The most prestigious law school admissions discussion board in the world.
|
 |
 
|
AI still can't read Blood Meridian and understand it
| https://i.imgur.com/YbMiOBE.png | 09/13/26 |
| Junko Enoshima | 09/13/26 |
| https://i.imgur.com/YbMiOBE.png | 09/13/26 |
Poast new message in this thread
Date: September 13th, 2026 7:52 PM Author: https://i.imgur.com/YbMiOBE.png because of the way it's written. You can paste the whole text into the context window and ask questions, and the model will only ever spit out wrong answers. You can also turn the text into a vector database to avoid context rot and you'll get the same outcome: the AI just can't understand who's talking or what's going on (http://www.autoadmit.com/thread.php?thread_id=5903643&forum_id=2.#50134246) |
Date: September 13th, 2026 8:03 PM Author: Junko Enoshima Try giving it shorter passages. Even ChatGPT 3.5 was able to help me understand passages of Proust. (http://www.autoadmit.com/thread.php?thread_id=5903643&forum_id=2.#50134261) |
 |
Date: September 13th, 2026 8:08 PM Author: https://i.imgur.com/YbMiOBE.png I'm always doing needle in the haystack or searching for broad overlapping themes, so I try to work with whole books as much as possible. I'm careful to remove indexes and introductions but I've had tons of luck turning big books into vector databases and doing semantic searches. I might give the AI 8 vector databases at once and ask "do these authors overlap thematically at all?" (http://www.autoadmit.com/thread.php?thread_id=5903643&forum_id=2.#50134269) |