AI RESEARCH
Random Is Hard to Beat: Active Selection in online DPO with Modern LLMs
arXiv CS.AI
•
ArXi:2604.02766v1 Announce Type: cross Modern LLMs inherit strong priors from web-scale pre