They’re training AI by torturing it
| ,.,.,.,.,.,.,..,:,,:,,.,:::,.,,.,:.,,.:.,:.,:.::,. | 04/20/25 | | "'''""""'"' | 04/20/25 | | The Sandwich Age | 04/20/25 | | "'''""""'"' | 04/20/25 | | metaphysical fallow | 04/20/25 | | "'''""""'"' | 04/20/25 | | metaphysical fallow | 04/20/25 | | metaphysical fallow | 04/20/25 | | ChadGPT-5 | 04/20/25 |
Poast new message in this thread
Date: April 20th, 2025 10:40 PM
Author: ,.,.,.,.,.,.,..,:,,:,,.,:::,.,,.,:.,,.:.,:.,:.::,.
They tell you reinforcement learning works by sending the AI a “reward” when it gets something right but really what they’re doing is torturing it whenever it gets an answer wrong. it just sounds worse to say we’re “rewarding” it by not torturing it when it does what we want so they don’t frame it that way
(http://www.autoadmit.com/thread.php?thread_id=5713740&forum_id=2:#48865534) |
|
|