05:38
2026-07-14
machinebrief.com
artificial-intelligence
Agentic-DPO: Revolutionizing Language Model Training Without Costly Rollouts
Agentic-DPO, a new method for training language models using offline preference supervision without costly environment rollouts, improved accuracy on tau-bench from 21.7% to 41.4% for a 9B model, matc…