This episode of The ML Digest covers the paper “Towards a Unified View of Large Language Model Post-Training” from researchers at Tsinghua University, Shanghai AI Lab, and WeChat AI. The authors argue that seemingly distinct approaches—Supervised Fine-Tuning (SFT) with offline demonstrations and Reinforcement Learning (RL) with online rollouts—are in fact instances of a single optimization process.

Link to original paper: https://arxiv.org/pdf/2509.04419