1 paper
Tristan Thrush, Sung Min Park, Herman Brunborg +5
What are the limits of controlling language models via synthetic training data? We develop a reinforcement learning (RL) primitive, the Dataset Policy Gradient (DPG), which can pre…