Presentation: Fine Tuning the Enterprise: Reinforcement Learning in Practice

Wenjie Zi and Will Hang from OpenAI present Agent RFT, a platform enabling real-time tool interactions to fine-tune reasoning models. The system addresses complex credit assignment challenges within context windows by utilizing custom reward signals instead of standard supervised learning methods. Enterprise users adopt this approach to eliminate long-tail token loops while driving extreme efficiency in production workloads.