@article{transformersprovablyimplementincontextre,
  title = {Transformers Provably Implement In-Context Reinforcement Learning with Policy Improvement},
  author = {Haodong Liang and Lifeng Lai},
  year = {2026},
  eprint = {2605.05755},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2605.05755},
}