pawnty
Re:Quiet-STaR: Language Models Can Teach Themselves to Think Before Sp...
on Friday March 15, 2024 @08:41AM (#39714865)
This is the missing piece to train AI which has the ability to reason. There are so many tasks whose answers are known but reason steps are missing. With this method, we can use less annotated data the reach the ability.
The interesting part(I imagine): the generated thought could be hard for human to understand while it is still way more helpful to get the correct answer! If that happens, we have created something more intelligent than ourselves.
Re:Linux TTY font for Chinese, but treat it as a syllabic writing
on Wednesday March 13, 2024 @10:08PM (#39699927)
Thanks. It helps to find some interesting phenomena in neuroscience.
Re:LaVague: Open-source Large Action Model to automate Selenium browsing
on Wednesday March 13, 2024 @09:43PM (#39699792)
A benchmark is helpful to show the success rate.