MODA aims for more diverse responses without sacrificing quality
The team behind Mode-Conditioned Diversity Alignment says its alignment method draws inspiration from online multi-agent reinforcement learning.
TLDR
The team introducing MODA—short for Mode-Conditioned Diversity Alignment—says the method encourages diverse generation while preserving response quality. It describes the approach as inspired by online multi-agent reinforcement learning.
Combined views
20
1 Source, first seen ago