The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Like 85% of my ChatGPT cites have been wholesale hallucinations lately.
Poast new message in this thread
Date: October 22nd, 2025 5:02 AM Author: purple party of the first part It's dropped off a cliff in like the past week. Worse than it's ever been before. Not just fake quotes, but not a single goddamn citation takes you to the case name GPT gives.
Anyone else getting this? |
Date: October 22nd, 2025 6:59 AM Author: Magenta Dingle Berry Go to settings and turn "Anticipation" to off (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49365679) |
 |
Date: October 22nd, 2025 10:54 AM Author: purple party of the first part Where is this on Pro? What does Anticipation do? (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49365987) |
|
Date: October 22nd, 2025 2:21 PM Author: Magenta Dingle Berry It's in settings, and it anticipates sources that will exist rather than merely sources that do exist. It's good for some things, but bad for legal research. (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49366456) |
|
Date: October 22nd, 2025 7:26 AM Author: Flickering submissive toaster Only use deep research function for the law (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49365701) |
|
Date: October 22nd, 2025 8:04 AM Author: mischievous juggernaut The State of Gaming (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49365753) |
|
Date: October 22nd, 2025 8:15 AM Author: mint self-absorbed brunch Seems to be very good still for me. You using the Pro version / mode? (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49365766) |
|
Date: October 22nd, 2025 10:53 AM Author: purple party of the first part Yes. I've been asking very specific shit -- questions about how the normal rules change in the context of an important, small, high-profile class of criminal cases that's increased in number recently, and that carries with it a whole shitload of new procedural rights unique to the class -- and GPT has been just totally unwilling to tell me "I dunno man, can't find a case on that." So it's just making shit up at a rate I've literally *never* seen, with any model. (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49365985) |
|
Date: October 22nd, 2025 11:05 AM Author: Flickering submissive toaster I have mostly not been able to get even the pro version to catch issues that a 2nd year associate should be able to identify lately. It’s good at giving deep dives into existing law and quickly putting together a pretty good memo that competently covers the basics/intermediates (which still is very helpful for me).
For example there’s an exception to the economic loss rule in tort cases where the breach of a public safety statute (in my jdx) allows recovery of economic losses. Only after 5-6 deep research queries did it identify that and I was asking it in a way that should have caught it much earlier.
I haven’t experienced hallucination on deep research mode though. |
|
Date: October 22nd, 2025 10:57 AM Author: pearl boyish partner stage Compare it to Google Gemini which hallucinates less (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49365990) |
|
Date: October 22nd, 2025 11:03 AM Author: carmine deer antler Agreed op, it's gotten noticeably worse lately. I think they are having inference compute problems. They've gotten so popular that they don't have enough compute for all users and so they are starting to skimp out
I canceled my subscription and I'm gonna try out gemini and see how it goes |
|
Date: October 22nd, 2025 12:35 PM Author: purple party of the first part It's been so goddamn terrible it has to be some kind of an acute issue. (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49366196) |
|
Date: October 22nd, 2025 12:47 PM Author: Coiffed nubile headpube theatre lol@ thinking compute power is the problem (http://www.autoadmit.com/thread.php?thread_id=5788578&forum_id=2\u0026mark_id=5310911#49366225) |