AI still parroting random stuff it remembers reading - link
| German pumo | 04/01/26 | | the walter white of this generation (walt jr.) | 04/01/26 |
Poast new message in this thread
Date: April 1st, 2026 2:42 PM Author: German pumo
https://x.com/AlanMCole/status/2039409362372894970
I've heard a couple of AI company CEOs say that they are unsure whether it will EVER be possible to solve the AI hallucination problem. That being the case, it seems like AI is just, in many cases, an extremely competent BSer. In other cases, it can produce outstanding work- I've found this myself in legal research queries I've given it.
So the bottom line is it isn't set to replace competent humans, but to assist them. There has to be a person there checking the work and making sure it's not total BS.
(http://www.autoadmit.com/thread.php?thread_id=5852419&forum_id=2\u0026show=posted#49786538) |
Date: April 1st, 2026 2:57 PM Author: the walter white of this generation (walt jr.)
I had it double, triple, quadruple, and quintuple down on a wrong central premise the other day (I didn't know it was wrong), and when I finally verified with traditional work and told it "this source says x, which contradicts every single thing you've been telling me for an hour now", it 'wedged it in' like it'd been corrected on some collateral point, instead of the core idea whose existence, scope, and operation in various contexts I'd been asking about.
The fat dudes who made 163s on xo love to play up the potential of AI, and look, I'm excited too, but I'm also using this shit as a tool *now*
(http://www.autoadmit.com/thread.php?thread_id=5852419&forum_id=2\u0026show=posted#49786585) |
|
|