@misc{indiciae75fbd677a667,
  title = {Why Does RL Generalize Better Than SFT? A Data-Centric Perspective on VLM Post-Training},
  author = {Aojun Lu and Tao Feng and Hangjie Yuan and Wei Li and Yanan Sun},
  year = {2026},
  url = {https://arxiv.org/abs/2602.10815},
  note = {Source identifier: 2602.10815}
}