May 2023
A preference-training method avoids explicitly fitting a separate reward model.
The toolbox for aligning language-model behavior expands.
Explore on Humanity Unfolds
