a text, as you suggest-- in another one of the seven posts from you in my email inbox today-- as a means of changing our relationship to language, is less human-readable, as it was not human-written. I am interested in your ideas, which is why I subscribed in the first place, but I cannot continue to read these posts, as for me personally the artificial prose is unbearable.
“"Show me the reasoning behind this determination" goes from unanswerable to a database query.”
The problem, and we’ve already seen it demonstrated, is that the person typing the query has no control of the query that actually runs. It’s a trivial exercise to set a flag on the data that says “unless the person running the query has this access, don’t even show the data exists.” It’s also a standard database feature called “query rewrite” where the query is transformed for any number of reasons.
Not “there are records you can’t see”, but “no record exists”. The FBI used it systematically over RussiaGate. So there are already controls over who can see the data.
Here’s a link to the FBI example: https://x.com/ProfMJCleveland/status/1928580853245898772?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E1928580853245898772%7Ctwgr%5Eecb4d0b3055ad425550f76860b28ed1452d5b50b%7Ctwcon%5Es1_c10&ref_url=https%3A%2F%2Fpublish.x.com%2F%3Furl%3Dhttps%3A%2F%2Ftwitter.com%2FProfMJCleveland%2Fstatus%2F1928580853245898772
Making citing the rule cheap requires it to be reliable. We've seen cases where it wasn't.
I wonder if those cases were so public at least partly to slow down adopting AI. "It's unreliable".
Though I've seen AI hallucinations in my own work (software). Hallucinatory precedents is entirely believable.
What bothers me most about reading AI-generated text is that it asks more time and energy of the reader than the writer. Even AI "layered over" a text, as you suggest-- in another one of the seven posts from you in my email inbox today-- as a means of changing our relationship to language, is less human-readable, as it was not human-written. I am interested in your ideas, which is why I subscribed in the first place, but I cannot continue to read these posts, as for me personally the artificial prose is unbearable.
Do I understand correctly that you believe this stack’s articles to be AI generated?