Conditional-Prediction
-
PAPERDPDM routes agent decisions through one frozen conditional predictor instead of prompting
A paper proposes DPDM, which pairs a language model encoder with a frozen pretrained predictor so agents learn from deployment history without any weight updates.