Crichton was talking about the news media, but the same thing's true of AI. Try to have a conversation or ask probing questions of ChatGPT Pro on areas where you have pre-existing expertise and/or already know the answers, and you'll immediately see that its rife with errors, oversimplifications, and failures to draw inferences that would be relevant to your prompt.

But it's fucking great at giving you primers on shit you don't know much about -- frankly equivalent to new attorneys (and superior to TTTT summer associates) who often get that exact kind of task, in .01% of the time.

Obviously AI is still useful -- just like the media, as an institution, is useful to society and helps me know about shit that I wouldn't otherwise know about -- but it does make me wonder how much overconfident horseshit is lurking in its seemingly 180^180 memos on supersedeas-bond mechanics, how to find/buy a patek philippe perpetual calendar, and how to tweak my daily supplement stack and workout routine.

(http://www.autoadmit.com/thread.php?thread_id=5714813&forum_id=2#48872145)

60 Minutes interviewer gasps, recoils in shock

(http://www.autoadmit.com/thread.php?thread_id=5714813&forum_id=2#48872156)

HUG-HMS

Blonde spinner wives

(http://www.autoadmit.com/thread.php?thread_id=5714813&forum_id=2#48872182)

AI is going through the same phase the internet is where at first people thought if something showed up in a google search it must be true, but after being burned a few times they realized it wasn't an oracle.

ai can be reasonably trusted to get to the right answer on things computers are good at like math or for questions you know are easy. it's a really great tool for proofreading. but anything complex it's best served as a jumping off point. just like how wikipedia is a good tool to begin researching something, but you gotta dig into the primary sources and be aware of bias on the platform.

(http://www.autoadmit.com/thread.php?thread_id=5714813&forum_id=2#48873547)

All of this still makes the quintessential junior associate theoretically replaceable by AI, since the associate is likely relying on the same internet-based information sources that AI can access. But it hasn't been super-helpful lately for the most front-of-mind professional questions I've been facing lately (my job mostly consists of, and I am mostly hired these days, because I hang out with judges and politicians and rich businessmen).

(http://www.autoadmit.com/thread.php?thread_id=5714813&forum_id=2#48875433)

I once had Pro give me a flat-out wrong (and dangerously so) legal assertion that it elaborated on immensely and made the linchpin of an entire argument. It was on the verge of shaking my confidence in the whole thing, but I pulled the sources and this exact error came directly from a (reputable, very much on-point) law review article. Hard to blame chatgpt for that.

And that's a topic that people *are* writing on. There's so, so much shit out there that still doesn't get memorialized for chatgpt to have any chance at seeing.

(http://www.autoadmit.com/thread.php?thread_id=5714813&forum_id=2#48876131)

Pass

(http://www.autoadmit.com/thread.php?thread_id=5714813&forum_id=2#48873587)